The relationship between context and health inequalities: Europe and Portugal as case studies

Universidade Nova de Lisboa Instituto de Higiene e Medicina Tropical The relationship between context and health inequalities: Europe and Portugal as case studies Inês Campos Matos DISSERTAÇÃO APRESENTADA PARA CUMPRIMENTO DOS REQUISITOS NECESSÁRIOS À OBTENÇÃO DO GRAU DE DOUTOR EM SAÚDE INTERNACIONAL, ESPECIALIDADE DE POLÍTICAS DE SAÚDE E DESENVOLVIMENTO JUNHO, 2017 Universidade Nova de Lisboa Instituto de Higiene e Medicina Tropical The relationship between context and health inequalities: Europe and Portugal as case studies Autora: Inês Ferreira Pita de Campos Matos Orientador: Professor Giuliano Russo Coorientadora: Professora Luzia Gonçalves Dissertação apresentada para cumprimento dos requisitos necessários à obtenção do grau de Doutora em Saúde Internacional, especialidade de Políticas de Saúde e Desenvolvimento, de acordo com o Regulamento Geral do 3.º Ciclo de Estudos Superiores Conducentes à Obtenção do Grau de Doutor pelo Instituto de Higiene e Medicina Tropical/Universidade Nova de Lisboa (n.º 474/2012) publicado no Diário da República, 2.ª série, n.º 223 de 19 de novembro de 2012. Apoio financeiro: Subsídio para Internos Doutorandos da Fundação para a Ciência e a Tecnologia Referência SFRH/SINTD/94891/2013 Bolsa Fulbright para Investigação em Saúde Pública da Comissão Fulbright Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 i Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 ii Publications from this dissertation Campos-Matos I, Kawachi I. Social mobility and health in European countries: does welfare regime type matter? Social Science and Medicine. 2015;142:241-248 DOI: https://doi.org/10.1016/j.socscimed.2015.08.035 Campos-Matos I, Subramanian SV, Kawachi I. The ‘dark side’ of social capital: trust and self-rated health in European countries. European Journal of Public Health. 2016;26(1):90-95. DOI: http://dx.doi.org/10.1093/eurpub/ckv089 Campos-Matos I, Russo G, Perelman J. Connecting the dots on health inequalities – a systematic review on the social determinants of health in Portugal. International Journal for Equity in Health. 2016;15(1):15-26. DOI: 10.1186/s12939-016-0314-z Campos-Matos I, Russo G, Gonçalves L. Shifting determinants of health inequalities in unstable times: Portugal as a case study. European Journal of Public Health. 2017. DOI: https://dx.doi.org/10.1093/eurpub/ckx080 Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 iii Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 iv Acknowledgements To my supervisors – Luzia Gonçalves and Giuliano Russo – and other people who supervised my work on this thesis – I. Kawachi, S.V. Subramanian, and J. Perelman. To colleagues and bosses in other areas of my life – in particular Jorge Nunes and Pedro Serrano. To everyone who took the time to teach me, I am deeply grateful. To my friends, who mostly distracted me from my work, and to whom I profoundly thank for that. To my family, in particular to my parents, who have always supported me. My love, gratitude, and admiration for you is never-ending. To Chris. Thanks for all the chocolate, patience, and ranting. Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 v Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 vi Prevention is the heart of Public Health. But equity is its soul. Dr Margaret Chan Opening address at the Executive Board special session on WHO reform, November 2011, when she was Director-General of the World Health Organization Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 vii Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 xiv 3.1.4. Shifting Determinants of Health Inequalities in Portugal ...................... 107 3.1.5. Summary ................................................................................................ 109 3.2. Limitations ....................................................................................................... 111 3.3. The Effect of Context on Health and on Health Distribution .......................... 113 3.3.1. First Mechanism: Changes in Overall Health ........................................ 113 3.3.2. Second Mechanism: Changes in Health Distribution ............................ 115 3.3.3. Conceptual Framework .......................................................................... 117 3.3.4 Summary ................................................................................................. 121 3.4. Application of the Conceptual Framework ...................................................... 122 3.4.1. Social Capital and Health in European Countries .................................. 122 3.4.2. Social Mobility and Health in European Welfare Regimes ................... 123 3.4.3. Shifting Determinants of Health Inequalities in Portugal ...................... 125 3.4.4. Summary ................................................................................................ 127 3.5. Contribution to Policy and Research ............................................................... 129 3.5.1. Regarding Health Inequalities ................................................................ 129 3.5.2. Regarding Europe .................................................................................. 130 3.5.3. Regarding Portugal ................................................................................ 131 3.5.4. Summary ................................................................................................ 132 3.6. Conclusions ...................................................................................................... 133 3.7. Discussion and Conclusion References ........................................................... 135 4. APPENDICES ........................................................................................................ 141 4.1. Appendix 1: Illustration of how changes in health inequalities affect absolute and relative measures – a hypothetical example ..................................................... 143 4.2. Appendix 2: Online supplementary data from the first publication ................. 145 4.3. Appendix 3: Online supplementary data from the third publication ............... 151 4.4. Appendix 4: Online supplementary data from the fourth publication ............. 171 Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 xv Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 xvi List of Figures Figure 1. Wider Determinants of Health Model, by Dahlgren and Whitehead ........... 3 Figure 2. WHO's Commission on the Social Determinants of Health Framework ..... 5 Figure 3. A framework for elucidating the pathways from the social context to health outcomes and for introducing policy interventions ..................................................... 29 Figure 4. Flow of information through the different phases of a systematic review proposed by the PRISMA statement ............................................................................ 46 Figure 5. Hypothetical concentration curve ................................................................. 49 Figure 6. First mechanism: changes in overall health ................................................. 114 Figure 7. Second mechanism: changes in health distribution ...................................... 116 Figure 8. The impact of context on health and health distribution: conceptual framework ..................................................................................................................................... 117 Figure 9. Effect of contextual social capital on health distribution ............................. 123 Figure 10. Effects of welfare regimes and social mobility on health and health distribution ................................................................................................................... 125 Figure 11. Effects of socioeconomic changes on health and health distribution ......... 127 Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 xvii Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 xviii List of Tables Table 1. Welfare regime types, their main characteristics, and example countries ........ 23 Table 2. Summary of dissertation publications, determinants tested, geographic context, time period, and main findings .................................................................................... 110 Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 xix Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 xx List of Abbreviations BMI Body Mass Index CC Concentration Curve CHD Coronary Heart Disease CIx Concentration Index CSDH Commission on the Social Determinants of Health DFID Department for International Development ESS European Social Survey EU European Union EU-SILC European Survey on Income and Living Conditions GDP Gross Domestic Product HI Health Inequalities HIV Human Immunodeficiency Virus ILO International Labour Organization OR Odds Ratio PRISMA Preferred Reporting Items for Systematic Reviews and Meta-Analyses RII Relative Index of Inequality SAH Self Assessed Health SDH Social Determinants of Health SES Socioeconomic Status SII Slope Index of Inequality SR Systematic Review UK United Kingdom US United States USSR Union of Soviet Socialist Republics WHO World Health Organisation Dissertação para a obtenção do grau de Doutor em Saúde Internacional Inês Campos Matos, Junho de 2017 xxi The relationship between context and health inequalities Chapter 1. Introduction 1 1. Introduction 1.1. The Study of Health Inequalities 1.1.1. Four Decades of Politics and Research on Health Inequalities Academia has a long tradition of seeing itself as an autonomous body, free from ideological thought and from social and political context (1). However, academic work is not done in isolation from its surrounding world. The story of how academic knowledge about health inequalities (HI) and the social determinants of health (SDH) has evolved in the past four decades is a reflection of how this knowledge and dominant political thought go hand in hand, sometimes one pushing forward more vigorously. The connection between socioeconomic determinants and health has been known for centuries, but there has been a growing interest in the subject in the last four decades. In Europe, this political and academic interest was shaped by a few landmark events. The first political landmark was the publication of the Black Report in the United Kingdom (UK) in 1980 (2). This report was set up by a Labour Secretary of State for Health who was concerned about the dimension of mortality inequalities between social classes (3). The report found that, despite three decades of a National Health Service, HI still existed and could largely be explained by differences in material conditions between social classes (2). But the final document was reported to a Conservative government, elected in 1979 under the commitment to reduce public spending, who dismissed it and failed to properly publish it (3). Despite this, the report played a pivotal role in setting the research agenda for two decades after it was published (3, 4). The findings of the Black Report were informed by a body of academic work published in the years before. Probably the most noticeable findings came from the Whitehall study (3), a longitudinal study of civil servants working in London that started in 1967 to analyse the ‘power of risk factors and indicators of coronary heart disease (CHD) to predict mortality’ (p. 1165). The Whitehall study showed a clear inverse relationship between grade of employment and CHD mortality that persisted ever after controlling for a wide range of cardiovascular risk factors (5). As one of the authors of the original study, Sir Michael Marmot, later commented in an interview, The relationship between context and health inequalities Chapter 1. Introduction 2 this was unexpected, as at the time common sense suggested that heart disease was more common in people with higher-grade, more stressful jobs (6). Overall, the Whitehall studies together with the Black Report reflected a change in the understanding of the determinants of health. The UK was seen as a success in terms of population health improvement, with extensive state-provided social support and universal healthcare provision. These analyses showed that, over and above the direct effect of the most basic determinants of health – such as diet or healthcare – strong social forces operated to create a health gradient, even among people who had access to all essential living conditions. The positioning of HI and SDH in the research agenda led to a proliferation of empirical analyses in the years ensuing the publication of the Black Report. In 1991, Dahlgren and Whitehead proposed the ‘Wider Determinants of Health Model’, probably the most widely known and used framework on this topic (7) (figure 1). This model describes the main influences of health, built in layers, one on top of the other. The structural environment is the overarching layer, which includes ‘general socioeconomic, cultural and environmental conditions’. This is followed by a layer of living and working conditions, which includes factors like employment and education. The next two layers refer to support from social networks and individual lifestyle, respectively. Finally, the central layer is made of individual unchangeable factors, such as age and sex. This model proved useful in presenting the main determinants of health to broad audiences, and was groundbreaking when it was first published, as it highlighted the importance of broader socioeconomic factors in the production of health. It also emphasized the cumulative nature of the determinants of health and provided a framework upon which to consider policy options, as each layer can be translated into a level of policy intervention. In 1997, after eighteen years of conservative rule, the UK elected a Labour government. This new government was elected with a strong commitment to reduce social inequalities, and quickly commissioned a report to ‘review and summarise inequalities in health in England and to identify priority areas for the development of policies to reduce them’ (8: p.5). This was materialized with the publication of the Acheson Inquiry in 1998, which revealed a wide range of HI throughout the lifecourse, some even increasing over the previous decades. The report clearly stated that HI were a consequence of socioeconomic factors, and as such could only be tackled The relationship between context and health inequalities Chapter 1. Introduction 9 responsible for the distribution of individual determinants of health, such as income or education (figure 2) (12). But how does context influence health and its distribution? The answer to this question has been undermined by a number of challenges facing this body of literature. As an emerging topic, one of the main issues has been the heterogeneous conceptualization of context (14). Different authors have used different definitions of what context is, including at what physical or geographical level it operates – household, neighbourhood or country will likely all influence health, but probably in different ways. Diderichsen et al. (28), for example, define context as a ‘catch-all phrase used to refer to the spectrum of factors in society that cannot be directly measured at the individual level’ (p. 19), encompassing the ‘structure, culture and function of a social system’ (p. 19). This vague definition highlights the difficulty to homogenise the concept. Porta’s dictionary of Epidemiology (33), on the other hand, defines context as ‘the location of a person by time and place’, referring to both ‘geographical location and to group membership’ (p. 58). This definition brings to focus two components of context that are commonly identified: the physical and the social environment. These environments can not only influence health, but may also influence each other (34). For example, the extent of physical space a community has available will influence how its individuals interact. This distinction brings to light that, by being physical or social, context is not restricted to a geographical definition; contextual characteristics can be defined within a network of peers who interact exclusively online, but who nonetheless share social norms that shape their elements’ health. Another common conceptualization of context distinguishes between ‘compositional’ and ‘contextual’ effects. This distinction arose from geographical analyses that think of place effects as a consequence of the characteristics of the people who reside in a certain place (‘compositional’) and of the characteristics of the place itself (‘contextual’). However, as Macintyre et al. (14) and Frohlich et al. (21) argue, this distinction is not necessarily useful nor correct. In fact, there are complex interdependencies between people (‘composition’) and places (‘context’), as individuals are not placed at random where they live or where they work. As Macintyre (35) put it: ‘people make places and places make people’ (p. 12). The relationship between context and health inequalities Chapter 1. Introduction 10 Another factor hampering the appropriate analysis of contextual influences on health is the lack of clear theorizing about the mechanisms by which context operates (14). This leads to a search for contextual determinants of health that has no basis on a strong theory of how these determinants operate. As Mitchell et al. (36) comment, ‘lack of theory has often resulted in a choice of variables with which to characterize an area which is guided more by what is available ‘off the shelf’ than by careful theoretical consideration’ (p. 68). As a result of unclear definition and theorizing, measurement of context has been, in the least, heterogeneous. This has also dismissed important questions such as which spatial or time scales are appropriate. Indeed, contextual influences can be measured as characteristics of a street, neighbourhood, city, or country, just to name a few. Pathways that link these characteristics to population health and HI will differ according to the geographical scale they are being measured at. For example, while a universal healthcare policy might be a good measure of healthcare access, it does not take into account regional inequalities in the distribution of healthcare services, which can only be detected with a smaller scale analysis. On the other hand, ideological views such as racism and other forms of discrimination may not be detected at such a small scale, but nonetheless be prominent in the country and have an important effect on the health of that population (37). Time scales are also oftentimes dismissed: most analyses measure contextual exposure and health outcome at the same moment in time, but this is often implausible, as exposures can take time to have an effect. For example, air pollution may take decades to have an impact on adult mortality, and this biological plausibility must be taken into account (14). Finally, as described in the first section, context has also been pushed aside from research as a consequence of dominant political views. As Margaret Thatcher famously put it, “there is no such thing as society” (38). Unclear definition, operationalization, and theorizing have considerably complicated the construction of a coherent theory on the influence of context on health and HI. This has led to important critiques, even comparisons with medieval medical theory – Sloggett and Joshi (39) called the contextual influence a ‘social miasma’ (p. 1473). However, despite the weaknesses in the current evidence on this topic, and the critiques of its existence as an issue at all, researchers would tend to agree that where people live matters for their health (40). In a brief review of the evidence, Macintyre The relationship between context and health inequalities Chapter 1. Introduction 11 et al. (14) conclude that ‘rather than there being one single, universal ‘area effect on health’ there appear to be some area effects on some health outcomes, in some population groups, and in some types of areas’ (p. 128). This conclusion suggests that context can sometimes have little impact on overall population health, and a tremendous impact on its distribution between population groups, and hence on HI. Thus, context appears to have an important influence on health and its distribution within a society, but current knowledge is considerably hampered by a number of issues, such as heterogeneous conceptualization and measurement. Additionally, context can have different effects depending on the individual, place, and time. This differential effect suggests that not only it is necessary to take context into account when studying HI, but that it may be key to their understanding. An appropriate theoretical framework that summarizes these relationships would provide the much needed basis on which empirical analysis could build evidence. 1.1.4. Summary Research in HI and the SDH has grown exponentially in Europe since the 1980s, when political interest in the topic first emerged. Academic and political interest in HI evolved through this period of time, with a clear shift of focus from individual determinants, individualistic methodology and ‘risky behaviours’, to social processes and context as determinants of HI. This shift is reflected not only on dominant political thought of industrialized societies, but also on dominant academic thought. This parallel course shows how the academic discourse is not separate from the world that surrounds it, as the views of researchers can be shaped by dominant normative views of the society they are in. The individualist approach that has dominated research on HI has constrained knowledge and hindered the creation of policies that effectively reduce them. This has happened because issues of heterogeneous conceptualization, measurement, and theoretical definition have undermined the study of contextual determinants, which has opened the field to considerable critique. However, contextual determinants are unavoidable, particularly when exploring HI, namely for measurement and conceptual issues. Ultimately, it is likely that context has a complex differential effect that interacts with individual characteristics. This makes context key in the understanding of HI. The relationship between context and health inequalities Chapter 1. Introduction 12 The notion that not all determinants of health are best conceptualized at the individual level has been called the new paradigm of public health. As a new idea, it is still in developing stage, and would benefit greatly from a solid theoretical basis that would allow build-up of knowledge and adequate policy choices. The relationship between context and health inequalities Chapter 1. Introduction 13 1.2. Measurement of Health Inequalities 1.2.1. Which Determinants? For a long time and across many countries, the poor have had worse health and shorter lives than the wealthy (41). This consistency is seen regardless of the major causes of death in society: it is true when communicable diseases are the main killers, and it is true for non-communicable diseases (42). It is seen through the life course, from gestation to birth, childhood, adolescence, adulthood and old age (43). It is seen using a number of different statistical methods and measurements, and it is seen by a number of individual SES measures, like income, education or occupation (44). But the study of contextual determinants has raised both conceptual and methodological questions that cannot be answered by traditional views of individual characteristics and traditional statistical methods alone. Conceptually, recognizing contextual effects on individual health implies a shift in the understanding of how risk factors operate. In traditional epidemiology, individual characteristics are identified as causes or risk factors for ill health, implying that interventions should be focused on the individual. Some critiques of this approach claim that an exclusive focus on the individual can lead to counterproductive processes of victim blaming (45). The new paradigm of contextual effects on health recognizes that the context can be, in and of itself, a determinant of health. Additionally, the acknowledgement of context as having an influence on health also implies that individual characteristics must be framed by the context they are in. In this sense, having a certain amount of money matters not only in absolute terms, but also considering the average wealth of everyone else, i.e., what you have matters, but what others around you have matters too. This implies that HI can be created, in part, of psychosocial mechanisms. Indeed, a noticeable finding in HI research has been that HI do not occur in a threshold effect. Systematic differences are not seen just below a certain point of income, wealth or occupational rank. Rather, there is a socioeconomic gradient in health, in which the poorest have worse health than the ones who earn an average income, who in turn have worse health than the richest (41). This gradient is seen for all SES determinants: financial resources, education and occupation. This gradient shows that socioeconomic determinants do not influence health only because they provide access to essential resources, otherwise there would be no difference once those essential resources were present (44). Rather, there are other effects at The relationship between context and health inequalities Chapter 1. Introduction 14 work through the social spectrum that produce systematic differences in health outcomes. These effects are a consequence of psychosocial mechanisms. While material conditions are widely recognized and important for the health gradient, as they are directly related to access to health promoting resources, such as quality food and housing, extensive research has shown that adverse social conditions can directly lead to adverse biological effects, regardless of access to resources (46). This view helps understand the socioeconomic health gradient. Within a structure of a society, everyone is beneath someone else by some measure – be it money, prestige, cultural capital, or others. This social hierarchy can create stress responses, as being of a lower social status may lead to feelings of inadequacy and lack of control (47). As such, since there is always someone better off, everyone suffers the consequences of this gradient, not only the poorest. These stress responses have a direct impact on biological functioning, and can lead to more health damaging behaviour, such as drinking and smoking. This leads to a health gradient that affects everyone, even after basic material conditions for a healthy life are satisfied. Noticeably, rather than being opposite explanations, material and psychosocial mechanisms act together to help explain the socioeconomic health gradient (44). Psychosocial explanations highlight that, further than being a question of absolute poverty, HI are also a question of relative deprivation. The international glossary of poverty defines relative poverty as the ‘absence or inadequacy of those diets, amenities, standards, services and activities which are common or customary in society’ (48: p. 169). This notion of poverty implies that it can be a socially defined concept, measured within the group the individual is in, as it depends on what is ‘customary in society’. The health gradient and psychosocial explanations show that the creation of HI is complex and a product of multiple determinants operating at the same time. Importantly, these determinants occur at both the individual and contextual level simultaneously, and neither should be ignored. a) Individual Determinants Historically, income, education, occupation and employment have dominated the analysis of HI as determinants of individual SES. These determinants are often used The relationship between context and health inequalities Chapter 1. Introduction 15 interchangeably, but research has shown that this is not necessarily correct, as they can reflect different underlying causal processes (49). i. Education Education is probably the most commonly used measure of SES and has extensively been related to various health outcomes. Comparative analyses between European countries show strong associations of education with cause-specific mortality (50), self-assessed health (SAH) (51), limiting long-standing illness (52), chronic conditions (53), smoking (54, 55) and obesity (56). As a measure of SES, education has a number of important advantages: it is easy to measure, shows high response rates in surveys, is fairly comparable across countries, applicable to both working and non-working individuals, tends to remain stable through life, and is not likely to be affected by reverse causation, since it is usually determined in young adulthood and remains stable throughout life (57). However, reverse causation cannot be completely excluded, since a healthy life expectancy might induce higher investments in education and ill children might be less able to complete education (58). Nonetheless, analyses of compulsory schooling laws in the US and Europe, which ‘force’ most people into education, regardless of their health prospects, suggest that education causes better health, despite the opposite also being true (59-61). Several