Modelling Mortality from Self-Reported Health and Disability Data: A Multi-Scale Report from Ireland, 2011-2016

Revista de Estudios Andaluces (REA) e-ISSN: 2340-2776. REA núm. 35 (2018). http://dx.doi.org/10.12795/rea.2018.i35 Modelling Mortality from Self-Reported Health and Disability Data: A Multi-Scale Report from Ireland, 2011-2016 Modelos de Mortalidad a Partir de Datos de Salud y Discapacidad Autopercibidos: Un Informe sobre Irlanda a Distintas Escalas, 2011-2016 Ronan Foley Maynooth University, Ireland [email protected] Formato de cita / Citation: Foley, Ronan (2018). Modelling Mortality from Self- Reported Health and Disability Data: A Multi-Scale Report from Ireland, 2011-2016. Revista de Estudios Andaluces, núm. 35, 179-192. http://dx.doi.org/10.12795/rea.2018.i35.07 Enlace artículo/to link to this article: http://dx.doi.org/10.12795/rea.2018.i35.07 Esta obra se distribuye con la licencia Creative Commons Reconocimiento-NoComercial- SinObraDerivada 4.0 Internacional http://editorial.us.es/es/revista-de-estudios-andaluces https://revistascientificas.us.es/index.php/REA Esta obra se distribuye con la licencia Creative Commons Reconocimiento-NoComercial- SinObraDerivada 4.0 Internacional 179 Revista de Estudios Andaluces, núm. 35 (2018) pp. 179-192 Modelling Mortality from Self-Reported Health and Disability Data: A Multi-Scale Report from Ireland, 2011-2016 Modelos de Mortalidad a Partir de Datos de Salud y Discapacidad Autopercibidos: Un Informe sobre Irlanda a Distintas Escalas 2011-2016 Ronan Foley Maynooth University, Ireland [email protected] Recibido: 11 de diciembre, 2017 Revisado: 25 de enero, 2018 Aceptado: 26 de enero, 2018 Abstract Relationships between mortality and morbidity are long established within geo-spatial modelling and GIS-based analyses. While there has always been a strong associational relationship between the two measures, this has been less tested at an aggregate areal scale than one might expect. From a geographical perspective this has been in part because access to data at meaningful spatial scales, especially for mortality, can be difficult. This paper presents newly collected data from Ireland on mortality and maps crosssectional associations with self-reported health and limiting long-term illness and disability conditions (LLTC) for the first time at an intermediate level geography. Data is also available for two different time-periods from administrative records and from five-yearly censuses. Mortality data was collected at a newly-created intermediate level geography (IA, n=410, average population=10,900) while the self-reported health/LLTC data was collected at a long-standing administrative scale (ED, n=3409, average population=1,350). Given there was a nested relationship between the two scales, redistricting techniques were used in GIS to enable direct comparisons. Mortality data was available for all deaths (SMR) and premature death (ASR), as well as for four different causes of death. Self-reported health was mapped in three ways; as a combined rate for poor health status; as a weighted health score and as a combined rate for LLTC. The associations were derived from correlation/regression modelling at the smaller IA scale. The results identified that the associations were statistically significant and of mixed magnitude, but had relatively low r-squared values. The associations were strongest for premature (under 75) mortality, while additional Revista de Estudios Andaluces, núm. 35 (2018) pp. 179-192. e-ISSN: 2340-2776 http://dx.doi.org/10.12795/rea.2018.i35.07 Esta obra se distribuye con la licencia Creative Commons Reconocimiento-NoComercial- SinObraDerivada 4.0 Internacional 180 Modelling mortality from self-reported health and disability data: a multi-scale report from Ireland, 2011-2016 correlations for cause of death were lower again. From this, we concluded that the self-reported health/LLTC statistics, while of some potential explanatory value, were not especially useful as predictive variable. Suggestions for improvement would be to weight the self-reported health data by age and additionally to take into account deprivation as a second explanatory factor operating within cross-sectional