mechanisms explain the pathway linking education to health. The effect seems to be mediated in part by income and occupation, although analyses show an educational gradient even after controlling for these factors (58). The remaining health differences can be strongly explained by behavioural factors, which in turn seem to be a consequence of better information and better cognitive abilities, which affects the ability to process information regarding healthy behaviours and disease management (62). Preferences also seem to play a part, as they vary systematically across educational groups (62). Finally, education can also provide an individual with a social network of similarly educated peers, which can have substantial health benefits (41). ii. Financial resources Financial resources, such as income or wealth, are also strongly correlated with health, independently of education. They can have an impact on health to the extent that they allow individuals to access health-producing resources, such as healthcare or The relationship between context and health inequalities Chapter 1. Introduction 16 better living conditions. However, assessing causality is difficult, since health strongly increases one’s ability to earn money and several confounding factors – such as education or preferences – might determine both better health and better income. Whereas the effect of health on income has been extensively shown in a number of societies, the effect of income on health remains more of an ‘open question’ (41, 58). Unlike education, financial resources can be more difficult to measure, as people may not be as happy to share this information, there are many components involved (income, wealth, savings, property, etc.), and comparability can be hampered by different currencies and differences in currency value, for example. Nonetheless, income-related health gradients have been shown for SAH (63, 64), functional limitations (64, 65) and smoking (55) in European countries. iii. Employment Employment, or lack of it, is central to most adults’ life and has been associated with health in a number of settings. This association is not surprising. First of all, employment provides income, which is essential for access to basic goods. Unemployment can lead not only to financial strain, but also uncertainty about the future. This leads to a second mechanism: stress. Unemployment, with the uncertainty it brings, leads to a feeling of lack of control, which has been extensively associated with adverse health outcomes (66). Employment also provides psychological benefits – like providing a structure to the day, self-esteem, status and a sense of contribution to a collective cause – that are absent in unemployment (67). Opportunities to socialize are also more common when one is employed, and social support and integration have extensively been linked to health (68). Finally, unemployed individuals seem to have an increased risk of health-damaging behaviours, such as smoking and drinking (69). This might occur because people who drink and smoke are more likely to become unemployed, because people who become unemployed drink and smoke more to deal with their stressful situation, as a consequence of a common causal factor, or a combination of any of these. Regardless, employment is a major factor in HI among working-age adults. Analyses in European countries have shown that unemployment is associated with worse SAH (63), chronic health conditions (70) and mortality (71). The relationship between context and health inequalities Chapter 1. Introduction 17 iv. Occupation The International Labour Organisation (ILO) (72) defines occupation as ‘a set of jobs whose main tasks and duties are characterised by a high degree of similarity’ (p. 1). Accordingly, occupational grades are usually classified according to tasks and responsibilities involved. Probably the most commonly used classification is ILO’s International Standard Classification of Occupations, which defines ten major occupational groups, from elementary occupations to managers, defined in terms of skill level and specialisation required for each occupation (73). Other classifications, such as manual and non-manual (53), administrators, professionals, executives and clerks (74) or white collar and blue collar (75) are sometimes used, but all reflect different degrees of skills that are required for the job. Occupational grade has been strongly associated with health outcomes. In European countries, it has been associated with overall mortality (53, 76), stroke and ischaemic heart disease mortality (77), infant mortality (78), child health (78), SAH, long term limitations and chronic conditions (79). The Whitehall studies are one of the most important contributions to the understanding of this relationship. These studies showed that people in higher ranks had a stronger sense of control over their health, their jobs and their lives (74), which is strongly associated with better health (66). However, occupation can also reflect an individual’s place in society, to a greater extent than financial resources, education or employment do. Having an occupation will usually grant adequate earnings and a certain degree of job security, so the health differences that remain can also be explained by the effect of social standing (rank) and subjective feelings towards one’s position in society (59). In fact, a number of authors have used occupation as a marker of ‘social class’ (80-82), possibly based on the understanding that occupation reflects more than just skill levels and specialisation. v. Social Mobility The SES of an individual can have an impact on their health at any given time and the movement between different social strata can too. Social mobility is the process of moving between social strata, either between generations (parents and children) or within the life-course of the individual (interand intragenerational social mobility, respectively). The relationship between context and health inequalities Chapter 1. Introduction 18 Social mobility has been associated with health in a variety of contexts (83, 84), but the way it operates can be difficult to tease out. One possibility is that it is an accumulation effect. Power et al. (85), for example, showed that in a birth cohort from the UK, the accumulation of unfavourable social circumstances was more important than social mobility per se in determining adult health. This means that, when comparing two people with the same SES, one of whom ‘moves down’ the social ladder and another who remains stable, the first will have worse health outcomes. However, this is not a consequence of the downward social movement itself, rather a consequence of having a lower SES at a given moment in life. Another explanation for the association between social mobility and health is an opposite causal effect: extensive empirical analysis has shown a ‘health selection effect’ that pushes people who are unhealthier down the social ladder (84, 86). This is not surprising, as people who are ill can be less capable to study and work, thus reducing their potential earnings and social position. Finally, some evidence also suggests that social movement in and of itself has some effect on health (41, 83). Surprisingly, it is not just the downward movement that seems to have a negative impact, but also upward movement, particularly within short periods of time, can also have a deleterious effect (41). This unexpected effect may be a consequence of an increase in unhealthy behaviour (such as smoking more because one has more available income), or of physiological and behavioural adaptations to a different social setting. b) Contextual Determinants A multitude of contextual characteristics have been analysed in the literature as potential determinants of health. Building on the definition of context outlined in the previous chapter, these can be classified as physical or social determinants – with the caveat that this is an oversimplified characterisation, as physical and social contextual characteristics often interact with each other. Physical determinants can be thought of in terms of natural environment – air, water, noise, green spaces – and of the built environment – houses, roads, infrastructures, and transport systems. Extensive research has shown a strong connection between the built environment and health. Examples include impacts on mental health (87), physical activity (88), eating habits (89), obesity (90), and drinking (91). This influence can work through the availability of green spaces that allow people to have a more physically active life, walkability of neighbourhoods that help provide a safe The relationship between context and health inequalities Chapter 1. Introduction 25 status measures and other variables that impact mortality (124). Moreover, its capacity to predict mortality has become increasingly more consistent between 1990 and 2002 in the US, possibly because of better health related information (125). On the other hand, despite its strong ability to predict mortality, this relation might vary between gender (126), age (127), and SES (128), possibly biasing results of HI analyses. However, SAH has value on top of its ability to predict mortality: it can be seen as a more comprehensive measure of health, which allows respondents to weight different aspects of their own health and value them according to their own preferences (129). An increasingly used alternative to SAH has been to use ‘health limitations’ as a general measure of health. This measure is also based on a survey question that asks respondents whether they are limited in their daily activities due to a health condition. Possible answers usually include ‘yes, severely limited’, ‘yes, somewhat limited’, and ‘no’. Some authors consider this a ‘quasi-objective’ indicator, more accurate than SAH (65), and some have used it as a proxy measure of disability (130). HI have been observed using both SAH and limitations as an outcome measure (65). These measures rely on self-reporting, but HI have also been observed in morbidity indicators that are based on objective measurements. For example, several cancers (although not all) show a socioeconomic gradient, as does the survival rate after cancer diagnosis (78). Measures of physical ability also tend to show a socioeconomic gradient (131), as does Body Mass Index (BMI) (132) and the metabolic syndrome (133), to mention only a few examples. c) Health Related Behaviours According to the Global Burden of Disease Study, the three risk factors that most contribute to disease burden in western and central Europe are high blood pressure, tobacco smoking, and high BMI (134). These reflect the four major behaviours related to non-communicable diseases: eating, drinking, smoking and exercising. These four behaviours have extensively shown a social pattern, such that people from lower SES tend to show less compliance with dietary and exercise recommendations, drink more, and smoke more (135). These health related behaviours occur within social structures and contexts that can facilitate or hinder them. For example, while ultimately smoking may be a matter of The relationship between context and health inequalities Chapter 1. Introduction 26 individual choice, factors such as the economic ability to buy cigarettes, having a more or less stressful life, and living in an environment where social norms support or shame smoking behaviour, can all exert an influence on that final ‘individual’ decision. Another particular form of health related behaviour is health care use. As with other behaviours, health care use and quality are also socially patterned (136). This can be related to factors like economic resources, access to information or geographical accessibility. However, at least in a European context, health care is far from being the most important cause of HI and probably contributes only slightly to these inequalities (137). 1.2.3. Which Measures? Despite being known for centuries, HI have not always been a unanimously accepted fact. The Black Report dedicated some of its pages into showing that HI were not a product of mathematical artefact (4), but it was still criticized for the measures it used (138). In fact, the measure one chooses to analyse HI can determine the result of the analysis, possibly even leading to contradicting results (138). One of the most commonly used methods are range measures, probably the most simple and easy to interpret. They compare the health status of two groups of the socioeconomic distribution. This is done by calculating absolute or relative differences (i.e., ratios). For example, if half (0.5) of the poorest quintile and a fifth (0.2) of the richest quintile have diabetes, the absolute difference is 30 percentage points (0.5-0.2=0.3) and the ratio is 2.5 (0.5/0.2=2.5). While inequality is summarized in one value, range measures show an incomplete picture, as they only compare two groups (usually top and bottom). Inequality between these two groups may remain the same, while the distribution within the middle groups dramatically changes (138). Another shortcoming of these measures is that they ignore the sizes of the groups. This is important especially in comparative analyses – across time or space – as two similar results may reflect two very different distributions. This is also an issue when socioeconomic groups do not have a fixed size, such as those defined by occupation or education. A range measure of inequality may remain stable over time, but if the group with more years of education increases, while the number of less educated decrease, the distribution is clearly different. The relationship between context and health inequalities Chapter 1. Introduction 27 To overcome some of these shortcomings, regression-based measures can be used. An example is the Slope Index of Inequality (SII): the regression coefficient (the slope) of a regression model of the health outcome on SES, ordered by SES. It can be interpreted as the absolute effect on the health outcome of moving from one SES category or value to the next (139). Despite its simplicity, this measure is sensitive to the population mean of the health outcome, which limits is comparability across populations and time. The Relative Index of Inequality (RII) overcomes this issue, as it is calculated by dividing the SII by the population mean of the health outcome. Measures based on the concentration curve (CC) are also used. These measures originated in the economic analysis of income distribution, famously summarized in the Gini Coefficient – a measure of how income or wealth of a population is distributed among its elements. Equally, when a health outcome measure is used, it is possible to summarize in one number how health is distributed, reflecting not only the experience of two groups, but of the whole population. Additionally, unlike the SII and the RII, measures based on the CC do not assume a linear relationship between independent and dependent variables. An important distinction that can be done in the methods used to calculate HI is the difference between relative and absolute measures. It has been argued that the use of only absolute or relative measures can be misleading, as it can influence the readers’ perception of the magnitude, significance and even direction of HI (140, 141). In fact, methodological reviews of the reporting of HI have shown that absolute and relative measures of the same effect can yield opposite results (141). Additionally, the choice of relative or absolute measures can also reflect different equity value judgments (142). When comparing two groups that get healthier at an equiproportionate rate – i.e., in both groups’ health improves at an X% rate – then a relative measure of HI will remain the same, while an absolute measure will change. On the other hand, if the groups’ health improves in a uniform way – i.e., both groups see an overall improvement of X percentage points – then an absolute measure of HI will not change, while a relative measure will change (142) 1. 1 This is better illustrated in the hypothetical example outlined in appendix 1. The relationship between context and health inequalities Chapter 1. Introduction 28 1.2.4. Which Mechanisms? a) Pathway Between Individual Socioeconomic Status and Health The study of the pathways that link SES to health is comparatively less common than of its determinants, outcomes or measures. In 2001, Diderichsen, Evans and Whitehead outlined a framework for understanding the social origins of HI (28) that summarizes these pathways (figure 3). At that time, literature on the SDH was just escalating and starting to touch the topic of context as an important determinant of health, shying away from the traditional conceptualization of risk factors as individual attributes. Diderichsen’s framework recognizes the role of context in the creation of HI, although it mostly focuses on individual pathways. Probably the most noticeable use of Diderichsen’s framework was its application by the WHO’s CSDH as a basis for their own framework (figure 2) (12, 13). Diderichsen’s framework has also been used as a basis for other frameworks (143); as a map to understand other social phenomena, such as intimate partner violence (144) and social consequences of disease (145); as a frame to present evidence of literature reviews on traffic injuries in youth (146), in cystic fibrosis in the UK (147) and in work-related health (148); as a basis for policy comparisons between countries (145, 149, 150); and as a map for empirical analysis of the health of lone mothers (151, 152) and smoking in adolescents (153). Diderichsen’s framework has not raised much academic discussion, and other authors who used it as a basis for their own work did not explicitly critique its applicability. Despite this, it provides a simple, yet complete summary of how SES influences health and vice-versa. The framework describes four mechanisms: (I) social stratification, (II) differential exposure, (III) differential susceptibility and (IV) differential consequences; and four entry points for policies to target those mechanisms: (A) influencing social stratification, (B) decreasing exposures, (C) decreasing vulnerability, and (D) preventing unequal consequences. With this framework, the authors aim to provide a model to understand the process of creation of HI, a systematization of what kind of policies might work in their mitigation, and a base to empirically test which HI producing mechanisms are more important in a society. The relationship between context and health inequalities Chapter 1. Introduction 29 This framework provides a simple, yet exhaustive, description of how SES and health outcomes are interrelated. It can be easily applicable to any SES measure and health outcome, as the variety of analyses it has been used for clearly shows. It also clearly identifies the steps in which differential effects occur, which in turn lead to HI. Figure 3. A framework for elucidating the pathways from the social context to health outcomes and for introducing policy interventions. Source: Diderichsen F, Evans T, Whitehead M. The social basis of disparities in health: Challenging inequities in health: from ethics to action. New York: Oxford University Press; 2001. i. First Mechanism: Social Stratification Social stratification is the way context determines individual social position (figure 3). ‘Context’ is understood by Diderichsen (28) as ‘the spectrum of factors in society that cannot be directly measured at the individual level’ (p. 19). As was later described by the author (154), the social stratification mechanism actually encompasses two mechanisms: ‘one that generates and distributes wealth and power to different social positions in society and one that stratifies individuals into the social positions’ (p. 59). Thus, the pattern of health across social positions is a result of both the characteristics of the positions and of the individuals occupying them. The relationship between context and health inequalities Chapter 1. Introduction 30 Under the author’s understanding, the process of social stratification occurs mostly during early childhood development (155); at this stage, a range of determinants operates to determine social position and health in adult life. For example, in a sample of the Portuguese population, height (as a proxy measure of childhood social circumstances) was strongly associated with several health outcomes, such as asthma and chronic pain, such that taller people (who tended to have had better childhood circumstances) were healthier (156). The existence of life-long determinants highlights the importance of a life-course approach to the analysis of HI. Policies that influence social stratification usually fall out of the realm of ‘health policies’. These are policies that decrease social inequalities, such as creating equal educational opportunities or redistributing wealth. Because social position is inextricably linked to health, reductions in social inequalities have the potential to also reduce HI. As poverty has also been extensively linked to health, these policies can be particularly important in protecting the most vulnerable – through, for example, illness pensions. ii. Second Mechanism: Differential Exposure Differential exposure refers to the unequal distribution in type, amount or duration of exposures that impact health on different social groups. Unequal exposures comprise environmental, biological or behavioural risk factors, which are commonly connected to social position. For example, an unskilled worker may have a low income that does not allow them to choose a healthy diet; a person living in an urban environment is exposed to more air pollution; and the stress of having a low income can make a person more prone to smoke. Whilst social stratification operates mainly in early life, differential exposures can occur in childhood or adult life, as a consequence of the social stratification process. Many health policies that employ risk reduction strategies – for example, media campaigns to promote healthy eating by informing people of what constitutes a healthy diet – do not differentiate between social groups. However, health campaigns reach individuals differently, as the more educated can be more exposed. Accordingly, analyses of these strategies have shown that, despite improving some people’s health, they sometimes increase HI, as they disproportionately improve the health of those that are healthier to start off with (157). The relationship between context and health inequalities Chapter 1. Introduction 31 To reduce exposures, one must ‘modify the effect of social position on the occurrence of the specific causes’ (154: p.60). This means strategies should aim to reduce the disproportionate amount of risk factors that some people face, focusing on particular groups that face them, for example, by protecting them from occupational risks, bad housing, or inadequate nutrition. Noticeably, differential exposures tend to cluster, as people in lower social strata are more likely to be exposed to multiple adverse risk factors. This clustering of adverse exposures through life constitutes the third mechanism: differential vulnerability. iii. Third Mechanism: Differential Vulnerability The effect of an exposure on an individual is not exclusively a function of the exposure, but also of the individual him/herself. For Diderichsen et al. (155), the added exposures to multiple risk factors increase the vulnerability of people in lower social positions. These exposures act synergistically, making the individual more susceptible to the effect of each of them. This means that even if a given risk factor is distributed evenly across social groups, its impact may be unevenly distributed among those groups due to different underlying vulnerability (28). Thus, the third mechanism is ‘mostly a question of clustering to lower socio-economic groups of causes in the same pathway’ (154: p.60). The success of policies that aim to decrease risky exposures also depends on the existence of other exposures and on the context in which the individual is in. For example, a policy might be put in place to provide housing to everyone, but a homeless individual with an incapacitating mental illness and no other form of support will not be able to navigate the administrative process to apply for the house they need. Reducing vulnerability depends on just that: tackling interacting exposures and not just focusing on a single one. However, vulnerability is not only about the additive or interacting effect of several exposures, it is also about contextual effects. As the Diderichsen points out (158), ‘children living in extreme poverty have very different mortality rates in different countries, which shows that the national policy context modifies the effect of poverty’ (p. 14). For example, a society with strong social cohesion can potentially reduce the effects of poverty or unemployment in an individual’s health, by reducing the stress associated with these situations. The relationship between context and health inequalities Chapter 1. Introduction 32 Thus, reducing vulnerability must include comprehensive strategies that tackle multiple exposures simultaneously, and providing people with an environment that helps mitigate the effects of individual risk factors. iv. Fourth Mechanism: Differential Consequences Health is a fundamental good, valuable not only for its own sake, but also for how it allows individuals to fulfil their expectations about their lives. Illness can have social and economic consequences, which can feed back into the mechanism of social stratification (mechanism I in figure 3). For example, loss of a limb may lead to extensive healthcare payments and also to lost income due to inability to work; these direct and indirect costs can have tremendous consequences in a family’s budget. However, this impact depends on other determinants, such as how well off the family was before the health event or the social support that is available – such as universal healthcare or disability insurance. Depending on these, the consequences can be more or less grave, changing the likelihood of ‘falling behind’. The consequences of disease can also have an effect on an aggregate level, as high rates of illness can influence a country’s social and economic development. An example of this is the impact of HIV in South Africa: in 2000, the World Bank projected that Gross Domestic Product (GDP) would be 17% lower in 2010 due to the high rates of the disease (159). Policies aimed at reducing the social consequences of disease include those related to the provision of healthcare (including primary, secondary and tertiary), and those aimed at mitigating the economic consequences of disease, such as providing work opportunities, benefits or insurance to people who are ill. These policies, by providing healthcare and promoting reintegration in the workforce after a person falls ill, have the potential to break the cycle between SES and health. b) Pathways Between Context and Health and its Distribution Despite extensive evidence connecting multiple contextual characteristics with individual and population health, the pathways between the two have not been particularly explored (160). Macintyre et al. (14) called this the ‘black box of places’ (p. 131), an unknown influence that we can see but cannot explain. The framework of the CSDH tries to fill this gap. According to this framework, context creates social stratification and distributes individuals through strata, which The relationship between context and health inequalities Chapter 1. Introduction 33 have an impact on their health (13). In this view, context is composed of governance, policy (macroeconomic, social and health), and cultural and societal norms and values (figure 2) (12). This framework took some ideas from Diderichsen’s framework, whose description of the pathways between context and health was more detailed. In fact, Diderichsen’s framework can be approached from a perspective of the individual or of society (150). When viewed from the perspective of society, Diderichsen’s framework identifies two processes through which society impacts health and HI: social stratification and policy entry-points. In Diderichsen’s words (28), ‘[the process of social stratification] both allocates power and wealth to social positions and allows individuals into different positions’ (p. 21). Diderichsen argues that the process of social stratification is ‘central’ to the issue of HI, in so much as it helps us to distinguish between ‘fair’ and ‘unfair’ inequalities (i.e., inequalities versus inequities). The author argues that, once one understands the process of social stratification, one can judge it as to whether it is ‘fair’ or ‘unfair’. As such, if power and wealth are distributed ‘fairly’ among social positions of a society and if individuals are ‘fairly’ allocated to those positions, HI stemming from that process can also be deemed ‘fair’ (i.e., not inequities2). It remains, nonetheless, that classifying something as ‘fair’ or ‘unfair’ is a normative exercise, and thus depends on views of justice. Context can also influence health and health distribution to the extent that it creates informal and formal rules that influence health, such as policies. For example, universal access to health care can help equalize opportunities, as it helps individuals who have fallen ill to recover, re-enter the workforce and avoid a fall in social position. Informal rules can also have a health impact, such as when expectations about a neighbourhood allow it to maintain high levels of violence. Furthermore, these rules can impact different groups differently. This influence of context is acknowledged in Diderichsen’s framework to a large extent through the policy ‘entry points’ he identifies (decreasing exposures, decreasing vulnerability and preventing unequal consequences). By identifying these entry-points, the author describes how 2 The distinction between health inequities and inequalities is not universally accepted, but it is generally considered to be a question of whether a moral judgement is made or not: while inequality and equality are purely descriptive – simply describe a difference between two groups – inequity and equity encompass a moral view – not only there is a difference, but it is an unjust difference. The relationship between context and health inequalities Chapter 1. Introduction 34 policies, by shaping the rules within a society, influence its health and health distribution. In summary, according to Diderichsen’s framework, context influences health and health distribution through two major mechanisms: social stratification and policy entry points. Social stratification operates through the creation of hierarchies in a society, whereas policy entry points influence the pathways between social position and health. However, as Whitehead, Burström and Diderichsen (150) point out, policy is only one component of social context that ‘may have an influence on the pathways between social position and health’ (p. 257). A comprehensive societal perspective needs to focus on how all contextual components – not just policy – influence health and its distribution. It remains that, despite the general agreement that context influences health and HI, lack of understanding of the mechanisms by which this happens undermines the evidence that is produced (14, 160). This is further reinforced by the observation that various public health policies seem to have failed to reduce HI, even when this was their explicit goal (25, 161). This shows a need to systematize the mechanisms by which context influences health and HI, to better understand and study them and to design policies that are appropriate for their outlined goals. 1.2.5. Summary The choice of health outcome, SES indicator, or measure for the analysis of HI, can reflect different research questions and different normative views, and can ultimately lead to different conclusions. Despite this, HI have been shown for a variety of health outcomes – from mortality, to morbidity, and health behaviours –, of SES indicators – both individual and contextual –, and using a variety of different measures. The mechanisms that link individual SES to health outcomes can be described in light of Diderichsen’s framework, which describes a pathway that starts in social stratification – the way that context determines individual social position. Social position then goes on to determine exposure to risk factors, vulnerability to those factors, and the consequences of health back on social position. Diderichsen summarises how context influences HI in two mechanisms: social stratification and policies. However, a comprehensive perspective on context must include other The relationship between context and health inequalities Chapter 1. Introduction 41 Portuguese constitution, interest in health equity has been limited. A WHO report on the Portuguese National Health Plan argued that HI were an ‘important policy gap’ in this plan (202). In an interesting analysis of this seeming lack of interest, Bago d’Uva argues that it is the absence of explicit and effective policies to tackle HI that allows them to remain so prevalent (203). 1.3.3. Summary Despite considerable improvements in overall health and in social support in European countries, HI remain high, in some cases even increasing in the last years. This has been called a ‘paradox’, as HI remain an important public health challenge in the continent. Portugal is a particularly interesting case study of HI, as it is one of the most unequal European countries, both in terms of income and health distribution. Current evidence suggests the existence of significant pro-poor inequality in most health outcomes and for most socioeconomic determinants, with a few noticeable exceptions, such as allergies and smoking. On top of this, Portugal has gone through important changes in the last decade, namely with the implementation of austerity measures that seem to have had an impact on the provision of social services, thus potentially increasing HI. Despite this, political attention to the issue in Portugal is still very limited, which may be exactly why HI remain so high. The relationship between context and health inequalities Chapter 1. Introduction 42 1.4. Objectives The overall aim of this dissertation is to contribute to the understanding of how characteristics of the context can have an impact on population health and health distribution, using Portugal and Europe as case studies. To accomplish this aim, the following objectives were pursued: • To analyse and interpret how changes in individual and contextual social capital, and the interaction between the two, were associated with changes in SAH in European countries between 2002 and 2012. • To analyse and interpret how the relationship between social mobility and HI varied between six welfare regime types in European countries between 2002 and 2012. • To collect, summarise, describe, and interpret available evidence about socioeconomic HI in Portugal. • To outline how social inequalities in health limitations changed in Portugal between 2004 and 2014, considering different measures of social status, and interpret the results in the light of macroeconomic changes in the country in that period. • Drawing on the findings from the previous analyses, to outline a theoretical framework that summarises how context influences population health and health distribution. The relationship between context and health inequalities Chapter 1. Introduction 43 1.5. Methodological Approaches This research used secondary data collected between 2002 and 2014 in European countries. According to the rules and regulations of the Ethical Council of the Institute of Hygiene and Tropical Medicine of the Nova University of Lisbon, this research did not require ethical approval from the Council (204). 1.5.1. Data Sources This work was based on data from two main databases: the European Social Survey (ESS) and the European Union Survey on Income and Living Conditions (EU-SILC). a) European Social Survey The ESS is a repeated cross-sectional survey that collects data on attitudes, beliefs and behaviour patterns (205). The survey is applied every two years in more than thirty European countries since 2002. Countries are free to participate or not every year the survey is performed. The survey is jointly funded by the EU and each participating country (206). Data is made freely available online for researchers, upon a simple registration process on the website (207). ESS aims to achieve a representative sample on each country of persons over 15 years old living in private households. Each country is given the freedom to choose their preferred sampling design, considering costs, experience and other country specific factors. However, a few requirements are applied to every country to ensure comparability of the samples, such as the use of strict random probability methods at every stage, a high response rate (minimum 70%), a full coverage of the population, and a minimum effective sample size (n=1,500 or n=800 in countries where the population is smaller than 2 million) (208). The ESS questionnaire consists of a core section and a rotating section. The core module fulfils ESS’s primary role, of monitoring change in values and attitudes in Europe through time. The rotating modules are selected based on a call for proposals made in the Official Journal of the EU. In 2014, the rotating modules were on ‘Social inequalities in health and their determinants’ and on ‘attitudes towards immigration and their antecedents’ (209). The ESS core questionnaire includes two health related questions: SAH (‘How is your health in general?’) and hampered in daily activities The relationship between context and health inequalities Chapter 1. Introduction 44 (‘Are you hampered in your daily activities in any way by any longstanding illness, or disability, infirmity or mental health problem? If yes, is that a lot or to some extent?’). ESS results have been widely used in academic research and in policy documents. The ESS bibliography includes hundreds of publications that show the range of use this survey has had (210), including not only analyses of survey results, but also broad discussions on survey methodology. b) European Union Survey on Income and Living Conditions EU-SILC is an annual survey carried out in several European countries with a mixed longitudinal and cross-sectional design. The survey is regulated by EU law as a way of compiling comparable data on income, poverty and social exclusion within the union, using harmonized methods and definitions (211). As such, every EU country is expected to participate, by setting up its own data collection or using existing surveys that comply with EU-SILC requirements. The EU funds the first four years of data collection for each member state. Data is made available for scientific purposes given the researchers’ compliance with a number of privacy requirements. To access this data, authorisation was sought from and provided by Eurostat’s Microdata Access Team. The survey’s longitudinal component follows a simple rotational design: in year one, a cross-sectional sample is selected; this sample is divided in four sub-samples, each itself representative of the population. In year two, one sub-sample is dropped, the other three are followed up, and one new sub-sample is added. As such, except for the three first sub-samples, every sub-sample is requested to be part of the survey for four consecutive years. In any one particular year, the overall sample made up of four subsamples, which make up the cross-sectional sample (212). EU-SILC aims to interview a representative sample of people aged 16 or over living in a private household. Sample size depends on country size; for example, minimum effective sample size for the cross-sectional sample in Portugal is 10,500 and 7,500 for the longitudinal component. This adds up to a minimum sample size of 166,000 per year in the cross-sectional sample, when all countries are combined (211). Just like the ESS, sampling procedures can be defined within each country, as long as they follow certain requirements that make the samples comparable, such as the use of probability sampling and of households as the basic unit of sampling (213). The relationship between context and health inequalities Chapter 1. Introduction 45 Most of the EU-SILC questionnaire concerns living conditions, poverty, and income. Three health related questions are included: • SAH: ‘How is your health in general?’ allowing following answers: o Very good / Good / Fair / Bad / Very bad / Don’t know / Refusal to answer • Chronic conditions: ‘Do you have any long-standing illness or health problem?’ allowing following answers: o Yes / No / Don’t know / Refusal to answer • Limitations by health conditions: ‘For at least the past six months, to what extent have you been limited because of a health problem in activities people usually do?’ allowing following answers: o Severely limited / Limited but not severely / Not limited at all / Don’t know / Refusal to answer Eurostat uses this last variable – limitations by health conditions – as a measure of disabilities (212). EU-SILC results are a fundamental source of data for the work of the European Commission. They have been extensively used in the production of books, scientific publications, political statements, and statistical working papers, among others (214). 1.5.2. Methods a) Systematic Review of the Literature A systematic review (SR) can be defined as a ‘review of the scientific evidence which applies strategies that limit bias in the assembly, critical appraisal, and synthesis of all relevant studies on the specific topic’ (33: p.276). SR are important given the outstanding number of scientific publications and the amount of existing evidence, sometimes contradictory, on a particular topic (215). Thus, the value of SR has long been established, especially for clinical practice, as they provide quick and easy access to a summary of the available evidence to busy clinicians (216). Public health interventions are usually more complex than the traditional randomized clinical trial. They can act on different levels simultaneously – directly with the individual or on a broader, contextual level – and through multiple mechanisms. Sometimes, randomized interventions are not feasible for practical or ethical reasons. The relationship between context and health inequalities Chapter 1. Introduction 46 This means that the traditional methodological guidance for SR can be insufficient to face this complexity (215), and public health researchers have to be creative when doing a SR. This includes resorting to alternative study designs, such as qualitative or observational studies (217). This presents its own issues, as observational studies can find spurious associations, and, in this situation, the exploration of heterogeneity might yield better insights than attempts to look for an overall measure of effect (218). To help overcome limitations associated with SR, the PRISMA guidelines systematise overarching principles that should be applied (PRISMA stands for Preferred Reporting Items for Systematic Reviews and Meta-Analyses). These guidelines (216) consist of a 27-item checklist and a flow diagram outlined by experts and, as much as possible, based on available evidence, that aim to ‘improve the reporting of SR and meta-analysis’ (p. 2). The proposed diagram (figure 4) suggests that all steps that lead to the inclusion of a certain number of studies should be recorded and explained in detail, including sources and reasons for exclusion. The 27-item checklist includes guidance on all elements of the SR, from the title, abstract, introduction, methods, results, discussion and funding. Figure 4. Flow of information through the phases of a systematic review proposed by the PRISMA statement. Source: Moher D et al. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ 2009;339(7716):332. The relationship between context and health inequalities Chapter 1. Introduction 47 b) Statistical Methods i. Logistic Regression Regression analyses are an important part of research in HI. Binary logistic regression analyses are a particular type of regression analyses that concern themselves with binary outcomes, i.e., when the health variable only takes two distinct values. Although the general principles of the logistic regression are similar to those of the linear regression, there are differences in the form of the model and its assumptions (219). The goal of any regression analysis is to find the best fitting, most parsimonious, and interpretable model of a relationship between a set of explanatory variables and an outcome variable (219). Once fitted, the selected model yields an estimate for each coefficient for each of the independent variables included. This coefficient can be interpreted as a rate of change – of ‘a function of the dependent variable per unit of change in the independent variable’ (220: p.49), holding all other independent variables constant. In practice, the coefficient is the difference between the log of the odds for two different values of the independent variable. In the analysis of HI, the independent variables are usually the SES variables of interest, such as income or education. Changes in the scales of the log-odds are not easy to interpret; however, logistic regression coefficients can be easily converted to OR. These, on the other hand, are easily interpreted as the ratio of the odds of the outcome variable between two groups defined by the independent variable (220). For example, if the health outcome is mortality (y=1 for death, y=0 for survival), and education is the independent variable (x=0 for less than high education, x=1 for high education), an OR=2 means that the odds of dying among the less educated are two times higher than the odds of dying among the more educated. This is also applicable for continuous independent variables, considering a 1-unit or an x-unit increase (220). Using the same example, but if instead of education we use income, an OR=2 may mean that for every increase in unit of income (1 €, 1.000 £, 100 $, depending on the scale used), the odd of death decrease by half. ii. Multilevel Models Ecological variables have been dismissed by epidemiology for a long time, particularly since the identification of the ecologic fallacy, which states that individual relationships cannot be inferred from ecologic relationships (15). The relationship between context and health inequalities Chapter 1. Introduction 48 Conversely, in the other extreme, completely ignoring the effects of contextual characteristics can lead to the individualistic (or atomistic) fallacy – ‘an erroneous inference about causal relationships in groups of people made on the basis of relationships observed in individuals’ (33: p. 11). To answer this need, statistical methods must include both individual and group level characteristics, i.e., they must be multilevel. Multilevel methods are regression-based models that take into account the hierarchical structure of the data, for example, at the individual and neighbourhood level. The concept of health determinants as both individual and contextual naturally leads to a multilevel perspective. Several other reasons also warrant the use of multilevel analyses when studying SDH. First, multilevel models allow for the simultaneous estimation of regression coefficients of variables at multiple levels. This is essential, as the result of a single level analysis might in fact be an artefact, reflecting a relationship that truly only exists at a different level (18). Second, they can disentangle complex questions of contextual and individual heterogeneity. While contexts can have an overall effect on population health, they might do so by affecting only a particular group (contextual heterogeneity); on the other hand, the variability of a health outcome within a context may be very different among different groups (individual heterogeneity). Third, multilevel models allow for the analysis of interactions between variables at different levels (individual-contextual interactions). These are important as the same context may sometimes have opposite health effects on different groups (221). Fourth, they can sequentially include multiple levels of hierarchical clustering, from individual, to households, communities, and regions, for example (222). They also allow for more complex data structures, such as crossclassified and multiple membership, which allow individuals to be assigned to multiple groups simultaneously (223). Finally, multilevel models can also take into account the time dimension, such as when observations are nested within time variables (such as year), which are then nested within region, for example (18). iii. Concentration Index The CIx is a measure of HI based on the CC. The CC is the result of plotting the cumulative percentage of individuals, ranked by income, with the cumulative percentage of the health variable. Figure 5 shows a hypothetical example of a CC. In plotting a CC, perfect equality is represented by a diagonal line, showing a perfectly The relationship between context and health inequalities Chapter 1. Introduction 49 equal distribution of health among the sample population, regardless of income. The CIx is calculated as twice the area between the CC and the line of perfect equality. When there is no inequality, health is distributed equally within the population, the CC coincides with the diagonal and the CIx is zero. In the other extreme, if health is concentrated in one person, the CC is shaped like an inverted ‘L’, and the CIx is 1 (or -1). Figure 5. Hypothetical concentration curve. The x-axis represents the cumulative proportion of the population, ranked from poorest to richest. The y-axis represents the cumulative proportion of the health variable. In this example, health is disproportionately concentrated among the richest, as, for example, the poorest half of the population only have about 25% of the health variable. Source: author’s own elaboration. In the particular case of binary health outcomes, the CIx is not limited by the (-1,1) range, but depends on the mean of the outcome variable in the population. As this limits comparability between different populations (across time or between areas, for example), Wagstaff (2005) proposed a ‘normalisation’ of the CIx, by which it is divided by 1 minus the mean, making it comparable (224). Wagstaff, van Doorslaer, and Watanabe (2003) have shown that the CIx can be decomposed into the contributions of individual factors to the income-related HI (225). These factors are usually demographic variables, such as age and sex, or other SES, such as education or occupation. The contribution of each factor depends on two characteristics: the elasticity of that factor with respect to the health variable, and the The relationship between context and health inequalities Chapter 1. Introduction 50 degree of income-related inequality of that factor (i.e., the CIx of that factor). The contribution of each factor to the overall CIx is the product of the health elasticity and the CIx of that factor. 1.5.3. Summary The existence of a social gradient in health shows how health is not determined only by individual SES, but depends on the context the person is in. This conceptual approach cannot be explored using only traditional statistical methods, but needs more advanced statistical models to consider the complexity and different dimensions. Multilevel analysis allows for the description of both individual and contextual characteristics and for the quantification of their impact on health. While not excluding other statistical methods, such as simple regression analyses and methods based on the CC, multilevel models can better reflect the hierarchical nature of health determinants. The CIx, on the other hand, allows for a different analysis of HI. While not including contextual determinants, it reflects the overall experience of the population. This research used two regression-based methods: single level and multilevel logistic regressions. It was also used measures based on the CC, namely the CIx and its decomposition. Furthermore, a SR of the literature was performed, following the PRISMA guidelines. The relationship between context and health inequalities Chapter 1. Introduction 57 113. Johansson E. A note on the impact of hours worked on mortality in OECD countries. The European Journal of Health Economics. 2004;5(4):335-40. 114. Ruhm CJ. Are recessions good for your health? The Quarterly journal of economics .115.2 (2000): 617-650. 115. Catalano R, Goldman-Mellor S, Saxton K, Margerison-Zilko C, Subbaraman M, LeWinn K, et al. The health effects of economic decline. Annual review of public health. 2011;32. 116. Karasek RA, Theorell T. Healthy work: stress, productivity, and the reconstruction of working life. Basic books; 1992. 117. Bruckner TA. Economic antecedents of prone infant sleep placement among black mothers. Annals of epidemiology. 2008;18(9):678-81. 118. McCrone A, Usher E, Sonntag-O’Brien V, Moslener U, Grüning C. Global trends in renewable energy investment 2012. Frankfurt School UNEP Collaborating Centre for Climate and Sustainable Energy Finance. 2012. 119. Buonocore J. Mitigation and health: Climate policy not so costly. Nature Climate Change. 2014;4(10):861-2. 120. Bacigalupe A, Escolar-Pujolar A. The impact of economic crises on social inequalities in health: what do we know so far? International journal for equity in health. 2014;13(1):52. 121. Chen L, Berlinguer G, Evans T, Whitehead M, Diderichsen F, Bhuiya A, et al. Health equity in a globalizing world. In: Evans T, editor. Challenging inequities in health: from ethics to action. New York: Oxford University Press; 2001. p. 34-44. 122. Wilkinson RG. Socioeconomic determinants of health. Health inequalities: relative or absolute material standards? BMJ. 1997;314(7080):591. 123. Zajacova A, Dowd JB. Reliability of self-rated health in US adults. American journal of epidemiology. 2011;174(8):977-83. 124. Idler EL, Benyamini Y. Self-rated health and mortality: a review of twenty-seven community studies. Journal of health and social behavior. 1997:21-37. 125. Schnittker J, Bacak V. The increasing predictive validity of self-rated health. PloS one. 2014;9(1):e84933. 126. Singh-Manoux A, Guéguen A, Martikainen P, Ferrie J, Marmot M, Shipley M. Self-rated health and mortality: short-and long-term associations in the Whitehall II study. Psychosomatic medicine. 2007;69(2):138-43. 127. Franks P, Gold MR, Fiscella K. Sociodemographics, self-rated health, and mortality in the US. Social science & medicine. 2003;56(12):2505-14. 128. Burström B, Fredlund P. Self rated health: Is it as good a predictor of subsequent mortality among adults in lower as well as in higher social classes? Journal of epidemiology and community health. 2001;55(11):836-40. 129. Bjorner JB, Fayers P, Idler E. Self-rated health. Assessing quality of life. 2005:309-23. 130. Eurostat. Disability statistics introduced 2015. Available from: http://ec.europa.eu/eurostat/statisticsexplained/index.php/Disability_statistics_introduced. [Accessed June 1st 2017] 131. Avendano M, Aro AR, Mackenbach JP. Socio-economic disparities in physical health in 10 European countries. Mannheim Research Institute for the Economics of Aging; 2005. 132. Rennie K, Jebb S. Prevalence of obesity in Great Britain. Obesity reviews. 2005;6(1):11-2. The relationship between context and health inequalities Chapter 1. Introduction 58 133. Brunner E, Marmot M, Nanchahal K, Shipley M, Stansfeld S, Juneja M, et al. Social inequality in coronary risk: central obesity and the metabolic syndrome. Evidence from the Whitehall II study. Diabetologia. 1997;40(11):1341-9. 134. Lim SS, Vos T, Flaxman AD, Danaei G, Shibuya K, Adair-Rohani H, et al. A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, 1990–2010: a systematic analysis for the Global Burden of Disease Study 2010. The Lancet. 2013;380(9859):2224-60. 135. Okechukwu C, Davison K, Emmons K. Changing health behaviors in a social context. In: Berkman LF, Kawachi I, Glymour MM, editors. Social epidemiology. Second edition. New York: Oxford University Press; 2014. p. 365-395. 136. Van Doorslaer E, Masseria C, Koolman X, Group OHER. Inequalities in access to medical care by income in developed countries. Canadian medical association journal. 2006;174(2):177-83. 137. Stirbu I. Inequalities in health, does health care matter? Social inequalities in mortality in Europe, with a special focus on the role of the health care system. Rotterdam: Erasmus MC; 2008. 138. Wagstaff A, Paci P, Van Doorslaer E. On the measurement of inequalities in health. Social science & medicine. 1991;33(5):545-57. 139. Regidor E. Measures of health inequalities: part 1. Journal of epidemiology and community health. 2004;58(10):858. 140. King NB, Harper S, Young ME. Use of relative and absolute effect measures in reporting health inequalities: structured review. BMJ. 2012;345:e5774. 141. Kelly MP, santé CdlOsldsdl, Bonnefoy J. The social determinants of health: developing an evidence base for political action. National Instititute for Clinical Excellence; 2007. 142. Allanson P, Petrie D. On The Choice Of Health Inequality Measure For The Longitudinal Analysis Of Income‐Related Health Inequalities. Health economics. 2013;22(3):353-65. 143. Due P, Krølner R, Rasmussen M, Andersen A, Damsgaard MT, Graham H, et al. Pathways and mechanisms in adolescence contribute to adult health inequalities. Scandinavian Journal of Public Health. 2011;39(6 suppl):62-78. 144. Larsen MM. Health Inequities Related to Intimate Partner Violence Against Women. The Role of Social Policy in the United States, Germany, and Norway. Switzerland: 2016. 145. Whitehead M, Hanratty B, Burstrom B. 3.1 Unequal consequences of ill health: researching the role of public policy. In: Graham H, editor. Understanding Health Inequalities. 2009. p.181-198. 146. Laflamme L, Diderichsen F. Social differences in traffic injury risks in childhood and youth—a literature review and a research agenda. Injury Prevention. 2000;6(4):293-8. 147. Taylor-Robinson D, Diggle P, Smyth R, Whitehead M. Health Trajectories in People with Cystic Fibrosis in the UK: Exploring the Effect of Social Deprivation. In: Burton-Jeangros C, Cullati S, Sacker A, Blane D, editors. A Life Course Perspective on Health Trajectories and Transitions. London: Springer; 2015. p. 85-110. 148. Toivanen S. Work-related inequalities in health. Centre for Health Equity Studies. 2007. 149. Phipps S. Health outcomes for children in Canfrrada, England, Norway and the United States. Social indicators research. 2007;80(1):179-221. The relationship between context and health inequalities Chapter 1. Introduction 59 150. Whitehead M, Burström B, Diderichsen F. Social policies and the pathways to inequalities in health: a comparative analysis of lone mothers in Britain and Sweden. Social science & medicine. 2000;50(2):255-70. 151. Fritzell S, Weitoft GR, Fritzell J, Burström B. From macro to micro: the health of Swedish lone mothers during changing economic and social circumstances. Social science & medicine. 2007;65(12):2474-88. 152. Burstrom B, Whitehead M, Clayton S, Fritzell S, Vannoni F, Costa G. Health inequalities between lone and couple mothers and policy under different welfare regimes–the example of Italy, Sweden and Britain. Social science & medicine. 2010;70(6):912-20. 153. Schnohr CW, Kreiner S, Rasmussen M, Due P, Diderichsen F. School-related mediators in social inequalities in smoking: a comparative cross-sectional study of 20,399 adolescents. International journal for equity in health. 2009;8(1):17. 154. Diderichsen F. Policies for health equity: an ethical and epidemiological framework and targets for a new national health policy in Sweden. Health Care Priority Setting. 2003:57-62. 155. Diderichsen F, Andersen I, Manuel C, Andersen A-MN, Bach E, Baadsgaard M, et al. Health Inequality-determinants and policies. Scandinavian journal of public health. 2012;40(8 suppl):12-105. 156. Perelman J. Are chronic diseases related to height? Results from the Portuguese National Health Interview Survey. Economics & Human Biology. 2014;15:56-66. 157. Freimuth VS. The chronically uninformed: Closing the knowledge gap in health. In: Ray EB, Donohew L, editors. Communication and health: Systems and applications. Hillsdale, New Jersey: Lawrence Erlbaum Associates; 1990:171-86. 158. Diderichsen F. Resource allocation for health equity: issues and methods. World Bank. 2004. 159. Arndt C, Lewis JD. The macro implications of HIV/AIDS in South Africa: a preliminary assessment. South African Journal of Economics. 2000;68(5):380-92. 160. Galea S. Integrative chapter: macrosocial determinants of population health. In: Galea S, editor. Macrosocial Determinants of Population Health. New York: Springer; 2007. p. 295-7. 161. Mackenbach JP. The persistence of health inequalities in modern welfare states: the explanation of a paradox. Social science & medicine. 2012;75(4):761-9. 162. World Health Organisation. European Health For All Database. Available from: http://data.euro.who.int/hfadb/. [Accessed on June 1st 2017]. 163. Glatzer M. Welfare state growth and the current crisis in Portugal: Social spending and its challenges. Institute of European Studies. 2012. 164. Guillén A, Álvarez S, Silva PAE. Redesigning the Spanish and Portuguese welfare states: the impact of accession into the European Union. South European Society and Politics. 2003;8(1-2):231-68. 165. Van Kerm P, Alperin MNP. Inequality, growth and mobility: The intertemporal distribution of income in European countries 2003–2007. Economic Modelling. 2013;35:931-9. 166. Hespanha P, Portugal S. Welfare Cuts and Insecurity under the Rule of Austerity: the Impact of the Crisis on Portuguese Social Services. Oñati Socio-Legal Series. 2015;5(4). 167. OECD. Social and Welfare Issues: Inequality 2014. Available from: http://www.oecd.org/social/inequality.htm. [Accesses June 1st 2017] 168. Carrilho MJ. O declínio recente da mortalidade infantil em Portugal. Revista do Centro de Estudos Demográficos. 1985;27:155-9. The relationship between context and health inequalities Chapter 1. Introduction 60 169. Lucas JS. Iniquidade social perante a doença e a morte em Portugal — 1985. Revista do Centro de Estudos Demográficos. 1986;n.º 28:29. 170. Pereira JACG. Equity, health and healthcare: an economic study with reference to Portugal. PhD dissertation. University of York; 1995. 171. Gotsens M, Marí-Dell'Olmo M, Pérez K, Palència L, Martinez-Beneito M-A, Rodríguez-Sanz M, et al. Socioeconomic inequalities in injury mortality in small areas of 15 European cities. Health & place. 2013;24:165-72. 172. Simões A, Paquete A, Araújo M. Equidade Horizontal no sistema de saúde português: sector público vs sector privado. Associação Portuguesa de Economia da Saúde. 2006. 173. Simões N, Crespo N, Moreira SB, Varum CA. Measurement and determinants of health poverty and richness: evidence from Portugal. Empirical Economics. 2016;50(4):1331–1358. 174. Ribeiro S. Desigualdades Socioeconómicas na Doença Cardiovascular em Portugal: Estudo baseado no Inquérito Nacional de Saúde 2005/06. Masters Thesis. Nova University of Lisbon; 2010. 175. Ribeiro O, Teixeira L, Araújo L, Afonso RM, Pachana N. Predictors of anxiety in centenarians: health, economic factors, and loneliness. International Psychogeriatrics. 2014:1-10. 176. Fraga S, Marques-Vidal P, Vollenweider P, Waeber G, Guessous I, Paccaud F, et al. Association of socioeconomic status with inflammatory markers: A two cohort comparison. Preventive medicine. 2014. 177. Lawlor DA, Harro M, Wedderkopp N, Andersen LB, Sardinha LB, Riddoch CJ, et al. Association of socioeconomic position with insulin resistance among children from Denmark, Estonia, and Portugal: cross sectional study. BMJ. 2005;331(7510):183. 178. Silva M. Health service usage in the Ribeira de Pena concelho of Portugal. PhD Dissertation. University of Exeter; 1988. 179. Alves J, Kunst AE, Perelman J. Evolution of socioeconomic inequalities in smoking: results from the Portuguese national health interview surveys. BMC public health. 2015;15(1):1. 180. Lopez AD, Collishaw NE, Piha T. A descriptive model of the cigarette epidemic in developed countries. Tobacco control. 1994;3(3):242. 181. Moreira PA, Padrão PD. Educational and economic determinants of food intake in Portuguese adults: a cross-sectional survey. BMC Public Health. 2004;4(1):58. 182. Santos A-C, Barros H. Prevalence and determinants of obesity in an urban sample of Portuguese adults. Public Health. 2003;117(6):430-7. 183. Santos AC, Ebrahim S, Barros H. Gender, socio-economic status and metabolic syndrome in middle-aged and old adults. BMC Public Health. 2008;8(1):1. 184. Nogueira H, Santana P, Santos R. Saúde: Vulnerabilidade e Oportunidade na Área Metropolitana de Lisboa. In: Santana P, editor. A Cidade e a Saúde. Coimbra: Edições Almedina. 2007. 185. Nunes B, Silva RD, Cruz VT, Roriz JM, Pais J, Silva MC. Prevalence and pattern of cognitive impairment in rural and urban populations from Northern Portugal. BMC neurology. 2010;10(1):1. 186. Alves L, Azevedo A, Silva S, Barros H. Socioeconomic inequalities in the prevalence of nine established cardiovascular risk factors in a southern European population. PLoS One. 2012;7(5):e37158. The relationship between context and health inequalities Chapter 1. Introduction 61 187. Pereira M, Lunet N, Paulo C, Severo M, Azevedo A, Barros H. Incidence of hypertension in a prospective cohort study of adults from Porto, Portugal. BMC cardiovascular disorders. 2012;12(1):1. 188. Alves E, Henriques A, Correia S, Santos AC, Azevedo A, Barros H. Cardiovascular risk profile of mothers of a Portuguese birth cohort: A survey 4years after delivery. Preventive medicine. 2013;57(5):494-9. 189. Sandes AR, Nascimento C, Figueira J, Gouveia R, Valente S, Martins S, et al. Aleitamento Materno: prevalência e factores condicionantes. Acta Médica Portuguesa. 2007;20:193-200. 190. Camões M, Lopes C, Oliveira A, Santos AC, Barros H. Overall and central obesity incidence in an urban Portuguese population. Preventive medicine. 2010;50(1):50-5. 191. Santos MR, Vale MS, Miranda L, Mota J. Socio-demographic and perceived environmental correlates of walking in Portuguese adults—A multilevel analysis. Health & place. 2009;15(4):1094-9. 192. Moreira P, Santos S, Padrão P, Cordeiro T, Bessa M, Valente H, et al. Food patterns according to sociodemographics, physical activity, sleeping and obesity in Portuguese children. International journal of environmental research and public health. 2010;7(3):1121-38. 193. Cavelaars AE, Kunst AE, Geurts JJ, Crialesi R, Grötvedt L, Helmert U, et al. Educational differences in smoking: international comparison. BMJ. 2000;320(7242):1102-7. 194. Veiga PA. Income-related health inequality in Portugal. University of Minho. 2005. 195. Ribeiro S, Furtado C, Pereira J. Association between cardiovascular disease and socioeconomic level in Portugal. Revista Portuguesa de Cardiologia (English Edition). 2013;32(11):847-54. 196. Vilela S, Oliveira A, Pinto E, Moreira P, Barros H, Lopes C. The influence of socioeconomic factors and family context on energy-dense food consumption among 2-year-old children. European journal of clinical nutrition. 2015;69(1):47-54. 197. Kunst AE, Bos V, Andersen O, Cardano M, Costa G, Harding S, et al. Monitoring of trends in socioeconomic inequalities in mortality: experiences from a European project. Demographic research. 2004;2:229-54. 198. Oliveira A, Maia B, Lopes C. Determinants of inadequate fruit and vegetable consumption amongst Portuguese adults. Journal of Human Nutrition and Dietetics. 2014;27(s2):194-203. 199. Scott KM, Al-Hamzawi AO, Andrade LH, Borges G, Caldas-de-Almeida JM, Fiestas F, et al. Associations between subjective social status and DSM-IV mental disorders: results from the world mental health surveys. JAMA psychiatry. 2014;71(12):1400-8. 200. European Commission. Research Findings – Social Situation Monitor - Income inequality in EU countries. Available from: http://ec.europa.eu/social/main.jsp?catId=1050&intPageId=1870&langId=en. Accessed on June 1st 2017. 201. Pereira J, Furtado C. Equidade em saúde: do conhecimento à acção. In: DGS/ENSP, editor. A Nova Saúde Pública: A saúde Pública na Era do Conhecimento. Lisbon: Gradiva; 2011. 202. World Health Organisation. WHO evaluation of the national health plan of Portugal (2004-2010). Copenhagen: WHO Regional Office for Europe. 2010. The relationship between context and health inequalities Chapter 1. Introduction 62 203. Bago d'Uva T. Equidade no sector da saúde em Portugal. In: Simões J, editor. 30 Anos do Serviço Nacional de Saúde - um percurso comentado. Lisbon: Simões, Jorge; 2010. 204. Conselho de Ética do Instituto de Higiene e Medicina Tropical da Universidade Nova de Lisboa. Pedido de apreciação ética para projetos e protocolos de investigação. Lisbon: CEIHMT-UNL; 2017. 205. ESS team. About the European Social Survey European Research Infrastructure – ESS ERIC. Available from: http://www.europeansocialsurvey.org/about/. [Accessed June 1st 2017] 206. ESS team. About ESS - Funding. Available from: http://www.europeansocialsurvey.org/about/funding.html. [Accessed June 1st 2017] 207. European Social Survey. Frequently Asked Questions. Available from: http://www.europeansocialsurvey.org/about/faq.html. [Accessed June 1st 2017] 208. The ESS Sampling Expert Panel. Sampling for the European Social Survey Round VII: Principles and Requirements. London: Centre for Comparative Social Surveys, City University London; 2014. 209. ESS team. Rotating Modules. Available from: http://www.europeansocialsurvey.org/methodology/questionnaire/rotating_questionna ire.html. [Accessed June 1st 2017] 210. ESS team. ESS Bibliography. Available from: http://www.europeansocialsurvey.org/bibliography/. [Accessed June 1st 2017] 211. European Commission. Regulation (EC) No 1177/2003 of the European Parliament and of the Council of 16 of June 2003 concerning Community statistics on income and living conditions (EU-SILC). Brussels; 2003. 212. Eurostat. Health variables in SILC – methodology. Available from: http://ec.europa.eu/eurostat/statisticsexplained/index.php/Health_variables_in_SILC_-_methodologyMain_concepts_and_definitions. [Accessed June 1st 2017] 213. European Commission. Commission Regulation (EC) No 1982/2003 of 21 October 2003 implementing Regulation (EC) No 1177/2003 of the European Parliament and of the Council concerning Community statistics on income and living conditions (EU-SILC) as regards the sampling and tracing rules, 1982/2003. Brussels; 2003. 214. Eurostat. Statistics on Income and Living Conditions – Publications. Available from: http://ec.europa.eu/eurostat/web/income-and-living-conditions/publications. [Accessed June 1st 2017] 215. Khan KS, Ter Riet G, Glanville J, Sowden AJ, Kleijnen J. Systematic reviews: CRD's guidance for undertaking reviews in health care. 3rd edition. London:NHS Centre for Reviews and Dissemination; 2009. 216. Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS med. 2009;6(7):1-6. 217. Petticrew M. Systematic reviews in public health: old chestnuts and new challenges. Bulletin of the World Health Organization. 2009;87(3):163-A. 218. Egger M, Smith GD, Schneider M. Systematic reviews of observational studies. In: Egger M, Smith GD, Altman DG, editors. Systematic Reviews in Health Care: Meta-Analysis in Context, Second Edition. 2001:211-27. 219. Hosmer DW, Lemeshow S. Introduction to the logistic regression model. In: Hosmer DW, Lemeshow S, Sturdivant RX ,editors. Applied Logistic Regression, Third Edition. 2013:1-33. The relationship between context and health inequalities Chapter 1. Introduction 63 220. Hosmer DW, Lemeshow S, Sturdivant RX. Interpretation of the Fitted Logistic Regression Model. In: Hosmer DW, Lemeshow S, Sturdivant RX, editors. Applied Logistic Regression, Third Edition. 2013:49-88. 221. Subramanian S, Kim DJ, Kawachi I. Social trust and self-rated health in US communities: a multilevel analysis. Journal of Urban Health. 2002;79(1):S21-S34. 222. Subramanian S, Delgado I, Jadue L, Vega J, Kawachi I. Income inequality and health: multilevel analysis of Chilean communities. Journal of epidemiology and community health. 2003;57(11):844-8. 223. Fielding A, Goldstein H. Cross-classified and multiple membership structures in multilevel models: An introduction and review. University of Bristol. Research Report RR791, 2006. 224. Wagstaff A. The bounds of the concentration index when the variable of interest is binary, with an application to immunization inequality. Health economics. 2005;14(4):429-32. 225. Wagstaff A, Van Doorslaer E, Watanabe N. On decomposing the causes of health sector inequalities with an application to malnutrition inequalities in Vietnam. Journal of econometrics. 2003;112(1):207-23. The relationship between context and health inequalities Chapter 1. Introduction 64 The relationship between context and health inequalities Chapter 2. Results 65 2. Results 2.1. The ‘dark side’ of social capital: trust and self-rated health in European countries Reference: Campos-Matos I, Subramanian SV, Kawachi I. The ‘dark side’ of social capital: trust and self-rated health in European countries. European Journal of Public Health. 2016;26(1):90-95. DOI: http://dx.doi.org/10.1093/eurpub/ckv089 Online supplementary data for this publication is in Appendix 2. The relationship between context and health inequalities Chapter 2. Results 66 The relationship between context and health inequalities Chapter 2. Results 73 2.2. Social mobility and health in European countries: does welfare regime type matter? Reference: Campos-Matos I, Kawachi I. Social mobility and health in European countries: does welfare regime type matter? Social Science and Medicine. 2015;142:241-248 DOI: https://doi.org/10.1016/j.socscimed.2015.08.035 The relationship between context and health inequalities Chapter 2. Results 74 Short communication Social mobility and health in European countries: Does welfare regime type matter? In^ es Campos-Matos a , b , * , Ichiro Kawachi b a Department of International Public Health and Biostatistics, Nova University of Lisbon, Lisbon, Portugal b Department of Social and Behavioral Sciences, Harvard School of Public Health, Boston, MA, USA article info Article history: Received 7 October 2014 Received in revised form 10 August 2015 Accepted 19 August 2015 Available online 22 August 2015 Keywords: Europe Welfare regimes Social mobility Self-rated health abstract Health inequalities pose an important public health challenge in European countries, for which increased social mobility has been suggested as a cause. We sought to describe how the relationship between health inequalities and social mobility varies among welfare regime types in the European region. Data from six rounds of the European Social Survey was analyzed using multilevel statistical techniques, stratified by welfare regime type, including 237,535 individuals from 136 countries. Social mobility among individuals was defined according to the discrepancy between parental and offspring educational attainment. For each welfare regime type, the association between social mobility and self-rated health was examined using odds ratios and risk differences, controlling for parental education. Upwardly mobile individuals had between 23 and 44% lower odds of reporting bad or very bad self-rated health when compared to those who remained stable. On an absolute scale, former USSR countries showed the biggest and only significant differences for upward movement, while Scandinavian countries showed the smallest. Downward social mobility tended to be associated with worse health, but the results were less consistent. Upward social mobility is associated with worse health in all European welfare regime types. However, in Scandinavian countries the association of upward mobility was smaller, suggesting that the Nordic model is more effective in mitigating the impact of social mobility on health and/or of health on mobility. ©2015 Elsevier Ltd. All rights reserved. 1. Introduction Despite sustained efforts put in effect across European countries, health inequalities persist as an important public health challenge (Mackenbach, 2012). A range of policy solutions has been tried, but so far with relatively little impact. Social mobility has been identified as an important driver of health inequalities. Social mobility can occur either between generations (parents and children) as well as within the life-course of the individual. Truncated intergenerational social mobility is of particular concern because it can result in the crystallization of wealth inequality as well as health inequalities. Ill health is a potent cause of both intraindividual and inter-generational mobility restriction. For example, childhood illness has been shown to adversely affect educational attainment (Case and Paxson, 2008), which will subsequently affect an individual's success in the labor market. Ill health in midlife can affect labor force participation (reduced working hours, job loss), resulting in downward income mobility. Restricted social mobility can be manifest in multiple dimensions e educational achievement, occupational status or earnings and income. Furthermore health selection can be both direct (e.g. depressive illness directly resulting in truncated educational achievement) as well as indirect ee.g. depressive illness resulting in reduced social mobility via intermediary factors such as stigma and discrimination (West, 1991). Accordingly, social protections such as universal access to health care or anti-discrimination legislation represent important policies to promote both intraindividual and inter-generational social mobility. An individual's socioeconomic position is a robust determinant of his/her health, both in terms of their current (or achieved) socioeconomic position, but also their lifetime trajectory (Marmot and Macmillan, 2004; Marmot and Wilkinson, 2005). This can reflect processes of accumulation or a direct impact of social mobility (Hallqvist et al., 2004). Studies on the effect of social mobility on health have not always produced clear-cut results, with *Corresponding author. Department of International Public Health and Biostatistics, Nova University of Lisbon, Lisbon, Portugal. E-mail address: [email protected] (I. Campos-Matos). Contents lists available at ScienceDirect Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed http://dx.doi.org/10.1016/j.socscimed.2015.08.035 0277-9536/©2015 Elsevier Ltd. All rights reserved. Social Science & Medicine 142 (2015) 241e248 some seeming to indicate that upward social mobility can be just as deleterious to health as downward social mobility (Hemmingsson et al., 1999; Liberatos et al., 1988). These mixed results might, however, be a consequence of the inconsistent ways in which social mobility has been operationalized in the empirical literature (Singhammer and Mittelmark, 2010). The use of different indicators to characterize social groups can also be of consequence, since different indicators, such as education, occupation or income, as well as the intergenerational movement between them, can have different meanings (Galobardes et al., 2006). Overall, there are both theoretical and empirical grounds to suggest that the causal relationship between health and social mobility is bidirectional: individuals have more or less opportunities for social mobility depending on their health endowment and their health achievement is affected by transitions between social strata. The extent of social mobility varies substantially between countries (Beller and Hout, 2006). Government actions, such as expanding access to schooling or investing in the health of children (e.g. via improved nutrition or vaccination programs) have the potential capacity to break the inter-generational transmission of social disadvantage. Considering the strong relationship between social mobility and health, these governmental actions, systematized in Fig. 1, can have an important impact on health inequalities. Welfare regime types, often used to categorize European countries, share common policies such as the ones outlined in Fig. 1. In this cross national comparative study, we sought to examine the relation between social mobility and population health among different types of welfare regimes in the European region, in order to understand how the welfare state might moderate the link between mobility and health. 2. Methods 2.1. Data sources and variables Individual data was collected from six rounds of the European Social Survey (ESS), between 2002 and 2012, from thirty selected countries. The ESS is a repeated cross-sectional survey that collects comparable data on individual socioeconomic characteristics and health status of several European countries (ESS ERIC, 2014). Data is available online at www.europeansocialsurvey.org. The outcome variable, self-rated health, was based on the survey participant's response to the question ‘How is your health in general?’, dichotomized so that 1 included ‘bad’or ‘very bad’(other possible answers were ‘fair’,‘good’or ‘very good’). Social mobility was measured in relation to mother and father's achieved level of education according to the International Standard Classification of Education (ISCED) levels. Although social mobility is usually measured on the basis of the fathers' social standing, the increasing participation of women in the workforce and the importance of the mothers' characteristics on children's health behaviors (Favaro and Santonastaso, 1995) support the importance of considering mothers' status in social mobility studies; therefore, this analysis was done separately. Social mobility was classified in three possible categories: ‘down’,‘stable’and ‘up’, according to whether the respondent had reached, respectively, a lower, the same, or higher educational level than his or her parent. Our measure of mobility controlled for the parent's educational achievement when the respondent was 14 (the same variable used to assess mobility). Failing to take into account the ‘social group of origin’has been a common pitfall in previous studies of intergenerational social mobility and health (Singhammer and Mittelmark, 2010). Controlling for parent's educational achievement yields mobility coefficients that can be interpreted as independent from social group of origin. Other individual-level variables included age (restricted to 25 years and up), gender, marital status, belonging to an ethnic minority group, self-perceived income, domicile and main occupational activity. Respondents who were in full-time education were excluded, since not having completed education did not permit comparison to parents' achievement. For all these variables a base category with contrasting indicator variables was specified, except age, which was centered around it's grand mean. Fig. 1. Entry points for reducing and eliminating health disparities. I. Campos-Matos, I. Kawachi / Social Science & Medicine 142 (2015) 241e248242 To reduce the possibility of confounding by economic development, Gross Domestic Product (GDP) per capita, converted to international dollars using purchasing power parity, was retrieved from the World Bank database (World Bank, 2014) and included as a country-level variable (specified per country, per year). Additionally, the Gini coefficient, which might also confound the association between social mobility and health, was retrieved from the Eurostat database (Eurostat, 2014) and used as a level 2 variable (country-year specific). However, this was only used as a sensitivity analysis, since the Gini coefficient was missing for many countries for several years. 2.2. Welfare regimes Countries were grouped by welfare regime type and analyzed separately. Welfare regime classification is a much-debated topic, not only with disputed typologies, but also regarding which characteristics should be used to for their classification (Bambra, 2007). Nonetheless, we started with a widely used typology that divides European countries into four regime types: (i) Scandinavian, characterized by universal and generous benefits and a strong redistributive social security system (Eikemo et al., 2008a; Fenger, 2007); (ii) Anglo-Saxon, with a low level of government spending on social protection, modest benefits, usually means-tested (Eikemo et al., 2008a; Fenger, 2007); (iii) Bismarckian, with benefits tied to employment, financed mainly by employer and employee, and minimal redistribution (Eikemo et al., 2008a); (iv) Southern, with a dualist system of welfare provision, which strongly protects part of the population while underprotecting another (19). This classification is primarily based on Esping-Andersen's et al. (1990) groundbreaking work, which operationalized three principles: decommodification, social stratification and the publicprivate mix, to classify the first three typologies (Eikemo et al., 2008a; Espig Andersen et al., 1990). Ferrera (1996) later added the Southern type, basing his classification on the coverage of social protection schemes (Ferrera, 1996). This typology has been replicated in other attempts to define welfare regime types (Bambra, 2007; Bonoli, 1997) and has been used previously in the health literature (Eikemo et al., 2008a, 2008b). The consideration of Central and Eastern European countries to the European Union adds further complexity to this classification. Historically, the trajectories of these countries' welfare transformation can be separated in two, depending on the extent to which the welfare effort collapsed in the 1990's (Cook and Press, 2010). This typology separates Central and Eastern European countries (including Poland, Hungary, Czech Republic and the Baltic states, among others) from the remaining former USSR states (such as Russia and Central Asian countries). However, using a hierarchical cluster analysis, Fenger (2007) showed that, based on similarities on government spending, social situation and political participation, these countries could be divided in the following way: (i) Former USSR, with generally low governmental spending on social programs, mostly financed through social contributions; (ii) Post-Communist European, very similar to the first type, but with higher levels of economic growth, inflation, social wellbeing and egalitarianism (Fenger, 2007). It is important to note that these characteristics do not necessarily describe the countries in absolute; in fact, most countries have a mix of different welfare regimes, but nonetheless have predominant characteristics of one type. 2.3. Statistical analysis Data was analyzed using multilevel logistic models based on a logit-link function with a first order quasi-likelihood estimation procedure. The models were run using MLwiN program version 2.28 (Rasbash et al., 2013). The data was analyzed considering its hierarchical structure in three levels: individuals (level 1), nested within years (level 2), nested within countries (level 3). Overall odds, odds ratios, overall probabilities and rate differences were calculated for each welfare regime. This was based in part in the methods used by Hemmingsson et al. (1999), although we applied a multilevel modeling technique. The use of multilevel statistical techniques allows for the analysis of the effect of both individual and contextual variables on the outcome of interest. In this analysis, the contextual variable was countryand year-specific, making it mandatory to include both as levels. Additionally, these models take into account the hierarchical nesting of individual observations within a year and within a country, correcting otherwise underestimated standard errors and allow the modeling of variability at each level of analysis (Subramanian et al., 2003). As a sensitivity analysis, the same models were run using different estimation procedures. 3. Results Table 1 outlines the distribution of the variables among welfare regime types. Each type includes between two and seven countries, ranging in sample size from 23,310 to 62,509 individuals. Countries with a Scandinavian welfare regime had the lowest proportion of people reporting bad or very bad health (5.4%), followed by AngloSaxon (5.8%), Bismarckian (6.9%), Southern (11.9%), PostCommunist European (14.5%) and Former USSR (19.6%). Most respondents had achieved a higher educational level than their parents: between 49.3 and 63.2% had improved in relation to their mother, and between 45.6 and 61.7% in relation to their father. Between one third and half of respondents remained in the same educational level as their parents and a smaller proportion ‘moved down’eonly 1.6% in southern countries in relation to mother's achievement, up to 12.1% in Bismarckian countries, in relation to father's. Table 2 shows the coefficients for the mobility variable (stable, upward, downward) from the multilevel models. The full models are available in Tables 3 and 4 as an online supplement. The general pattern of association between social mobility and health was similar across all regime types, i.e. upward mobility was protective, while downward mobility was detrimental for self-rated health. Figs. 2 and 3 show the probabilities and risk differences in the different mobility groups in each welfare regime type. As was noted in Table 1, the overall probability of bad self-rated health differs significantly between welfare regime types. The benefits of upward mobility are present in all regime types, but risk differences are particularly high for Southern, Post Communist European and Former USSR regimes and significantly different only in the latter. Scandinavian countries showed the lowest risk differences between upwardly mobile and stable individuals when considering either mothers' or fathers' education. Downward mobility is generally detrimental. Post Communist European countries show the biggest absolute differences, with downward mobility leading to an absolute difference of 2.9 and I. Campos-Matos, I. Kawachi / Social Science & Medicine 142 (2015) 241e248 243 3.6% in the risk of bad health measured on the basis of maternal and paternal achievement, respectively. In Scandinavian, Anglo-Saxon, Bismarckian and former USSR countries, this risk difference ranges from 0.1 to 1.9%. Overall, Scandinavian countries showed the smallest absolute difference in health between stable and downward mobility when measured by the father's achievement, and former USSR countries when measured by the mother's. As a sensitivity analysis, the same models were run using different estimation procedures (second order marginal quasilikelihood, first and second order predictive quasi-likelihood and Markov Chain Monte Carlo). All resulted in models with the same effect of social mobility in the different welfare regimes. A model was also run with the Gini coefficient as a level 2 variable; this had very little effect in the odds ratios or risk differences and no effect in the statistical significance of the results. 4. Discussion This study sought to describe differences in the relationship between social mobility and health within different welfare regimes to better understand the effect that different regime types might have. Different welfare regimes had a substantially different proportion of individuals with bad or very bad self-rated health, lowest in countries in the Scandinavian regime (5.4%), followed by Anglo-Saxon (5.8%), Bismarckian (6.9%), Southern (11.9%), PostCommunist European (14.5%) and finally Former USSR (19.6%). Table 1 Composition of each welfare type. Scandinavian Anglo-Saxon Bismarckian Southern Post-Communist European Former USSR Countries Denmark Finland Norway Sweden Iceland United Kingdom Ireland Austria Belgium Switzerland Germany France Luxembourg Netherlands Spain Greece Italy Portugal Czech Republic Hungary Poland Slovenia Slovakia Croatia Bulgaria Estonia Latvia Lithuania Russia Ukraine n (level 2) 23 12 33 18 34 16 n (level 1) 37,975 23,310 62,509 31,789 51,698 30,072 Level 1 variables % bad or very bad health 5.4 5.8 6.9 11.9 14.5 19.6 Mobility, mother (%) Down 6.3 9.3 4.7 1.6 4.0 9.9 Stable 30.4 35.2 32.9 49.0 38.0 29.2 Up 63.2 55.5 62.4 49.3 58.0 60.9 Mobility, father (%) Down 10.1 10.2 12.1 3.4 6.8 9.4 Stable 34.6 36.7 42.3 50.9 46.1 28.9 Up 55.4 53.1 45.6 45.6 47.0 61.7 Women (%) 50.0 55.0 52.9 56.6 55.1 61.7 Mean age (years) 51.7 51.7 51.6 52.4 51.8 52.7 Urban (%) 64.3 66.0 56.9 65.0 61.3 70.2 Main activity (%) Paid work 63.5 49.7 53.9 48.0 48.9 50.6 No activity 31.3 35.3 32.9 35.5 42.0 42.0 Other 5.2 15.0 13.3 16.5 9.1 7.3 Feeling about income (%) Living comfortably 49.5 35.4 38.1 15.3 11.3 4.8 Coping 41.4 45.3 46.3 45.0 45.3 37.7 Difficult 7.0 14.3 12.3 27.7 29.0 37.0 Very difficult 2.0 5.0 3.3 12.1 14.5 20.5 Minority (%) 2.4 5.0 4.6 3.0 5.7 12.7 Marital status (%) Married 57.5 55.0 59.8 63.7 61.9 52.9 Separated/divorced 12.2 11.2 11.9 6.5 9.5 14.7 Widow 7.0 10.9 9.3 12.2 14.2 19.8 Single 23.3 22.9 19.1 17.6 14.3 12.7 Level 2 variables GDP per capita (international dollars) 39,459 36,705 37,075 26,396 19,924 16,599 Note: ISCED International Standard Classification of Education. Table 2 Mobility odds ratios and 95% confidence intervals from the multilevel models in each welfare regime type. Scandinavian Anglo-Saxon Bismarckian Southern Post-Communist European Former USSR Mother n (level 1) 30,458 19,752 53,644 29,030 46,556 24,959 Stable 11111 1 Down 1.39 (1.06,1.83) 1.40 (1.11,1.78) 1.09 (0.89,1.31) 0.72 (0.43,1.21) 1.35 (1.13,1.60) 1.01 (0.84,1.22) Up 0.77 (0.68,0.86) 0.70 (0.60,0.81) 0.77 (0.71,0.84) 0.56 (0.49,0.62) 0.69 (0.65,0.75) 0.67 (0.61,0.74) Father n (level 1) 29,837 19,184 52,326 28,417 45,500 23,036 Stable 11111 1 Down 1.18 (0.95,1.45) 1.51 (1.19,1.91) 1.20 (1.06,1.37) 1.24 (0.92,1.68) 1.45 (1.28,1.65) 1.17 (0.98,1.39) Up 0.76 (0.67,0.86) 0.68 (0.59,0.79) 0.76 (0.69,0.82) 0.57 (0.51,0.64) 0.73 (0.68,0.78) 0.68 (0.62,0.75) Note: bold indicates OR significant at p <0.05. I. Campos-Matos, I. Kawachi / Social Science & Medicine 142 (2015) 241e248244 This difference was reproduced in the multilevel models, which controlled for several socioeconomic individual and country characteristics. Both on a relative and on an absolute scale, upward mobility was associated with better health, regardless of welfare regime type. However, on the relative scale, these were significant for all regime types, whereas on the absolute scale there was only a true difference in countries from the Former USSR. Downward mobility was generally associated with worse health, but to differing extents and following a less clear pattern. It has been argued before that the use of only absolute or relative measures can be misleading, and our findings reiterate this argument (King et al., 2012; Kelly et al., 2007). In fact, relative measures of health inequalities are insensitive to equiproportionate changes, while absolute measures are insensitive to uniform changes, which reflects different equity value judgments implied in the empirical analysis (Allanson and Petrie, 2013). Thus, different results between relative and absolute scales might be a consequence of different levels of overall ill-health: former USSR countries had the highest prevalence of bad or very bad self-rated health, making absolute differences more likely to emerge. This study is sensible to a number of limitations. The outcome measure, self-rated health, is very culturally-sensitive, complicating cross-national comparisons (Jylh€ a et al., 1998). Nevertheless, it is an important predictor of mortality in every society where it has been examined (Idler and Benyamini, 1997), making it a much used and valued health measure. Additionally, social mobility doesn't have a unanimously accepted operationalization. Differences in educational achievement are not necessarily a reflection of different societal prestige or access to different social resources. Indeed, occupational mobility is often preferred (Beller and Hout, 2006), but the occupational measures available in the ESS were crude and difficult to compare between respondents and their parents. Also importantly, although the analyses controlled for parental education, this operationalization of social mobility might be measuring processes of accumulation. The welfare regime classification is also debatable. Although most of the regime types that were used in this study have been extensively used before, and despite both level 2 and level 3 variability being very low and nonsignificant (hinting to a high homogeneity between countries and country-years), they might not reflect the characteristics of welfare regimes that have an impact on the relationship between social mobility and health. Finally, it is not possible to assess causality between social mobility and health, considering that our analyses are based on cross-sectional data. Indeed, it is possible that the health of participants in our sample was already affected by mobility in the previous time period. The association between health and socioeconomic status is likely dynamic and bidirectional across the life course. Overall, it is interesting to note that although all welfare regime types show relative differences in bad self-rated health for upwardly mobile individuals, on an absolute scale the Scandinavian regime shows the smallest differences and the former USSR group the largest. Although welfare regime type seems to account for an important part of the variation in self-perceived health among European countries (Eikemo et al., 2008b), the extent to which it impacts health inequalities has been questioned. Mackenbach et al. (2008) reported a surprisingly high degree of health inequalities in northern European countries, showing that, despite egalitarian policies, lifestyle-related risk factors remain an important cause of mortality inequalities (Mackenbach et al., 2008). Eikemo et al. (2008a) also showed a clear gradient of health inequalities Table 3 Multilevel models: mobility from mother's education. Scandinavian Anglo-Saxon Bismarckian Southern Post communist Former USSR n (level 1) 30,458 19,752 53,644 29,030 46,556 24,959 OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) Fixed Parameters Intercept 0.008 (0.005,0.012) 0.009 (0.006,0.027) 0.011 (0.007,0.018) 0.019 (0.009,0.037) 0.016 (0.011,0.022) 0.036 (0.026,0.050) Individual Level Mobility (reference: stable) Down 1.39 (1.06,1.83) 1.40 (1.11,1.78) 1.09 (0.89,1.31)* 0.72 (0.43,1.21)* 1.35 (1.13,1.60) 1.01 (0.84,1.22)* Up 0.77 (0.68,0.86) 0.70 (0.60,0.81) 0.77 (0.71,0.84) 0.56 (0.49,0.62) 0.69 (0.65,0.75) 0.67 (0.61,0.74) Gender (ref: male) 1.03 (0.92,1.15)* 0.89 (0.77,1.02)* 1.07 (0.98,1.15)* 1.59 (1.45,1.75) 1.06 (0.99,1.13)* 1.21 (1.12,1.32) Age 0.99 (0.99,1.00)* 1.00 (0.99,1.01)* 1.01 (1.01,1.01) 1.04 (1.04,1.05) 1.04 (1.04,1.04) 1.05 (1.04,1.05) Domicile (ref: urban) 0.94 (0.84,1.05)* 0.85 (0.74,0.98) 0.88 (0.82,0.95) 1.14 (1.05,1.24) 1.11 (1.04,1.18) 0.97 (0.89,1.05)* Main activity (ref: paid work) No activity 6.29 (5.39,7.35) 6.17 (5.00,7.61) 4.35 (3.90,4.84) 2.54 (2.22,2.89) 3.35 (3.06,3.67) 2.63 (2.35,2.93) Other 2.83 (2.21,3.62) 2.55 (1.96,3.31) 1.82 (1.59,2.09) 1.91 (1.64,2.21) 1.97 (1.74,2.24) 1.56 (1.31,1.86) Income (ref: living comfortably) Coping 1.40 (1.24,1.59) 1.49 (1.26,1.78) 1.52 (1.38,1.67) 1.38 (1.15,1.64) 1.55 (1.33,1.81) 1.10 (0.84,1.45)* Difficult 2.97 (2.49,3.54) 2.61 (2.12,3.21) 3.29 (2.94,3.68) 2.28 (1.90,2.73) 3.00 (2.57,3.51) 1.76 (1.35,2.29) Very difficult 4.57 (3.56,5.86) 4.09 (3.18,5.27) 5.21 (4.47,6.06) 3.97 (3.28,4.82) 5.37 (4.55,6.32) 2.86 (2.19,3.75) Minority (ref: no) 1.44 (1.05,1.97) 0.86 (0.61,1.21)* 1.27 (1.08,1.49) 1.03 (0.79,1.34)* 0.97 (0.86,1.09)* 1.02 (0.91,1.15)* Marital status (ref: married) Separated/divorced 1.37 (1.18,1.59) 2.02 (1.67,2.43) 1.28 (1.14,1.42) 1.23 (1.03,1.47) 1.03 (0.93,1.14)* 1.28 (1.14,1.43) Widow 1.07 (0.89,1.27)* 1.13 (0.92,1.37)* 1.08 (0.97,1.21)* 1.06 (0.95,1.17)* 1.06 (0.98,1.14)* 1.18 (1.08,1.30) Single 1.02 (0.87,1.19)* 1.19 (0.99,1.43)* 1.17 (1.05,1.31) 1.29 (1.12,1.51) 1.09 (0.97,1.21)* 1.31 (1.13,1.52) Mother's education (ref: ISCED V/VI) ISCED I 2.39 (1.76,3.26) 1.63 (1.18,2.25) 2.14 (1.66,2.77) 1.23 (0.79,1.92)* 2.48 (1.93,3.21) 2.02 (1.68,2.43) ISCED II 2.00 (1.49,2.71) 1.25 (0.91,1.72)* 1.72 (1.34,2.21) 0.67 (0.39,1.13)* 1.89 (1.48,2.41) 1.76 (1.45,2.12) ISCED III 1.67 (1.25,2.23) 1.20 (0.82,1.77)* 1.34 (1.04,1.71) 1.03 (0.62,1.71)* 1.26 (0.99,1.59)* 1.25 (1.04,1.49) ISCED IV 1.64 (1.04,2.59) 0.37 (0.17,0.79) 1.58 (1.09,2.27) 0.53 (0.15,1.88)* 1.54 (1.05,2.26) 1.12 (0.91,1.38)* Level 2 GDP 0.99 (0.99,1.00)* 0.99 (0.99,1.00)* 1.00 (0.99,1.00)* 0.99 (0.99,0.99) 0.99 (0.99,0.99) 0.99 (0.99,0.99) Random Parameters Level 3 variance ( s vo ) 0.08 (0.06)* 0.19 (0.19)* 0.22 (0.12)* 0.23 (0.17)* 0.07 (0.04)* 0.03 (0.02)* Level 2 variance ( s uo ) 0.01 (0.01)* 0 (0) 0 (0) 0.01 (0.01)* 0.03 (0.01) 0.01 (0.01)* Notes: * not significant at p <0.05. ISCED International Standard Classification of Education. OR Odds ratio. CI Confidence interval. I. Campos-Matos, I. Kawachi / Social Science & Medicine 142 (2015) 241e248 245 between European welfare states, from Southern (with the highest inequalities) to Bismarckian (with the lowest) (Eikemo et al., 2008a). This ‘paradox’was examined by Mackenbach (2012), who postulated that social mobility might be one of the drivers of health inequalities in Western European welfare states. Some studies have indeed shown that increased social mobility is associated with stronger health inequalities (Simons et al., 2013;  Asgeirsd ottir and Ragnarsd ottir, 2013; Elstad, 2001). However, Brekke, Grunfeld and Kverndokk (2014), showed that higher health inequalities in more egalitarian countries might be solely a consequence of a more equal health distribution, since the concentration index is more sensitive to health-contingent income transfers than to income-contingent health transfers (Brekke et al., 2012). Our findings suggest that some welfare states are in fact more Table 4 Multilevel models: mobility from father's education. Scandinavian Anglo-Saxon Bismarckian Southern Post communist Former USSR Level 1 n 29,837 19,184 52,326 28,417 45,500 23,036 OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) OR (95% CI) Fixed Parameters Intercept 0.009 (0.007,0.014) 0.009 (0.005,0.019) 0.013 (0.009,0.018) 0.014 (0.008,0.024) 0.017 (0.013,0.023) 0.033 (0.023,0.047) Individual Level Mobility (reference: stable) Down 1.18 (0.95,1.45)* 1.51 (1.19,1.91) 1.20 (1.06,1.37) 1.24 (0.92,1.68)* 1.45 (1.28,1.65) 1.17 (0.98,1.39)* Up 0.76 (0.67,0.86) 0.68 (0.59,0.79) 0.76 (0.69,0.82) 0.57 (0.51,0.64) 0.73 (0.68,0.78) 0.68 (0.62,0.75) Gender (ref: male) 1.03 (0.92,1.15)* 0.89 (0.77,1.03)* 1.05 (0.97,1.14)* 1.56 (1.42,1.72) 1.07 (1.00,1.14) 1.26 (1.15,1.37) Age 0.99 (0.99,1.00)* 1.00 (0.99,1.01)* 1.01 (1.01,1.01) 1.04 (1.04,1.05) 1.04 (1.04,1.04) 1.05 (1.04,1.05) Domicile (ref: urban) 0.95 (0.85,1.06)* 0.88 (0.76,1.02)* 0.88 (0.82,0.95) 1.14 (1.04,1.24) 1.11 (1.04,1.18) 0.95 (0.87,1.03)* Main activity (ref: paid work) No activity 6.11 (5.23,7.14) 5.87 (4.76,7.24) 4.44 (3.97,4.95) 2.54 (2.22,2.91) 3.29 (2.99,3.60) 2.70 (2.41,3.04) Other 2.67 (2.07,3.45) 2.47 (1.89,3.22) 1.19 (1.64,2.17) 1.94 (1.67,2.26) 1.91 (1.68,2.17) 1.58 (1.32,1.90) Income (ref: living comfortably) Coping 1.41 (1.24,1.61) 1.45 (1.21,1.72) 1.50 (1.36,1.65) 1.35 (1.13,1.61) 1.55 (1.32,1.81) 1.08 (0.82,1.43)* Difficult 2.86 (2.39,3.42) 2.54 (2.06,3.13) 3.25 (2.90,3.65) 2.24 (1.87,2.69) 2.94 (2.51,3.45) 1.71 (1.29,2.26) Very difficult 4.66 (3.62,6.01) 3.74 (2.89,4.85) 5.05 (4.33,5.90) 3.94 (3.24,4.78) 5.31 (4.49,6.27) 2.83 (2.13,3.76) Minority (ref: no) 1.46 (1.06,2.01) 0.97 (0.69,1.35)* 1.33 (1.13,1.57) 1.03 (0.79,1.35)* 0.97 (0.86,1.09)* 1.01 (0.89,1.15)* Marital status (ref: married) Separated/divorced 1.42 (1.34,1.96) 2.06 (1.69,2.49) 1.26 (1.13,1.41) 1.23 (1.03,1.48) 1.04 (0.93,1.16)* 1.25 (1.11,1.42) Widow 1.08 (0.84,1.34)* 1.17 (0.96,1.43)* 1.06 (0.95,1.19)* 1.06 (0.95,1.18)* 1.07 (0.99,1.16)* 1.18 (1.07,1.30) Single 1.03 (0.88,1.22)* 1.19 (0.99,1.44)* 1.15 (1.03,1.29) 1.31 (1.13,1.53) 1.08 (0.96,1.21)* 1.15 (0.98,1.35)* Education (ref: ISCED V/VI) ISCED I 2.18 (1.70,2.79) 1.59 (1.19,2.12) 1.92 (1.61,2.28) 1.70 (1.24,2.33) 2.27 (1.87,2.74) 2.12 (1.77,2.54) ISCED II 1.67 (1.29,2.16) 1.09 (0.82,1.45)* 1.59 (1.34,1.88) 1.33 (0.94,1.89)* 1.72 (1.44,2.06) 1.79 (1.49,2.15) ISCED III 1.42 (1.13,1.78) 1.16 (0.82,1.64)* 1.36 (1.17,1.57) 0.99 (0.68,1.47)* 1.21 (1.02,1.43) 1.39 (1.17,1.67) ISCED IV 1.67 (1.21,2.29) 0.93 (0.54,1.62)* 1.26 (0.98,1.61)* 1.11 (0.52,2.38)* 0.83 (0.58,1.18)* 1.17 (0.94,1.45)* Level 2 GDP 0.99 (0.99,1.00)* 1.00 (0.99,1.00)* 1.00 (0.99,1.00)* 0.99 (0.99,0.99) 0.99 (0.99,0.99) 0.99 (0.99,0.99) Random Parameters Level 3 variance ( s vo ) 0.09 (0.07)* 0.19 (0.19)* 0.21 (0.12)* 0.24 (0.17)* 0.08 (0.05)* 0.03 (0.02)* Level 2 variance ( s uo ) 0.01 (0.01)* 0 (0) 0 (0) 0.003 (0.004)* 0.03 (0.01) 0.01 (0.01)* Notes:* not significant at p <0.05. ISCED International Standard Classification of Education. OR Odds ratio. CI Confidence interval. Fig. 2. Probability of ‘Bad’or ‘Very Bad’self-rated health per to mobility group, defined from mother's educational achievement, per welfare type (error bars are 95% confidence intervals) and risk difference. I. Campos-Matos, I. Kawachi / Social Science & Medicine 142 (2015) 241e248246 effective in separating social mobility from health, namely Scandinavian countries exhibit smaller differences while former USSR societies the largest. This is not surprising considering that, for example, comprehensive social policies seem to be associated with fewer inequalities in ‘sickness’in European countries, as well as lower rates of non-employment (van der Wel et al., 2011). It is understandable then, that Scandinavian countries will manage to disassociate social mobility from health more effectively, leading to the small risk differences we found for upward mobility in these countries. Importantly too, Central and Eastern European countries underwent considerable transitions in the last decades, with important consequences to their social structures (Saar et al., 2012). Our results for the post-Communist European and former USSR welfare regimes, which tended to show the largest absolute differences in health, might reflect, at least partially, these important structural changes and not just relative social mobility. Our findings also reinforce the need to assess health inequalities using both relative and absolute measures, since the use of only one might be very misleading. Interestingly too, when comparing the association of downward mobility with health as assessed in reference to paternal versus maternal achievement, the former was larger in every welfare regime except the Scandinavian region. Considering that Scandinavian countries have the best indicators of gender equality (European Institute for Gender Equality, 2013), a possible explanation for this is that in other, less gender-egalitarian countries, the father's status is more decisive in determining the family's socioeconomic status, and therefore a downward mobility from his social position has a greater impact. To the best of our knowledge, this is the first research into the moderating effect of welfare regimes on the relationship between social mobility and health, and to measure social mobility separately based on maternal and paternal achievement. Further exploration of our findings would benefit from measuring occupational social mobility in addition to educational mobility. It would also benefit from a separate analysis for each gender, since the effect might be different for women and men and might help explain the differences in the association of downward mobility when measured on the basis of maternal and paternal achievement. Previous studies have questioned the contribution of the welfare regime in mitigating the extent of health inequalities and identified increased social mobility as a possible cause for this (Mackenbach, 2012). However, the present results show that important systematic differences exist between regime types with regard to upward mobility and health, with a notably attenuated association on the absolute scale in Scandinavian countries and stronger association in the former USSR regimes. This suggests that social mobility is not a cause of high health inequalities found in Scandinavian countries in previous analyses. Acknowledgments I. Campos Matos is supported by a Fulbright grant for Public Health research, awarded by the U.S. Department of State and the Portuguese Government, and by a studentship from the Portuguese Foundation of Science and Technology. Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.socscimed.2015.08.035. References Allanson, P., Petrie, D., 2013. On the choice of health inequality measure for the longitudinal analysis of income-related health inequalities. Health Econ. 22 (3), 353e365.  Asgeirsd ottir, T.L., Ragnarsd ottir, D.  O., 2013. Determinants of relative and absolute concentration indices: evidence from 26 European countries. Int. J. Equity Health 12, 53. Bambra, C., 2007. ‘Sifting the wheat from the Chaff’: a two-dimensional discriminant analysis of welfare state regime theory. Soc. Policy &Adm. 41 (1), 1e28. Beller, E., Hout, M., 2006. Welfare states and social mobility: how educational and social policy may affect cross-national differences in the association between occupational origins and destinations. Res. Soc. Stratif. Mobil. 24 (4), 353e365. Bonoli, G., 1997. Classifying welfare states: a two-dimension approach. J. Soc. policy 26 (03), 351e372. Brekke, K.A., Grunfeld, L., Kverndokk, S., 2012. Asymmetric Impacts of Transfers for the Concentration Index: Explaining the Health Equality Paradox. HERO. University of Oslo. Case, A., Paxson, C., 2008. Height, health, and cognitive function at older ages. Am. Econ. Rev. 98 (2), 463. Cook, L., 2010. In: Press, O.U. (Ed.), Eastern Europe and Russia. The Oxford Handbook of The Welfare State, New York. Eikemo, T.A., Huisman, M., Bambra, C., Kunst, A.E., 2008. Health inequalities according to educational level in different welfare regimes: a comparison of 23 Fig. 3. Probability of ‘Bad’or ‘Very Bad’self-rated health per to mobility group, defined from father's educational achievement, per welfare type (error bars are 95% confidence intervals) and risk difference. I. Campos-Matos, I. Kawachi / Social Science & Medicine 142 (2015) 241e248 247 European countries. Sociol. Health &Illn. 30 (4), 565e582. Eikemo, T.A., Bambra, C., Judge, K., Ringdal, K., 2008. Welfare state regimes and differences in self-perceived health in Europe: a multilevel analysis. Soc. Sci. Med. 66 (11), 2281e2295. Elstad, J.I., 2001. Health-related mobility, health inequalities and gradient constraint discussion and results from a Norwegian study. Eur. J. Public Health 11 (2), 135e140. Espig Andersen, G., 1990. In: Donati, P., Prandini, R. (Eds.), The Three Worlds of Welfare Capitalism, Cit. ESS ERIC, 2014. About the European Social Survey European Research Infrastructure eESS ERIC [cited 2014 April 22nd]. Available from: http://www. europeansocialsurvey.org/about/index.html. European Institute for Gender Equality, 2013. Gender Equality Index: Country Profiles. Eurostat, 2014. Gini Coefficient of Equivalised Disposable Income [cited 2014 March 30]. Available from: http://ec.europa.eu/eurostat/tgm/table.do? tab¼table&init¼1&language¼en&pcode¼tessi190&plugin¼1. Favaro, A., Santonastaso, P., 1995. Effects of parents' psychological characteristics and eating behaviour on childhood obesity and dietary compliance. J. Psychosom. Res. 39 (2), 145e151. Fenger, H., 2007. Welfare regimes in central and Eastern Europe: incorporating post-communist countries in a welfare regime typology. Contemp. Issues Ideas Soc. Sci. 3 (2). Ferrera, M., 1996. The'Southern model'of welfare in social Europe. J. Eur. Soc. policy 6 (1), 17e37. Galobardes, B., Shaw, M., Lawlor, D.A., Lynch, J.W., Smith, G.D., 2006. Indicators of socioeconomic position (part 1). J. Epidemiol. Commun. Health 60 (1), 7e12. Hallqvist, J., Lynch, J., Bartley, M., Lang, T., Blane, D., 2004. Can we disentangle life course processes of accumulation, critical period and social mobility? an analysis of disadvantaged socio-economic positions and myocardial infarction in the Stockholm Heart Epidemiology Program. Soc. Sci. Med. 58 (8), 1555e1562. Hemmingsson, T., Lundberg, I., Diderichsen, F., 1999. The roles of social class of origin, achieved social class and intergenerational social mobility in explaining social-class inequalities in alcoholism among young men. Soc. Sci. Med. 49 (8), 1051e1059. Idler, E.L., Benyamini, Y., 1997. Self-rated health and mortality: a review of twentyseven community studies. J. Health Soc. Behav. 21e37. Jylh€ a, M., Guralnik, J.M., Ferrucci, L., Jokela, J., Heikkinen, E., 1998. Is self-rated health comparable across cultures and genders? Journals Gerontology Ser. B Psychol. Sci. Soc. Sci. 53 (3), S144eS152. Kelly M.P., World Health Organization Commission on the Social Determinants of Health and Bonnefoy J., The Social Determinants of Health: Developing an Evidence Base for Political Action, 2007, Universidad del Desarrollo, Chile, and National Institute for Health and Clinical Excellence, United Kingdom. King, N.B., Harper, S., Young, M.E., 2012. Use of relative and absolute effect measures in reporting health inequalities: structured review. BMJ 345 e5774. Liberatos, P., Link, B.G., Kelsey, J.L., 1988. The measurement of social class in epidemiology. Epidemiol. Rev. 10 (1), 87e121. Mackenbach, J.P., Stirbu, I., Roskam, A.J.R., Schaap, M.M., Menvielle, G., Leinsalu, M., Kunst, A.E., 2008. Socioeconomic inequalities in health in 22 European countries. N. Engl. J. Med. 358 (23), 2468e2481. Mackenbach, J.P., 2012. The persistence of health inequalities in modern welfare states: the explanation of a paradox. Soc. Sci. Med. 75 (4), 761e769. Marmot, M., 2004. Men and women behaving badly? In: Macmillan (Ed.), The Status Syndrome: How Social Standing Affects Our Health and Longevity. New York. Marmot, M., Wilkinson, R., 2005. Social Determinants of Health. Oxford University Press. Rasbash, J., Charlton, C., Browne, W.J., Healy, M., Cameron, B., 2013. MLwiN Version 2.28. Centre for Multilevel Modelling, University of Bristol. Saar, E., Unt, M., 2012. Labour market returns to tertiary education in post-socialist countries. In: Lambert, P. (Ed.), Social Stratification: Trends and Processes. Ashgate Publishing, Ltd, England. Simons, A.M., Groffen, D.A., Bosma, H., 2013. Socio-economic inequalities in allcause mortality in Europe: an exploration of the role of heightened social mobility. Eur. J. Public Health ckt145. Singhammer, J., Mittelmark, M.B., 2010. Standard measures of inter-generational social mobility distort actual patterns of mobility and health behaviour: evidence for a better methodology. Crit. Public Health 20 (2), 223e232. Subramanian, S., Jones, K., Duncan, C., 2003. Multilevel Methods for Public Health Research: Neighborhoods and Health. Oxford University Press, New York. van der Wel, K.A., Dahl, E., Thielen, K., 2011. Social inequalities in ‘sickness’: European welfare states and non-employment among the chronically ill. Soc. Sci. Med. 73 (11), 1608e1617. West, P., 1991. Rethinking the health selection explanation for health inequalities. Soc. Sci. Med. 32 (4), 373e384. World Bank, 2014. GDP per capita, PPP (Current International Dollars) [cited 2014 March 30th]. Available from: http://data.worldbank.org/indicator/NY.GDP.PCAP. PP.CD. I. Campos-Matos, I. Kawachi / Social Science & Medicine 142 (2015) 241e248248 Table 1 Description of main results of eligible publications, according to SES variable and health outcome used Health outcome Medical indicators Functional indicators Subjective health Socioeconomic determinants Place of residence Physical health tended to be better among rural adolescents (Machado-Rodrigues, 2012, Machado-Rodrigues, 2011) and less deprived neighborhoods (Bastos, 2013). Parental perceptions of better neighborhood environments also tended to show an association with better physical (Nogueira, 2013a, Machado-Rodrigues, 2014) but worse mental health (Carvalho, 2014) among children. The only study (Nunes, 2010) showed no association between place of residence and cognitive ability. One study (Humboldt, 2014) showed that life satisfaction was better in rural areas. Race/ethnicity/ culture/language Migrants showed higher mortality (Harding, 2008, Williamson, 2009), worse oral health (Pereira, 2013) and a higher percentage of small preterm births (Harding, 2006b). On the other hand, migrant adolescents had less mental health problems (Neto, 2009 and Neto, 2010) and better cardiorespiratory fitness (Santos, 2011). There were differences in SRH among nationalities in one study (Dias, 2013), but all other studies showed no association between migration, ethnicity or nationality and subjective health (Malmusi, 2014 and Humboldt, 2014). Occupation Most studies showed a strong association between unemployment or less differentiated occupations and worse health (see, for example, Fraga, 2014 or Santos, 2008), although some found no association (for example Alves, 2012 or Bastos, 2013). None found an opposite result. One study (Azevedo, 2012) found people who were unemployed or retired were more likely to suffer from chronic pain. Silva (2014) showed strong associations between employment and more differentiated occupations with SRH. On the other hand, Humboldt (2014) found no association between employment and life satisfaction. Gender/sex Almost all studies showed an association between being female and worse health (see, for example, Santos, 2011 or Bulhões, 2013). Some studies found no gender differences (see, for example, Bastos, 2013 or Neto, 2010) and two found the opposite association (Perelman, 2012 and Stewart-Knox, 2012). Women were more likely to take sickness absence (Masterkaasa, 2014 and Perelman, 2012) and report chronic pain (Azevedo, 2012 and Perelman, 2012), and one study showed men reported more bed days (Perelman, 2012). Cognitive abilities differed between genders, depending on the test used (Martins, 2012, Santos, 2014a). Almost every study showed women had worse subjective health outcomes (see, for example, Bambra, 2009, Dias, 2013 or Pereira, 2011). Religion One study showed no association between religion or spirituality and the onset of major depression (Leurent, 2013). One study showed religious people showed higher life satisfaction (Humboldt, 2014), and another showed no association between religion or spirituality and quality of life or well-being (Vilhena, 2014). Education Lower education tended to show a strong association with worse health in almost all studies (see, for example, Bastos, 2013 or Santos, 2010). There were two exceptions: Lawlor, 2005, who showed that insulin resistance was more common in children of more educated parents and Costa, 2008, who showed girls whose parents were more Education was strongly associated with cognitive ability (Martins, 2012, Nunes, 2010 and Santos, 2014a), chronic pain (Azevedo, 2012) and functional limitations (Eikemo, 2008, Knesebeck, 2006). Better SRH was associated with higher education in all studies (see, for example, Knesebeck, 2006 or Silva, 2014) except one, that showed the opposite (Humboldt, 2014). Campos-Matos et al. International Journal for Equity in Health (2016) 15:26 Page 5 of 10 Table 1 Description of main results of eligible publications, according to SES variable and health outcome used (Continued) educated had more eating disorder symptomatology. Socioeconomic status Married individuals tended to show better health outcomes (see, for example, Harding, 2008 or Williamson, 2009), but had higher odds of being obese (Alves, 2012 and Goulão, 2015). Income, deprivation and financial difficulties showed conflicting results: while most studies tended to show worse health outcomes for more deprived people (see, for example, Pereira, 2013 or Alves, 2012) or no association at all (see, for example, Correia, 2014 or Pimenta, 2011), there were some exceptions that showed, for example, lower prevalence of obesity among homeless people (Oliveira, 2012) or more insulin resistance among children with richer parents (Lawlor, 2005). One study (Azevedo, 2012) found no association between marital status and chronic pain. Early life SES, as measured by height, was strongly associated with chronic pain in women (Perelman, 2014). Objective income (Humboldt, 2014, Silva, 2014) and perceived income (Dias, 2013) were found to be associated with subjective health, but not marital status (Humboldt, 2014) or height, as a measure of early life SES (Perelman, 2014). Social capital One study (Ferreira-Valente, 2014) showed that social support was associated with better psychological functioning. One study (Ferreira-Valente, 2014) showed that social support had a strong association with physical functioning, but not pain intensity. Number of activities outside the home was the only social capital indicator that showed an association with SRH (Silva, 2014). Other analyses showed no association (Vilhena, 2014, Silva, 2014). Note: no eligible publication explored the relationship between ‘race/ethnicity/culture/language’or ‘religion’and functional indicators Legend: SRH Self Rated Health. SES Socioeconomic Status Fig. 2 Diagram representing main results of the associations found in the eligible publications. The visual aspect of the diagram, but not the rules for its construction, was based on the diagram built by Ashley EA et al., “Clinical assessment incorporating a personal genome”The Lancet 375(2010): 1525-35. Note: Font size of health outcomes and circle size of socioeconomic determinants are proportional to the number of eligible publications in which they featured. Black arrows represent strong evidence of an association between socioeconomic indicator and health outcome; grey arrows represent weak evidence and dashed arrows represent evidence of the “negative”associations. In the results obtained, “negative”includes migrant populations having better mental health and married individuals having higher prevalence of obesity. Evidence of all other associations had a “positive” direction, i.e., ill health was associated with lower education, lower income, female gender, unemployment, deprivation, having less differentiated occupations and living in an unfavourable or urban area. Details on how this diagram was constructed are in the online Additional file 3 Campos-Matos et al. International Journal for Equity in Health (2016) 15:26 Page 6 of 10 Education was the most frequently studied determinant of health and for which most evidence exists of health inequalities. Evidence of educational inequalities in obesity was particularly common, especially for women, as the two studies that stratified the analysis by gender found only women showed significant inequalities [43, 44]. This suggests educational inequalities in overweight/obesity are found mostly or exclusively in women. This is not unique for Portugal: Roskam et al. (2010) found that other southern European countries also show high education inequalities in overweight and obesity only for women [57]. In this analysis, Portugal had the highest educational inequalities in overweight and obesity among women in all the countries analysed. This can be a consequence of various factors, such as inequalities in physical activity, dietary patterns or parity. However, both men and women seem to show the same extent of educational inequalities in physical activity and diet in Portugal [58, 59], which makes them unlikely factors in explaining inequalities in obesity seen mostly in women. On the other hand, women with lower education in Portugal have a higher fertility index [60], and since higher parity is strongly associated with obesity [61], this might be the most suitable explanation for the high educational inequalities in overweight and obesity seen for women in Portugal. Education was also strongly associated with SRH [28, 39, 47–49], which is consistent with other international analyses [49, 62]. Interestingly a European comparison among 22 countries found that Portuguese men showed the highest education inequalities in SRH when compared to other countries [49]. However, educational inequalities in SRH should be interpreted with caution. As Huisman, Lenthe and Mackenbach (2007) pointed out, the predictive ability of SRH for mortality varies significantly among educational groups for men [63]. This probably reflects educational differences in men’s health perception, biasing the answers to questions on subjective health. Our review also suggested strong gender inequalities in both SRH and mental health symptoms. Gender-related health inequalities is a broad and complex topic. Despite the prevailing notion that men have higher mortality and women higher morbidity [64], this has been challenged in the literature, and contradictory patterns continue to appear [65, 66]. Additionally, gender inequalities in health are probably a result of multiple factors, including biological and social [67], which raises questions of whether they should be considered as unfair or as unavoidable. Despite this, almost every publication that explored gender differences in our review showed strongly favourable results for men, particularly for mental health symptoms and SRH [32]. Noticeably, no publication explored gender differences in mortality. Academic attention to health inequalities in Portugal has tended to focus on specific topics. Gender and education are by far the most commonly used SES indicators, possibly because they are the most easily measurable, commonly used in surveys with high response rates and high validity of answers and are less affected by reverse causation. Twelve publications also looked at health inequalities between migrants and Portuguese natives; this is surprising considering Portugal is one of the European countries with the lowest proportion of migrant population among its residents [68]. This could be imputed to both the ease of measurement of this variable and the presence of research groups in the country investigating this subject. Other SES indicators appear to have been overlooked. For example, despite the growing literature on the effect of place in health, only a few publications explored this topic, most of which focused on rural/urban differences. There was also a notable deficiency of studies of social capital and poverty, despite Portugal’s high income inequality [6] and considerable risk of poverty and social exclusion [69]. Additionally, despite the growing recognition of the time dimension in the building of health inequalities [70], no publication took a life course approach to how SES indicators might affect health. This, coupled with the scarcity of longitudinal studies, substantially precludes the possibility of assessing causal relationships. This also speaks to a very scarce focus on the elderly - of the 71 eligible publications, only 7 focused on older people, which is surprising in a country where the old-age dependency ratio was the fifth highest in Europe in 2014 [71]. In 2013, the major causes of death in Portugal were diseases of the circulatory system (30), malignant tumours (24), diseases of the respiratory system (12), and endocrine, nutritional and metabolic diseases (5 %) [72]. In this sense, despite malignant tumours being the second most common cause of death, after circulatory diseases, there are strikingly few publications focusing on this health issue (four, of which two are ecological). This might again reflect the absence of a nationally oriented research policy, in part attributable to absence of political attention to this issue [3, 4, 8]. This is also the case for respiratory diseases, which are also almost absent from our analysis. In a recent report of a consortium published by the European Commission on Health Inequalities, Portugal was described as having “[clear] difficulties in measuring and analysing health inequalities”[73] (page 129). Interestingly, the current Portuguese National Health Plan identifies the reduction of child obesity as one of its four goals for 2020, but with no focus on its unequal distribution among socioeconomic groups [74]. This plan does mention the importance of the social determinants of health, but focuses almost exclusively on the access to health care services as a remedy for health inequalities [74]. The limited attention given to health inequalities in Portugal can only be explained with an extensive Campos-Matos et al. International Journal for Equity in Health (2016) 15:26 Page 7 of 10 exploration of multiple factors, but one of these factors is probably the engrained belief that the National Health Service, as a universal and (relatively) inexpensive service at point of care, is enough to face these inequalities. However, this is apparently not true, as this review has shown there are still important health inequalities in Portugal. Tackling these inequalities will demand an important effort to build an organized research and policy strategy that will have to go beyond the National Health Service. It is important to notice that Portugal is amongst the most unequal countries in Europe, so that it could benefit from a more progressive taxation scheme and higher social protection to the poorest, which are major evidence-based and consensual measures to fight inequalities in health [75]. Limitations This review tried to bring together analyses not always comparable among them. In fact, many of these publications focused on specific populations –migrants, children or certain regions in Portugal –that might have particular patterns of health inequalities. This might have hidden inequalities that are not apparent when all groups are pooled together. Our search strategy might have also excluded important publications, namely international comparisons that included a Portuguese sample not specifically mentioned. However, we tried to overcome this by searching for publications by researchers known for having published in this area. The quality of the analyses in the reviewed publications was found to be heterogeneous, with some presenting highly reliable analyses and others relying on ‘convenience samples’,oronsmall sample sizes. Following the PRISMA guidelines, we chose not to score nor select the publications based on ‘quality’, but to carry out a brief assessment of strengths and limitations on each (table in Additional file 2). Also, we focused our review on papers published in indexed peer-reviewed journals according to good practices of scientific research, but this may have excluded important publications, in particular from the grey literature. Finally, we restricted our analysis to health outcomes, and did not consider mediating factors such as lifestyle and healthcare use. Also, we did not consider studies on interventions to decrease inequalities in health. We adopted this strategy to avoid a too large scope for the review, which would have complicated the identification of general trends and interpretations. Further research should focus on these connected issues. Along this paper, we referred to “inequalities”in health instead of other possible terms such as “inequity”or “differences”. In particular, inequity refers to differences that are unjust, unfair and avoidable [76]. This option was made because the concept of inequality is more neutral in terms of interpretations and value judgements, whereas the term “inequity”implies strong assumptions about the causes of differences, which none of the reviewed papers could confirm. Additionally, most reviewed papers referred to inequalities in health, so we opted to be faithful to authors’interpretations. Conclusions We have shown that there is strong evidence of socioeconomic health inequalities in Portugal and comparative analyses show that these are possibly one of the highest among European countries. We identified education and gender as the main determinants of health inequalities, affecting mostly the distribution of obesity, self-rated health and mental health symptoms. The publications we identified also reflect the absence of a nationally oriented research strategy on health inequalities focusing on the most prevalent diseases (such as malignant tumours and respiratory diseases), determining factors of inequalities (living contexts, poverty or social capital) and vulnerable populations (such as the elderly). We hope this review will help guide decision-making to tackle these issues, as has long been recommended. Additional files Additional file 1: Detailed search strategy. (PDF 79 kb) Additional file 2: Table of extracted data from the 71 eligible publications. (PDF 196 kb) Additional file 3: Description of rules adopted to build diagram in Fig. 2 of the main text. (DOCX 16 kb) Additional file 4: Complete list of the seventy one eligible publications identified by the systematic review, by alphabetical order. (PDF 70 kb) Competing interests The authors have no competing interests to declare. Authors’contributions ICM contributed to study conceptualisation and design, data acquisition, selection and analysis, and drafted the manuscript. GR contributed to study conceptualisation and design, data selection and manuscript revision. JP contributed to study conceptualisation and design, data selection and extraction and manuscript revision. All authors approved the final manuscript. Acknowledgements The authors wish to thank the contribution of the researchers from the Nova Healthcare Initiative, who provided comments on a late draft of this paper. The final paper is the responsibility of the authors. This project has been financed by the Fundação para Ciência e Tecnologia (Grant VIH/SAL/0065/2011). Author details 1 Instituto de Higiene e Medicina Tropical, Universidade NOVA de Lisboa, Lisbon, Portugal. 2 Centro de Investigação em Saúde Pública, Lisbon, Portugal. 3 Global Health and Tropical Medicine, Instituto de Higiene e Medicina Tropical, NOVA University of Lisbon, Lisbon, Portugal. 4 Escola Nacional de Saúde Pública, Universidade NOVA de Lisboa, Lisbon, Portugal. Received: 23 November 2015 Accepted: 1 February 2016 Campos-Matos et al. International Journal for Equity in Health (2016) 15:26 Page 8 of 10 References 1. Marmot M, Friel S, Bell R, Houweling TA, Taylor S. Closing the gap in a generation: health equity through action on the social determinants of health. Lancet. 2008;372(9650):1661–9. 2. Black D, Morris J, Smith C, Townsend P. Inequalities in health: report of a Research Working Group. London: Department of Health and Social Security; 1980. p. 19. 3. Pereira J, Furtado C. Equidade em saúde: do conhecimento à acção. In: A Nova Saúde Pública: A saúde Pública na Era do Conhecimento, editor. DGS/ENSP. Lisbon: Gradiva; 2011. 4. World Health Organization. WHO evaluation of the national health plan of Portugal (2004–2010). Copenhagen: WHO Regional Office for Europe; 2010. 5. World Health Organization. Portugal Health System Performance Assessment. WHO and Alto Comissariado da Saúde [Portuguese High Comissioner for Health]. 2010. 6. European Commission. Social Situation Monitor - Income inequality in EU countries. 2015. http://ec.europa.eu/social/main.jsp?catId=1050&intPageId= 1870&langId=en. Accessed 22 November 2015. 7. Mackenbach JP, Stirbu I, Roskam AJ, Schaap MM, Menvielle G, Leinsalu M, et al. Socioeconomic inequalities in health in 22 European countries. N Engl J Med. 2008;358(23):2468–81. 8. Bago d’Uva T. Equidade no sector da saúde em Portugal. In: Almedina E, editor. 30 Anos do Serviço Nacional de Saúde - um percurso comentado. Lisbon: Simões, Jorge; 2010. 9. Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. BMJ. 2009;349:g7716. 10. O’Neill J, Tabish H, Welch V, Petticrew M, Pottie K, Clarke M, et al. Applying an equity lens to interventions: using PROGRESS ensures consideration of socially stratifying factors to illuminate inequities in health. J Clin Epidemiol. 2014;67(1):56–64. 11. Blaxter M. A comparison of measures of inequality in morbidity. Gower: Health inequalities in European countries Aldershot; 1989. p. 199–230. 12. Public Health Institute of the University of Porto [Instituto de Saúde Pública da Universidade do Porto]. Cohorts: EPIPorto Porto, Portugal: ISPUP. http://ispup.up.pt/research/research-structures/cohorts/?lang=en. Accessed 22 November 2015. 13. Alda de Caetano Carvalho & José Pereira Miguel. Inquérito Nacional de Saúde 2005/2006. Instituto Nacional de Estatística and Instituto Nacional de Saúde Dr. Ricardo Jorge. 2009. 14. Public Health Institute of the University of Porto [Instituto de Saúde Pública da Universidade do Porto]. Cohorts: Generation XXI Porto, Portugal: ISPUP. http://ispup.up.pt/research/research-structures/cohorts/geracaoxxi/. Accessed 22 November 2015. 15. Public Health Institute of the University of Porto [Instituto de Saúde Pública da Universidade do Porto]. Cohorts: EPITeen Porto, Portugal: ISPUP. http://ispup.up.pt/research/research-structures/cohorts/epiteen/. Accessed 22 November 2015. 16. Miranda MJ, Costa C, Santana P, Barrozo LV. Associação espacial entre variáveis socioeconômicas e risco relativo de nascimentos pré-termo na Região Metropolitana de São Paulo (RMSP) e na Área Metropolitana de Lisboa (AML). Saúde Soc. 2014;23(4):1142–53. 17. Harding S, Boroujerdi M, Santana P, Cruickshank J. Decline in, and lack of difference between, average birth weights among African and Portuguese babies in Portugal. Int J Epidemiol. 2006;35(2):270–6. 18. Oliveira C, Economou T, Bailey T, Mendonça D, Pina M. The interactions between municipal socioeconomic status and age on hip fracture risk. Osteoporos Int. 2015;26(2):489–98. 19. von Humboldt S, Leal I, Pimenta F. Living well in later life: the influence of sense of coherence, and socio-demographic, lifestyle and health-related factors on older adults’satisfaction with life. Applied Research in Quality of Life. 2014;9(3):631–42. 20. Bastos J, Peleteiro B, Barros R, Alves L, Severo M, Fátima Pina M, et al. Sociodemographic determinants of prevalence and incidence of Helicobacter pylori infection in Portuguese adults. Helicobacter. 2013;18(6):413–22. 21. Williamson LM, Rosato M, Teyhan A, Santana P, Harding S. AIDS mortality in African migrants living in Portugal: evidence of large social inequalities. Sex Transm Infect. 2009;85(6):427–31. 22. Harding S, Teyhan A, Rosato M, Santana P. All cause and cardiovascular mortality in African migrants living in Portugal: evidence of large social inequalities. Eur J Cardiovasc Prev Rehabil. 2008;15(6):670–6. 23. Neto F. Predictors of mental health among adolescents from immigrant families in Portugal. J Fam Psychol. 2009;23(3):375. 24. Neto F. Mental health among adolescents from returned Portuguese immigrant families from North America. N Am J Psychol. 2010;12(2):265. 25. Santos DA, Silva AM, Santa-Clara H, Matias CN, Fields DA, Sardinha LB. Determinant factors of cardiorespiratory fitness in Portuguese adolescents of different ethnicities. Revista Brasileira de Cineantropometria & Desempenho Humano. 2011;13(4):243–9. 26. Fraga S, Marques-Vidal P, Vollenweider P, Waeber G, Guessous I, Paccaud F, et al. Association of socioeconomic status with inflammatory markers: A two cohort comparison. Preventive medicine. 2014. 27. Harding S, Santana P, Cruickshank JK, Boroujerdi M. Birth weights of black African babies of migrant and nonmigrant mothers compared with those of babies of European mothers in Portugal. Ann Epidemiol. 2006;16(7):572–9. 28. Silva PA. Individual and social determinants of self-rated health and wellbeing in the elderly population of Portugal. Cadernos de saude publica. 2014;30(11):2387–400. 29. Sousa-Ribeiro M, Sverke M, Coimbra JL. Perceived quality of the psychosocial environment and well-being in employed and unemployed older adults: the importance of latent benefits and environmental vitamins. Econ Ind Democr. 2014;35(4):629–52. 30. Azevedo LF, Costa-Pereira A, Mendonça L, Dias CC, Castro-Lopes JM. Epidemiology of chronic pain: a population-based nationwide study on its prevalence, characteristics and associated disability in Portugal. The journal of pain. 2012;13(8):773–83. 31. Alves E, Correia S, Barros H, Azevedo A. Prevalence of self-reported cardiovascular risk factors in Portuguese women: a survey after delivery. International journal of public health. 2012;57(5):837–47. 32. Perelman J, Fernandes A, Mateus C. Gender disparities in health and healthcare: results from the Portuguese National Health Interview Survey. Cadernos de saude publica. 2012;28(12):2339–48. 33. Stewart-Knox B, Duffy ME, Bunting B, Parr H, de Almeida MDV, Gibney M. Associations between obesity (BMI and waist circumference) and sociodemographic factors, physical activity, dietary habits, life events, resilience, mood, perceived stress and hopelessness in healthy older Europeans. BMC Public Health. 2012;12(1):424. 34. Carvalho J, Nobre P. Gender issues and sexual desire: the role of emotional and relationship variables. J Sexual Med. 2010;7(7):2469–78. 35. Costa C, Ramos E, Severo M, Barros H, Lopes C. Determinants of eating disorders symptomatology in Portuguese adolescents. Archives of pediatrics & adolescent medicine. 2008;162(12):1126–32. 36. Lawlor DA, Harro M, Wedderkopp N, Andersen LB, Sardinha LB, Riddoch CJ, et al. Association of socioeconomic position with insulin resistance among children from Denmark, Estonia, and Portugal: cross sectional study. BMJ. 2005;331(7510):183. 37. Correia S, Barros H. Small-for-gestational age Portuguese babies: the effect of childhood social environment, growth and adult socioeconomic conditions. Prev Med. 2015;70:102–7. 38. Mello T, Antunes I, Waldman E, Ramos E, Relvas M, Barros H. Prevalence and severity of dental caries in schoolchildren of Porto, Portugal. Community Dent Health. 2008;25(2):119–25. 39. Schütte S, Chastang J-F, Parent-Thirion A, Vermeylen G, Niedhammer I. Social differences in self-reported health among men and women in 31 countries in Europe. Scandinavian journal of public health. 2013;41(1):51–7. 40. Correia S, Rodrigues T, Barros H. Socioeconomic variations in female fertility impairment: a study in a cohort of Portuguese mothers. BMJ open. 2014;4(1):e003985. 41. Pimenta F, Leal I, Maroco J, Ramos C. Perceived control, lifestyle, health, socio-demographic factors and menopause: Impact on hot flashes and night sweats. Maturitas. 2011;69(4):338–42. 42. Ferreira-Valente MA, Pais-Ribeiro JL, Jensen MP. Associations between psychosocial factors and pain intensity, physical functioning, and psychological functioning in patients with chronic pain: a cross-cultural comparison. Clin J Pain. 2014;30(8):713–23. 43. Camões M, Lopes C, Oliveira A, Santos AC, Barros H. Overall and central obesity incidence in an urban Portuguese population. Prev Med. 2010;50(1):50–5. 44. Santos A-C, Barros H. Prevalence and determinants of obesity in an urban sample of Portuguese adults. Public Health. 2003;117(6):430–7. 45. Nogueira H, Ferrão M, Gama A, Mourão I, Rosado Marques V, Padez C. Perceptions of neighborhood environments and childhood obesity: Campos-Matos et al. International Journal for Equity in Health (2016) 15:26 Page 9 of 10 evidence of harmful gender inequities among Portuguese children. Health & place. 2013;19:69–73. 46. Bingham DD, Varela‐Silva MI, Ferrão MM, Augusta G, Mourão MI, Nogueira H, et al. Socio‐demographic and behavioral risk factors associated with the high prevalence of overweight and obesity in portuguese children. Am J Hum Biol. 2013;25(6):733–42. 47. Dias S, Gama A, Martins MO. Health status and preventative behaviors of immigrants by gender and origin: a Portuguese cross‐sectional study. Nursing Health Sci. 2013;15(3):309–17. 48. Eikemo TA, Huisman M, Bambra C, Kunst AE. Health inequalities according to educational level in different welfare regimes: a comparison of 23 European countries. Sociology of health & illness. 2008;30(4):565–82. 49. Von Dem Knesebeck O, Verde PE, Dragano N. Education and health in 22 European countries. Soc Sci Med. 2006;63(5):1344–51. 50. Pereira M, Canavarro MC. Gender and age differences in quality of life and the impact of psychopathological symptoms among HIV-infected patients. AIDS Behav. 2011;15(8):1857–69. 51. Bambra C, Pope D, Swami V, Stanistreet D, Roskam A, Kunst A, et al. Gender, health inequalities and welfare state regimes: a cross-national study of 13 European countries. J Epidemiol Community Health. 2009;63(1):38–44. 52. Ribeiro O, Teixeira L, Araújo L, Afonso RM, Pachana N. Predictors of anxiety in centenarians: health, economic factors, and loneliness. International Psychogeriatrics. 2014. p. 1–10. 53. Amaral MOP, de Figueiredo Pereira CM, Martins DIS, Sakellarides CT. Prevalence and risk factors for insomnia among Portuguese adolescents. Eur J Pediatr. 2013;172(10):1305–11. 54. Carvalho M, de Matos MG. Psychosocial determinants of mental health and risk behaviours in adolescents. Global J Health Sci. 2014;6(4):22. 55. Bulhões C, Ramos E, Lindert J, Dias S, Barros H. Depressive symptoms and its associated factors in 13-year-old urban adolescents. International journal of environmental research and public health. 2013;10(10):5026–38. 56. Santos NC, Costa PS, Cunha P, Portugal-Nunes C, Amorim L, Cotter J, et al. Clinical, physical and lifestyle variables and relationship with cognition and mood in aging: a cross-sectional analysis of distinct educational groups. Frontiers in aging neuroscience. 2014. p. 6. 57. Roskam A-JR, Kunst AE, Van Oyen H, Demarest S, Klumbiene J, Regidor E, et al. Comparative appraisal of educational inequalities in overweight and obesity among adults in 19 European countries. Int J Epidemiol. 2010;39(2):392–404. 58. Gal DL, Santos A-C, Barros H. Leisure-time versus full-day energy expenditure: a cross-sectional study of sedentarism in a Portuguese urban population. BMC Public Health. 2005;5(1):16. 59. Moreira PA, Padrão PD. Educational and economic determinants of food intake in Portuguese adults: a cross-sectional survey. BMC Public Health. 2004;4(1):58. 60. Oliveira IT. Fecundidade e educação. Revista de Estudos Demográficos. 2007;40:5–19. 61. Davis EM, Zyzanski SJ, Olson CM, Stange KC, Horwitz RI. Racial, ethnic, and socioeconomic differences in the incidence of obesity related to childbirth. Am J Public Health. 2009;99(2):294. 62. Kunst AE, Bos V, Lahelma E, Bartley M, Lissau I, Regidor E, et al. Trends in socioeconomic inequalities in self-assessed health in 10 European countries. Int J Epidemiol. 2005;34(2):295–305. 63. Huisman M, Van Lenthe F, Mackenbach J. The predictive ability of selfassessed health for mortality in different educational groups. Int J Epidemiol. 2007;36(6):1207–13. 64. Oksuzyan A, Juel K, Vaupel JW, Christensen K. Men: good health and high mortality. Sex differences in health and aging. Aging Clin Exp Res. 2008;20(2):91–102. 65. McDonough P, Walters V. Gender and health: reassessing patterns and explanations. Soc Sci Med. 2001;52(4):547–59. 66. Macintyre S, Hunt K, Sweeting H. Gender differences in health: are things really as simple as they seem? Soc Sci Med. 1996;42(4):617–24. 67. Rieker PP, Bird CE. Rethinking gender differences in health: why we need to integrate social and biological perspectives. J Gerontol Ser B Psychol Sci Soc Sci. 2005;60(Special Issue 2):S40–S7. 68. Eurostat. Migrants in Europe. Belgium: A statistical portrait of the first and second generation; 2011. 69. European Commission. People at risk of poverty or social exclusion. http://ec.europa.eu/eurostat/statistics-explained/index.php/People_at_risk_ of_poverty_or_social_exclusion. Accessed 22 November 2015. 70. Kawachi I, Subramanian S, Almeida-Filho N. A glossary for health inequalities. J Epidemiol Community Health. 2002;56(9):647–52. 71. Eurostat. Old-age dependency ratio per 100 persons. http://ec.europa.eu/ eurostat/tgm/table. do?tab=table&init=1&language=en&pcode=tsdde510&plugin=1. Accessed 3 Feb 2015. 72. Direção-Geral da Saúde. A Saúde dos Portugueses. Perspetiva 2015 Lisbon: Direção-Geral da Saúde. 2015. 73. Marmot M. Health inequalities in the EU. Final report of a consortium. Health inequalities in the EU Final report of a consortium. 2013. 74. Direcção-Geral da Saúde. Plano Nacional de Saúde: Revisão e Extensão a 2020. Lisbon: Direcção-Geral da Saúde; 2015. 75. Smith KE, Eltanani MK. What kinds of policies to reduce health inequalities in the UK do researchers support? Journal of Public Health. 2015;37(1):6–17. 76. Braveman P, Gruskin S. Defining equity in health. J Epidemiol Community Health. 2003;57(4):254–8. • We accept pre-submission inquiries • Our selector tool helps you to find the most relevant journal • We provide round the clock customer support • Convenient online submission • Thorough peer review • Inclusion in PubMed and all major indexing services • Maximum visibility for your research Submit your manuscript at www.biomedcentral.com/submit Submit your next manuscript to BioMed Central and we will help you at every step: Campos-Matos et al. International Journal for Equity in Health (2016) 15:26 Page 10 of 10 The relationship between context and health inequalities Chapter 2. Results 95 2.4. Shifting determinants of health inequalities in unstable times: Portugal as a case study Reference: Campos-Matos I, Russo G, Gonçalves L. Shifting determinants of health inequalities in unstable times: Portugal as a case study. Accepted for publication in the European Journal of Public Health in May 2017. DOI: https://dx.doi.org/10.1093/eurpub/ckx080 Online supplementary data for this publication is in Appendix 4. The relationship between context and health inequalities Chapter 2. Results 96 European Journal of Public Health, 1–6 ßThe Author 2017. Published by Oxford University Press on behalf of the European Public Health Association. All rights reserved. doi:10.1093/eurpub/ckx080 ......................................................................................................... Shifting determinants of health inequalities in unstable times: Portugal as a case study Ine ˆs Campos-Matos 1,2 , Giuliano Russo 3 , Luzia Gonc¸alves 1,4 1 Instituto de Higiene e Medicina Tropical, Nova University of Lisbon, Lisbon, Portugal 2 Centro de Investigac¸a ˜o em Sau ´de Pu ´blica, Lisbon, Portugal 3 Department of Primary Care and Public Health, Queen Mary University of London, London, UK 4 Centro de Estatı ´stica e Aplicac¸o ˜es da Universidade de Lisboa, Lisbon, Portugal Correspondence: Ine ˆs Campos-Matos, Departamento de Sau ´de Internacional e Bioestatı ´stica, Instituto de Higiene e Medicina Tropical, Rua da Junqueira, n100, 1349-008 Lisboa, Portugal, Tel: +44 750 282 3003, e-mail: [email protected] Background: We explore how health inequalities (HI) changed in Portugal over the last decade, considering it is one of the most unequal European countries and has gone through major economic changes. We describe how inequalities in limitations changed considering different socioeconomic determinants, in order to understand what drove changes in HI. Methods: We used cross-sectional waves from the European Survey on Income and Living Conditions database to determine how inequalities in health limitations changed between 2004 and 2014 in Portugal in residents aged 16 years and over. We calculated prevalence estimates of limitations and differences between income terciles, the concentration index for each year and its decomposition and multiple logistic regressions to estimate the association between socioeconomic determinants and limitations. Results: The prevalence of health limitations increased in Portugal since 2004, especially after 2010, from 35 to 47%. But the difference between top and bottom income terciles decreased from 23 to 10 percentage points, as richer people experienced a steeper increase. This was driven by an increase in prevalence among economically active people, who, from 2011 onwards, had more limitations (OR and 95% CI were 2.42 [2.13–2.75] in 2004 and 0.71 [0.65–0.78] in 2014). Conclusion: These results suggest worsening health in Portugal in the last decade, possibly connected to periods of economic instability. However, absolute HI decreased considerably in the same period. We discuss the possible role of diverse adaptation capacity of socioeconomic groups, and of high emigration rates of young, healthier people, reflecting another side of the ‘migrant health effect’. ......................................................................................................... Introduction Socioeconomic health inequalities (HI) are ubiquitous. They have been observed worldwide as long as data have been available. It appears that, regardless of place and time, health tends to follow the patterning of socioeconomic differences (1). Various socioeconomic indicators—education, financial resources, employment or occupation—determine HI, operating through different pathways. Education leads to better information, cognitive abilities and determines preferences (2); financial resources, such as income or wealth, allow individuals to access health-producing resources, such as healthcare or housing. Employment not only provides income, but also a sense of control over one’s life, lack of which is strongly associated with important stress reactions, which can deteriorate health (3). People with higher occupational grades also tend to have a stronger sense of control over their health, their jobs and their lives (4), but occupation can also reflect an individual’s place in society, showing the effect of rank and subjective feelings towards one’s position in society (2). The simultaneous analysis of various socioeconomic determinants of HI can provide clues as to which processes are more important in the creation of HI (5). A better understanding of which processes shape HI will help to build a base to design policies that tackle them effectively. Portugal is a particularly interesting case study for HI. The country has had low economic growth (6), and despite substantial investments in social protection, education and healthcare (7,8), remains one of the most unequal European Union countries in income distribution (9). This is reflected in health distribution: several analyses found Portugal to have some of the highest HI among European countries (10–12). Additionally, Portugal has gone through a period of economic crisis and implementation of austerity measures in the last years, that have led to a spike in emigration (13) and a deterioration of public social services (14). A recent review of the impact of economic crises found that they tended to aggravate HI in a variety of countries (15). However, the review noted that results were variable, perhaps due to differing welfare policies, or the diversity of health and socioeconomic variables. Poor understanding of how economic crises shape HI hinders the interpretation of these results. This work aims to support policy choices that attempt to mitigate the effect of economic crises or other contextual changes on HI. To do this, we describe how HI changed in Portugal over the last decade, in light of the important social and macroeconomic changes that the country has been through, and how the socioeconomic determinants of these inequalities changed. We used data from the cross-sectional waves of the European Survey on Income and Living Conditions (EU-SILC), from 2004 to 2014. Portugal is used as a case study, but this analysis is applicable to other countries as it describes how determinants of HI can be shaped by contextual transformations. This is particularly useful considering that many countries have recently gone through similar macroeconomic changes as Portugal. Methods This analysis was performed using data from the Portuguese crosssectional waves of EU-SILC between 2004 and 2014 (provided by Eurostat in December 2015). EU-SILC is an annual survey carried out in several European countries with a mixed longitudinal and cross-sectional design. Despite this mixed design, cross-sectional samples are representative of the target population when appropriate weights are used (16). Portugal participates since 2004 using a stratified, multi-stage, household-based sample. The survey collects data on living conditions and includes three health related questions: limitations in daily activities due to health problems, self-reported health (SRH) and chronic conditions. We used ‘limitations’ as our health outcome. Individuals were asked if they were limited in activities they usually did because of health problems. Possible answers included ‘Yes, strongly limited’, ‘Yes, limited’ or ‘No’. The first two options were collapsed, creating a binary variable (1 = ‘with limitations’, 0 = ‘without limitations’). This health outcome was chosen as it provides an objective measure than SRH and should capture health status more accurately (17). The initial descriptive analysis was also done for the other two health variables: SRH and chronic conditions. SRH is a widely used survey measure in which respondents rate their overall health; we used SRH as a binary variable in which ‘bad’ and ‘very bar’ health were the outcome. ‘Chronic conditions’ is a self-assessed question in which respondents are asked whether they have a chronic condition; this was also used as a binary variable, in which having a chronic condition was the outcome. The following variables were included in the analysis: Age at interview (in years). Sex (male or female). Income: yearly household equivalised disposable income, in euros, deflated using the harmonised index of consumer prices (18). Education: defined by highest International Standard Classification of Education (ISCED) level attained (19), categorised into ‘primary or less’ or ‘more than primary’. Occupation: based on the International Standard Classification of Occupations (ISCO) used in EU-SILC, occupations were categorised in white or blue collar, following previous work (ISCO codes 1–5 were white collar, 6–9 blue collar and armed forces were excluded) (20). Activity: based on the EU-SILC variable ‘self-defined current economic status’, people were categorised as ‘active’ if they defined themselves as being employed (part or full time), in training or studying, or fulfilling domestic tasks; and ‘inactive’ if they were unemployed, retired, unfit to work or in the ‘other inactive’ category. Savings: EU-SILC further asks households about their capacity to face unexpected financial expenses and to afford one-week annual holiday away from home. These two variables were merged and transformed into a binary variable so that the value ‘0’ was attributed to households who could afford both and ‘1’ to the remaining households. We used the complete sample of residents aged 16 and over. The proportion of individuals who had limitations was calculated for each year in the overall sample, within each income tercile, and stratified by age groups. Income terciles were calculated according to the distribution of income for each year. The concentration index (CIx) for income-related inequalities in limitations was calculated for each year. The CIx is a measure of inequalities based on the health concentration curve. This curve is the result of plotting of the cumulative percentage of individuals, ranked by income, with the cumulative percentage of limitations. In this plot, perfect equality is represented by a diagonal line, showing an equal distribution of limitations among the population, regardless of income. The CIx is calculated as twice the area between the concentration curve and the line of perfect equality. When there is perfect equality, the CIx is zero. By convention, if all limitations are concentrated in the richest (poorest) person, the CIx is 1 (-1). However, with dichotomous outcome variables, the CIx is not within the [-1,1] range and between-year comparability may be limited; following Wagstaff (21), to minimise this limitation, we normalised the CIx by dividing it by 1 minus the proportion of respondents reporting limitations in each year. Wagstaff et al. (22) showed that the CIx can be decomposed into contributions of individual factors to the income-related HI. This analysis allows for the quantification of how each factor (i.e. each socioeconomic variable) contributes to the overall distribution of the health outcome among income ranks. The contribution of each factor is the product of the elasticity of that factor with respect to the health variable (i.e. the proportional change of a specific factor in relation to a proportional change in the health variable) and the CIx of that factor (i.e. the degree of incomerelated inequality of that factor). Finally, we performed a multiple logistic regression for each year, using the dichotomous health variable (limitations) as an outcome. We included all the demographic and socioeconomic variables listed above as explanatory variables: age, sex, income, education, occupation, activity and savings. These were all added to the model simultaneously. Analyses were weighed by a personal cross-sectional weight provided by the EU-SILC database, which controls for geographical, household size, gender, and age group distribution, and nonresponse within each household. Analyses were done on SPSS Statistics v21 and in ADePT Software v6.0 using a non-linear model for the CIx. Results Table 1 summarises the sample characteristics. Yearly sample size ranged from 9947 individuals in 2007 to 14 650 in 2014. Average age increased from 46.3 to 49.0-years-old from 2004 to 2014. The proportion of individuals with limitations also increased from 35.2 to 47.3%. Median income increased between 2004 and 2012, from 5869 to 8366 euros per year, and dropped to 8265 euros in 2014. There was also an increase in the proportion of people with secondary and tertiary education and in white-collar occupations, both representing approximately half the sample in 2014. The proportion of active people decreased from 70% in 2004 to 58.5% in 2014. Figure 1 shows the proportion of individuals with limitations by year. This proportion was stable at around 30% until 2011, when it increased to 43%, and then increased again in 2014 to 47%. These changes occurred in all income terciles, but a few differences were noticeable: (i) in almost every year, the proportion of people with limitations was higher in the first tercile (the lowest-income population group), followed by the second, and lowest in the third; (ii) this difference was stable until 2011, when the proportion of limitations increased in all terciles, most markedly in the second and third; (iii) this led to a decrease in the absolute difference in limitations inequalities between the first and third income terciles. Figure 1 points out the absolute differences between the first and third income terciles in four years (23% age points in 2004 and 2010, 16 in 2011, and 10 in 2014). When stratified by age groups, the analysis presented in figure 1 shows that inequalities in limitations were highest in the older age groups, the increase in limitations in 2011 occurred in younger age groups, and the oldest age groups showed a decrease in limitations in 2012 (Supplementary figure S1A). The CIx was negative every year, as the prevalence of limitations was higher in poorer people (figure 2). The CIx ranged between 0.15 and 0.18 (in absolute values) until 2010 and dropped in 2011 to 0.09 and to 0.05 in 2014. Until 2010, every socioeconomic variable had a negative contribution to the CIx, meaning that they all contributed to pro-poor inequality in the distribution of limitations. However, after 2010 there were a few noticeable changes. First, activity now gave a positive contribution to the CIx. Detailed analysis of the contribution of each variable (Supplementary table S1A) showed that the elasticity of limitations with respect to activity changed in 2011, from positive to a negative contribution; the CIx of activity, on the other hand, remained stable. This means that, in all years, inactive people had lower incomes when compared with active people. However, while limitations were more prevalent in inactive people until 2010, they were more prevalent in active people after this year. 2of6 European Journal of Public Health The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 105 We used data from six rounds of the ESS, between 2002 and 2012, which included 36 countries and 237,535 individuals. Data was organized in three levels – individuals, years and countries – and analysed using multilevel statistical techniques. Countries were grouped according to their welfare regime type and analyses were done separately for each group. Each individual was attributed one of three social mobility paths: upward, stable, or downward, according to the differences between their own and their mother and father’s educational achievement (analyses were done separately for mother and father). Health was operationalized as a binary variable: 1 for ‘bad’ or ‘very bad’ SAH and 0 for ‘fair’, ‘good’, and ‘very good’. In order to ensure that the social movement itself was analysed and the effects of childhood circumstances were excluded, analyses were controlled for parental education. We calculated relative (OR) and absolute (risk difference) measures of the association between social mobility and SAH for each welfare regime type. Results showed that upward mobility (when compared to being socially ‘stable’) was positively associated with better health in all welfare regime types, measured both from mother and father’s achievement, using absolute and relative measures. On a relative scale, these results were statistically significant for p<0.05. On an absolute scale, former USSR countries showed the biggest and only significant difference for upward movement (4.1 and 3.8% difference, when social mobility was measured from the mother’s or the father’s achievement, respectively). Scandinavian countries showed the smallest and not significant differences: 0.8 and 0.9% difference, from mother or father’s, respectively. Overall, this analysis showed that social mobility was associated with differences in health in all welfare regime types, but Scandinavian countries showed the smallest association. Despite having high levels of social mobility, these countries seemed to efficiently separate it from health, more so than countries from other welfare regime types. These results suggest that the ‘paradox’ of high HI in northern European countries is unlikely to be due to social mobility differences among welfare regime types. This analysis did not test other hypotheses on why HI remain high in northern European countries, but showed that welfare regimes play a role in determining both levels of population health and health distribution in European countries. The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 106 3.1.3. Social Determinants of Health in Portugal As one of the most unequal countries in Europe, both in terms of income distribution (13) and health distribution (14), Portugal presents an interesting case study for understanding HI. Despite the high levels of inequality, WHO identified HI as an ‘important policy gap’ in the Portuguese National Health Plan (15), which reflects the low political interest in the subject and the absence of a national strategy in place to tackle these inequalities. It was thus important to systematize current knowledge on socioeconomic HI in Portugal, to set a stepping-stone towards a possible strategy to tackle HI in the country. For this, we carried out a SR of the literature that gathered the existing evidence about socioeconomic HI in Portugal. The PRISMA statement was used to guide and report the review (16). The review began by defining what measures of SES would be included. For this, the PROGRESS2 framework was followed, standing for Place of residence, Race / ethnicity / culture / language, Occupation, Gender/sex, Religion, Education, Socioeconomic status and Social capital (17). Both individual and contextual determinants were included. Healthcare utilization or access, and health related behaviours were excluded from the analysis, as we were interested solely on inequalities in health outcomes. We included every study that quantified an association between the socioeconomic and health variables, and controlled for, at least, gender and age. Studies that used data from 2000 onwards and from the Portuguese resident population (regardless of nationality) were included. We excluded qualitative studies. Articles written in Portuguese and English were included. We searched Scopus, Web of Science and Pubmed for papers that met the eligibility criteria. Additionally, we scoped publications of researchers in Portugal who regularly publish research in this area for publications that met the eligibility criteria, to complement our online search. Data was extracted from the selected articles and a brief quality evaluation was performed. Results were presented using a narrative description and a diagram that summarized the findings. The final selection included seventy-one papers, all of which reported observational studies, and most of which used cross-sectional data. Most publications reported 2 The PROGRESS framework was created as an aide-memoir to help researchers apply an equity lens to their research, and public health professionals to consider all potentially inequitable circumstances in public health interventions. The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 107 significant HI that favoured individuals or groups with better social standing. Some associations were particularly common and showed strong evidence of HI: lower education with obesity and with subjective ill health; and female gender with mental health symptoms and subjective ill health. Of the seventy-one selected papers in the SR, seventeen analysed, in one way or another, the effect of a contextual variable. Seven of these studies had an ecological design (18-24). Four investigated how an individual’s perceptions of their neighbourhood were associated to their health (25-28). Four other studies explored the difference between residents of urban and rural settings (29-32). The two remaining studies looked at differences between deprived and affluent neighbourhoods (33) and municipalities (34). Overall, these were very heterogeneous studies, looking at different topics, different populations, and with differing degrees of quality. This review offered a systematization of current evidence on HI in Portugal, which had not been done before. Its results show that, while there are a considerable number of publications touching on the subject, they tend to focus on specific topics, reproducing similar results; they do not necessarily focus on what are the most prevalent health and social issues in Portugal; and they tend to use similar methodologies. This analysis was also able to identify what are the most important HI in Portugal for which evidence is available – education and gender inequalities in obesity, SAH and mental health – and the major gaps in the research literature regarding this topic – analyses focused on the most prevalent health issues, the most important SES factors, and the most vulnerable populations. It is clear from these results that, at the present moment, there is no nationally oriented research strategy which would be crucial to guide research in a country where HI remain particularly high. 3.1.4. Shifting Determinants of Health Inequalities in Portugal Many individual socioeconomic variables can determine HI. Education, one of the most commonly used indicators, can determine better cognitive abilities and better knowledge, that allow individuals to prevent illness, have better health, and better manage disease (35). Financial resources can allow individuals to acquire healthproducing resources (36). Employment provides income, but also a sense of purpose, The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 108 a structure to the day, and social connections, which are all associated with better health (37). Occupation, on the other hand, is a reflection of social rank, and has also been strongly connected to multiple health outcomes; famously, the Whitehall studies showed that these differences were a consequence of feelings of control over one’s life, which are stronger among people in higher occupational ranks (38). This variety of SES indicators is a reflection of the multitude of pathways that lead to HI. Observing which indicators are more important, and how these change over time, can give important clues to which processes are the most relevant in creating HI. Portugal has been through important changes over the last decade, with periods of political instability and implementation of austerity measures (39). These transformations provide a unique opportunity to observe how contextual changes can lead to changes in overall health, health distribution, and to the determinants of HI. With the fourth publication, we aimed to describe how the determinants of HI changed in Portugal over the last decade, in order to understand what processes created HI and how these changed over time. This understanding can hopefully lead to informed policies that can successfully tackle HI. The fourth and last publication of this dissertation used data from the cross-sectional waves of the Portuguese sample of EU-SILC from 2004 to 2014 was used, with limitations in daily activities as the outcome variable. We calculated the prevalence of limitations in each income tercile in each year and absolute differences between first and third terciles for selected years. We then calculated the CIx for each year and its decomposition in various SES indicators: occupation, employment, education, income, savings, age, and sex. Finally, we ran a multiple logistic regression analysis for each year, to determine the OR for each of these indicators. The prevalence of limitations was found to have increased in Portugal in the last decade, especially after 2010. However, this increase was steeper in richer terciles, which led to a decrease in both absolute and relative inequalities in limitations. Analysis of the CIx decomposition and of the OR showed that professional activity was the main determinant of the decrease in inequality – active people had fewer The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 109 limitations than inactive people before 2011, but this pattern reversed from this year, as active people had more limitations3. This analysis was the first that looked at the change of HI in Portugal over time. Despite focusing on one country alone, several lessons can be applied to other contexts for research and policy purposes. Firstly, HI changed considerably in Portugal over the last decade, possibly related to changes in the country’s social and economic circumstances. In 2011 Portugal went through a period of considerable uncertainty with the request of a €78 billion bailout from the EU and the International Monetary Fund, resignation of the ruling government and snap elections, and constant talks of austerity measures that would force the Portuguese to ‘tighten their belt’ in the near future (40). This might have influenced overall health and changes in its distribution, which highlights the importance of contextual determinants on HI and adds knowledge to the growing body of evidence of the impact of economic crises on HI (41). Secondly, it shows that HI, even within one country, are neither static nor determined by the same factors over time. On the contrary, contextual socioeconomic changes can have substantial impacts on HI and on what drives them. Finally, this understanding of the drivers of HI can be extremely useful to outline policies to tackle them. In our analysis, the main driver of decreasing HI was an inversion in the prevalence of limitations among active and inactive people. We hypothesized this may be a consequence of high emigration rates in the country (which led to an exit of healthier, richer people) or of different adaptation capacity among socioeconomic groups (as groups of higher social standing may be less capable of adapting to worse socioeconomic circumstances). 3.1.5. Summary These four publications paint a picture of how contextual determinants interact with individual characteristics to influence the distribution of health in Portugal and in its wider European context. The analyses and their results are summarized in table 2. Contextual social capital was found to have no impact on population health in European countries, but had an effect for a particular group of people – those with low individual levels of interpersonal trust. Welfare regime types were also associated 3 People were considered ‘active’ if they were employed (part or full time), in training or studying, or fulfilling domestic tasks; ‘inactive’ people were unemployed, retired, unfit to work or in the ‘other inactive’ category. The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 110 with the magnitude of the health impact of social mobility, with Scandinavian countries showing the smallest absolute differences. Finally, important contextual changes in Portugal over the last decade seem to have influenced health and its distribution in the country. From a different perspective, the SR sought to summarize the knowledge on HI in Portugal up to date. This analysis showed that the study of contextual determinants of HI is still uncommon in Portugal and focused on a limited number of determinants. Results of this paper must be interpreted with caution as they aggregate all the data on HI in Portugal over the last decade; as the fourth publication showed, this was a time of intense changes in the distribution of health in Portugal, so aggregating the results over this period of time may have hidden important information. Table 2. Summary of dissertation publications, determinants tested, geographic context, time period, and main findings. Publication Social Capital and Health in European Countries Social Mobility and Health in European Welfare Regimes Social Determinants of Health in Portugal Shifting Determinants of Health Inequalities in Portugal Contextual determinant Social capital Welfare regime n/a Macroeconomic context Individual determinant Social capital Social mobility n/a Socioeconomic determinants (a) Context Europe Europe Portugal Portugal Time period 2002-2012 2002-2012 2000-2014 2004-2014 Main findings Important crosssectional interaction – low trust individuals have worse health in high trust contexts Scandinavian countries had the smallest association between social mobility and health Few studies on contextual determinants of health. HI decreased in Portugal after 2010, mostly due to worse health among professionally active people (a) Five individual determinants were tested: occupation, activity, income, education, and savings. HI: Health Inequalities The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 111 3.2. Limitations The publications included in this analysis are subject to a number of limitations. Limitations that are specific to each publication have been spelled out in the results section. Some limitations are common to two or more publications. The first is that the data that that was used was cross-sectional. EU-SILC has a longitudinal component that could have been used, but panels only last four years, and this was not considered a sufficient amount of time to assess changes in health outcomes. As such, it was decided to use only cross-sectional data. This decision limited the study’s ability to determine whether health is an outcome of the determinant under study, or the other way around. However, it was tried, when appropriate, to consider both directions of the association. For example, by acknowledging both processes of health selection and of the impact of social mobility in health in the second publication. Another possible source of limitations is from the used health variables. The use of SAH and of health limitations is open to criticism, but it must also be considered that these have important value, and there is a reason why most population surveys ask these questions specifically. SAH has been shown to be a reliable measure of overall health (42), and even a good predictor of mortality (43). Arguably, SAH is a better measure of health than an ‘objective’ one, such as a diagnosed illness, as it incorporates the individual’s perception of their own health (42). Self-reported limitations in daily activities are also a reflection of the individual’s perceptions, but provide a more objective measure than SAH. This outcome has been used by other authors as an objective measure of functional limitations (44, 45) and Eurostat uses it as measure of disability (46). The two first publications must also be interpreted with caution as they consist of cross-country comparisons of subjective measures, which may be interpreted differently in each country. This is particularly important for the first publication, as interpersonal trust showed considerable variation between countries. We tried to address this issue by using countries as fixed effects, thus removing from the model time-invariant country characteristics. As a whole, the four publications may also occasionally seem to bear only a loose connection, as they focus on different geographical areas, on different determinants of The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 112 health, and are sometimes framed in ways that do not seem to be connected. Indeed, the SR of the literature was slightly different in its goals, but was considered a necessary first step, in order to aggregate all available knowledge about HI in Portugal before carrying on further research. Excluding the SR, all the publications look at contextual characteristics – be it welfare regime, social capital, or an economic crisis – and how these have an impact on the distribution of health within particular population groups. The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 113 3.3. The Effect of Context on Health and on Health Distribution As was outlined in the introduction, the study of context and how it influences health and HI has been undermined by a lack of a theoretical basis on how contextual determinants work. Both the CSDH and Diderichsen have recognised the role of context in creating HI, but only insofar as it is responsible for social stratification and for the production of policies (47, 48). While helpful for policy formulation, this is an incomplete view of how context can influence health and HI. The framework proposed here seeks to fill this gap by outlining the mechanisms by which context influences health and health distribution. It does this by describing two mechanisms: (i) changes in overall heath and (ii) changes in health distribution. This distinction is important because, just as is argued for the CSDH’s framework (48), policies aimed at improving population health do not always have a positive impact on its distribution. Hence, when seeking to influence HI, it is not enough to implement a policy to improve overall health. Rather, it is important to consider how that policy can change health distribution also. It is hoped that the framework proposed here contributes to the absence of this theoretical foundation, by outlining the mechanisms by which context influences health and health distribution, providing a basis for policy choices and empirical analyses. This framework was substantially inspired by Diderichsen’s framework, drawing on the individual pathways that the author outlines to form a basis of how context can influence those pathways. Unlike that framework, however, this one focuses on the role of context, and highlights its impact on population health, on the one hand, and on health distribution, on the other. These impacts are not mutually exclusive, as any one change in context or contextual characteristic can influence health and health distribution through multiple pathways; however, it is hoped this distinction supports a reflection on how context operates and helps fill the ‘black box’ of contextual effects on health (49). 3.3.1. First Mechanism: Changes in Overall Health The first mechanism leads to changes in the overall health of a population. Using the individual pathway of Diderichsen’s framework as a base, this mechanism can operate by changing SES, exposure to risk factors, or health status of individuals. Figure 6 shows these three effects. The core of the framework starts from social The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 114 position, which determines differential exposures; exposures lead to differential vulnerability; and finally disease or illness lead to differential consequences, which can have an impact back on social position. Figure 6. First mechanism: changes in overall health. The first effect refers to how individuals’ social position can be altered by context. For example, when a country’s finances improve, poverty tends to decline, and individual economic status of its citizens tends to improve, leading to overall improvements in health. The second effect reflects changes in the exposure to risk factors; examples of this are water fluoridation or reductions in air pollution, which can reduce the exposure to risk factors in an entire population, thus improving its overall health. The third effect refers to events that change health status, such as when an innovative treatment to a prevalent disease is discovered and made available. All these effects have the potential to improve overall population health, either directly, by affecting individuals’ health or indirectly, by affecting social position or exposure The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 121 programs (contextual characteristic) have the potential to reduce this inequality, as they provide easier access to screening to people who otherwise might not use it (69). 3.3.4 Summary The framework proposed here outlines how context influences health and health distribution. It is proposed that context operates through two mechanisms, one that changes overall population health (though effects on social position, exposure to risk factors, and disease or illness) and another that changes health distribution (through effects on social stratification, differential exposure, differential vulnerability, and differential consequences). This framework seeks to fill a gap in the research literature, by which the pathways between the context and health and HI have not been outlined before. It takes a step forward from other conceptual frameworks, as it acknowledge the role of context on several different steps between social position and health outcomes. In this framework, context does more than just stratify individuals to their social position. The framework will hopefully be used as a basis for future policy and empirical analyses, helping clarify the mechanisms by which context influences health and its distribution. The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 122 3.4. Application of the Conceptual Framework The framework described in the previous section (figure 8) outlines the mechanisms by which contextual determinants can influence health and health distribution. This framework seeks to summarize how this influence operates, and can potentially be used as a basis for policy and empirical analyses. In this chapter, the framework is applied to three research papers from this dissertation, in order to illustrate how it can be used in the interpretation of evidence. It is not applied to one publication – the SR – as this did not look at how one contextual determinant interacts with individual characteristics but rather set out the background for the last piece of investigation on Portugal (see table 2). 3.4.1. Social Capital and Health in European Countries This analysis showed a complex effect of national-level social capital on individual health, making it an interesting case study for the application of the framework. The framework focuses on how the contextual determinant (contextual social capital) can have an impact on population health and health distribution, operating through an individual characteristic (individual social capital). The key results were: • Contextual social capital was not associated with individual health, and • High contextual social capital was associated with worse health in low trust individuals and better health in high trust individuals. These results show that contextual social capital had no impact on overall population health. Within the conceptual framework (figure 8), this means that social capital will not operate through the full arrows (the first mechanism – changes in overall population health). On the other hand, contextual social capital had a differential effect on different social groups, leading to changes in health distribution (dashed arrows, second mechanism – changes in health distribution). This is probably a reflection of how contextual social capital is not a resource enjoyed equally by all individuals – when high trust individuals are the majority, the social capital they produce between them is not shared with the minority, low trust individuals. Moreover, not only is this resource out of their reach, but discrimination from the majority and dissemination of ‘bad social capital’ (such as reinforcing social norms that are harmful to health) might further damage a group that is already vulnerable. This differential effect can thus be a The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 123 consequence of processes of network closure, discrimination, and dissemination of ‘bad social capital’. All these processes in essence reflect a differential exposure to social capital: high trust individuals have access to contextual social capital, while low trust individuals do not; furthermore, low trust individuals may also be exposed to ‘bad’ social capital, further damaging their health. Figure 9. Effect of contextual social capital on health distribution: results of the first publication (Campos-Matos I, Subramanian SV, Kawachi I. The ‘dark side’ of social capital: trust and self-rated health in European countries. European Journal of Public Health. 2016;26(1):90-95). It is important to note that applying the framework to only one analysis does not exhaust all possible mechanisms by which one contextual determinant operates. Contextual social capital can impact HI, as people are not equally exposed to its effects, and it can also impact overall population health – as some authors argue and indeed have shown (4, 70). This framework is helpful to clearly identify the mechanisms under study in an empirical analysis, and further suggests other mechanisms that may also exist and not have been detected in that particular analysis. 3.4.2. Social Mobility and Health in European Welfare Regimes This analysis focused on how the association between social mobility and health can differ between European welfare regimes. Importantly, it provided some evidence that The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 124 health selection does not seem to be an important determinant of HI in countries with high levels of social mobility. To apply the framework to this analysis, the focus will be on how the contextual determinant (welfare regime) can have an impact on health and health distribution through the individual determinant (social mobility). The key messages regarding the impact of welfare regime through social mobility were: • There were significant differences in overall health between different welfare regime types – Scandinavian countries showed the best and former USSR countries the worse results; and • Individuals who were upwardly mobile showed better health than those who were socially stable, but upward mobility had a small impact in Scandinavian countries, compared to a large impact in former USSR countries. First, it is clear that welfare regime is associated with differences in overall population health. There are many ways in which certain welfare regimes can have an impact on health – it can be related to economic development, health services provision, and cultural aspects, among others. One possible way is through social mobility. Political choices, such as how to provide education or how to redistribute wealth, have the potential to break the inter-generational transmission of social disadvantage and improve the social position of many, regardless of their parents’ social standing. Thus, the extent of social mobility varies substantially between welfare regimes (71) and can potentially lead to better socioeconomic circumstances across the population. This is evident by the observation that countries where social mobility is highest also tend to have higher overall educational levels (72). It is by this effect on social mobility that welfare regimes, through improvements in people’s social position, can lead to a better (or worse) level of overall population health (figure 10). Second, upward mobility was associated with better health in all countries, but the difference was considerably bigger in former USSR countries and smallest in Scandinavian countries. This suggests that Scandinavian countries are more effective at separating social mobility from health. This means that when individuals climb up (down) the social ladder, they are more likely to have better (worse) health, especially if they live in former USSR countries. This can reflect a mechanism of social stratification, as welfare regimes determine how individuals are placed in a society and, consequently, their health. From a health selection perspective, it may also mean The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 125 that individuals who are ill (healthy) are much more likely to fall (climb) in the social ladder in former USSR countries, when compared to Scandinavian countries. This second possibility can reflect a mechanism of differential consequences, as a person’s health leads to changes in their social position. Figure 10. Effects of welfare regimes and social mobility on health and health distribution: results of the second publication (Campos-Matos I, Kawachi I. Social mobility and health in European countries: does welfare regime type matter? Social Science and Medicine. 2015;142:241-248). The results of this analysis reflect social mobility’s effects on health and HI, since more generous welfare regimes not only show better overall health, but also fewer inequalities between social mobility groups. Thus, welfare regime, through its effects on social mobility, can impact both the health and the health distribution of a population (figure 10). This analysis, of course, does not include other effects that welfare regimes can have both on health and HI, as it focuses exclusively on the effects that are mediated by social mobility. 3.4.3. Shifting Determinants of Health Inequalities in Portugal The focus of this analysis was on how HI have changed in Portugal over the last decade. Although we did not test the effect of any contextual determinant, we interpreted the results considering the substantial economic changes that were The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 126 happening in the country at the time. For the application of the framework, the following key points summarize the most relevant results of this analysis: • Overall population health deteriorated in Portugal between 2010 and 2011; • This deterioration was steeper in the richest tercile, thus decreasing inequalities between income terciles; and • The main driver of this decrease was the change between active and inactive groups: while limitations were more prevalent in inactive people until 2010, they became more prevalent in active people after this year. The overall deterioration in population health in Portugal between 2010 and 2011 is likely to be related to overall economic changes happening at the time. These events, such as implementation of highly publicized austerity measures, can operate through a stress-determined pathway, as a climate of uncertainty can lead to stress and, consequently, poorer health, leading to worse health outcomes across the population. In the framework, this can be interpreted as an effect on exposure to a risk factor: economic changes are the contextual determinant that lead to an increase in the exposure to a risk factor – stress – thus having an effect on the whole population’s health (figure 11). However, not everyone reacted the same way to this uncertainty. In fact, richer people (the richest tercile) seemed to suffer the greatest hit. This was greatly mediated by the fact that active people – who also tended to be richer – had a higher prevalence of limitations after 2010 than inactive people. Despite not having formally tested this, we hypothesized that two mechanisms might be behind this change. First, people from higher socioeconomic classes (who tend to have higher incomes) may not be as used to dealing with uncertainty as people from lower socioeconomic classes are. To them, the prospect of uncertain times ahead could have led to more intense stress reactions and to an inability to deal with practical day-today problems on a more restricted budget. This first mechanism can be identified in the framework as differential vulnerability, as all groups were exposed to uncertainty and stress, but – perhaps counter intuitively – high SES people were more vulnerable, at least during a certain period of the time. Second, considering the high emigration rates Portugal was going through at the time (73), the group of active people who was ‘left behind’ might have had disproportionately high rates of limitations, as migration is known to be a selective process by which healthier people tend to migrate more (74). This second mechanism is a consequence of changes in the composition of the The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 127 population. This can be seen as a change in social stratification, as changes in the economic context led to changes in how wealth and power were distributed to different social positions. Figure 11. Effects of macroeconomic changes on health and health distribution: results of the fourth publication (Campos-Matos I, Russo G, Gonçalves L. (accepted for publication). Shifting determinants of health inequalities in unstable times: Portugal as a case study. Accepted for publication in the European Journal of Public Health in May 2017. DOI: https://dx.doi.org/10.1093/eurpub/ckx080). Overall, this analysis suggests that economic and social changes in a country can lead to changes in health and health distribution through three mechanisms: social stratification, exposure to risk factors, and differential vulnerability to those risk factors (figure 11). Once again, this does not provide an exhaustive description of how economic crises impact health and health distribution, but suggests some of the mechanisms that may operate and frames the findings in a larger context. 3.4.4. Summary The framework proposed here aims to outline the mechanisms that connect contextual characteristics to population health and health distribution. It was applied to the analysis of three contextual determinants: social capital, welfare regime, and macroeconomic changes. Each of these determinants influenced health outcomes and their distribution through various mechanisms: differential exposure (social capital); The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 128 social stratification, social position, and differential consequences (welfare regimes); and social stratification, exposure, and differential vulnerability (economic crisis). The framework proved a useful tool to frame these publications, in which contextual determinants were explored. In particular, it acknowledged the importance of context in several steps in order to influence both health and HI. In future research, it can provide a structure to facilitate a reflection about the theoretical basis that underscores analyses, thus strengthening them and the arguments they propose. The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 129 3.5. Contribution to Policy and Research 3.5.1. Regarding Health Inequalities This research has presented a new framework on how context can shape health and health distribution. This framework is built on a framework proposed by Diderichsen et al. (figure 3) (52), and incorporates a few important differences from the original framework. Firstly, it explicitly focuses on context. For Diderichsen, context was responsible for social stratification and policies. In the framework put forward here, it is proposed that context is seen is a broader perspective, encompassing also physical and other social elements (besides policies). It is also proposed that context not only contributes by creating a system of social stratification and policies, but also by influencing all the other steps of the pathway between social position and health outcomes. Furthermore, this framework highlights that context can influence health and health distribution, and it does so by different mechanisms. Another important distinction from Diderichsen’s work is that in the former framework differential vulnerability is the accumulation of harmful risk factors in a particular population group. In applying the new framework to the analyses, it was found that differential vulnerability does not necessarily mean that the same individuals are always vulnerable. Similarly, differential exposures and consequences do not necessarily mean that the poorest or least educated will always be more exposed to risk factors or suffer greatly the consequences of ill-health. In our work, it is proposed to redefine the term ‘differential’ to mean only that effects are different in different socioeconomic groups, without suggesting a direction of the effect. Hopefully, this framework can be used in future research projects to collect evidence of the effect of a contextual determinant. For example, to perform a review of how economic growth can impact on health and health distribution, the collected evidence can be organized using this framework. This framework could also help researchers outline a clear theoretical basis for their work, thus building a stronger evidence base around the topic of their research. This is particularly important considering the lack of a clear theory about the mechanisms by which context operates has been one of the factors hampering the study of contextual influences on health (75). Regarding its contribution to policy making, as it builds on and further the current theories on contextual effects on health, the framework may also help policy makers The relationship between context and health inequalities Chapter 3. Discussion and Conclusions 130 understand the options available to tackle specific aspects of HI. Particularly, it could prove a useful instrument for equality impact assessments – the analysis of policies that tries to ensure they do not discriminate against any vulnerable group (76). Besides being a practical instrument for researchers and policy makers, the framework will hopefully contribute to the rising trend of a different thinking about HI. Namely, it emphasizes how HI are not a product of individual characteristics alone, but a consequence of the interaction between context and individuals. This can have important implications on how HI are tackled – from a focus on individual behaviour change, policy makers should also think about enabling and disabling characteristics that contexts can provide to influence those behaviour changes. Another contribution of the present research is the concurrent analysis of several individual SES characteristics as an opportunity to explore the mechanisms behind HI. This type of analysis is not common, but proved very productive. In this analysis, this allowed for the exploration of what might have been behind the decrease in HI in Portugal after 2010, thus laying groundwork for further research on the topic. This method provided interesting levels of analysis, and could be reproduced in future analyses in other contexts. 3.5.2. Regarding Europe This research also provided some contribution to the understanding of HI in European countries. The persistence of HI in Europe has been called a ‘paradox’ and social mobility has been put forward as one of a few possible explanations (10). The present findings on social mobility in different welfare regimes in Europe show that this is highly unlikely, or countries with high social mobility would show at least as high inequalities in health between mobility groups as countries with low mobility, which was not the case. This can help move forward the exploration of the ‘paradox’, as other explanations are now more likely and should be further investigated. In terms of policy-making, these findings suggest that it is possible to mitigate the association between social mobility and health. Although all welfare regimes showed some kind of association between social mobility and SAH, some were significantly smaller, suggesting that it is possible to separate the two more efficiently, and that the answer lies in the differences between the welfare regimes.