work. Finally, modelling at different spatial scales might act as a useful guide for comparative analysis in Andalucía and other regions of Spain, where spatial scales may be similar in terms of size and scale. Keywords: mortality, morbidity, spatial patterns, modelling, Ireland. Resumen Desde hace mucho tiempo, se vienen realizando análisis geoespaciales utilizando herramientas SIG para establecer la relación entre mortalidad y morbilidad. A pesar que siempre ha existido una estrecha relación entre estas dos dimensiones, esta ha sido escasamente evaluada a escala agregada. Desde una perspectiva geográfica, esto no ha sido posible en parte debido a que el acceso a datos, tales como mortalidad, ha sido difícil para determinadas escalas espaciales. Este artículo presenta por primera vez, a un nivel geográfico intermedio, información y mapas recientes sobre mortalidad en Irlanda, asociándola de forma transversal con la salud auto-percibida, la limitación a causa de una larga enfermedad y el nivel de la discapacidad (LLTC). Los datos utilizados provienen de dos fuentes, registros administrativos y censos quinquenales y pertenecen a dos periodos diferentes temporales. Los datos de mortalidad se recopilaron en base a un nivel geográfico intermedio (IA, n = 410, población media = 10,900), mientras que los datos de salud auto-percibida/LLTC se recopilaron en base a una antigua escala administrativa (DE, n = 3409, población media = 1,350). Ante la falta de homogeneidad entre las dos escalas, se utilizaron técnicas de reordenación mediante SIG para realizar comparaciones directas. Los datos de mortalidad estaban disponibles para todas las muertes (SMR) y las muertes prematuras (ASR), para cuatro causas diferentes de mortalidad. La salud autopercibida fue cartografiada de tres formas distintas: como una tasa combinada para el mal estado de la salud; como una puntuación ponderada de la salud y como una tasa combinada para LLTC. Las relaciones fueron derivadas desde un modelo de correlación/ regresión a una escala más pequeña IA. Los resultados han puesto de manifiesto que las relaciones eran estadísticamente significativas y de magnitud mixta, pero tenían valores r-cuadrados relativamente bajos. La relación fue más significativa para las muertes prematuras (menor de 75), mientras que las correlaciones relacionadas con la causa de muerte fueron bajas. A partir de esto, se llegó a la conclusión de que las estadísticas respecto a la salud auto percibida /LLTC, aunque poseen algún valor explicativo, no se pueden utilizar como variables predictores. Para mejorar los resultados en la realización de estudios transversales, se sugiere que la información sobre la salud auto-percibida se divida por grupos de edad y, además, se Revista de Estudios Andaluces, núm. 35 (2018) pp. 179-192. e-ISSN: 2340-2776 http://dx.doi.org/10.12795/rea.2018.i35.07 Ronan Foley Revista de Estudios Andaluces, núm. 35 (2018) pp. 179-192. e-ISSN: 2340-2776 http://dx.doi.org/10.12795/rea.2018.i35.07 Esta obra se distribuye con la licencia Creative Commons Reconocimiento-NoComercial- SinObraDerivad a 4.0 Internacional 181 tengan en cuenta la deprivación como un segundo factor explicativo. Finalmente, el modelado a diferentes escalas espaciales podría servir como una guía útil para realizar análisis comparativos en Andalucía y otras regiones de España, donde las escalas espaciales pueden ser similares en términos de tamaño y escala. Palabras clave: mortalidad, morbilidad, patrones espaciales, diseñar, Irlanda. Esta obra se distribuye con la licencia Creative Commons Reconocimiento-NoComercial- SinObraDerivada 4.0 Internacional 182 Modelling mortality from self-reported health and disability data: a multi-scale report from Ireland, 2011-2016 1. INTRODUCTION Relationships between mortality and morbidity are well established but are always open to further scrutiny within geo-spatial modelling and GIS-based analyses (Gatrell and Elliot, 2009). Typically, places that have high levels of morbidity tend to produce high levels of mortality, with the most common causal factors associated with low socio-economic status (SES), high levels of poverty and deprivation and additional spatial justice factors such as unhealthy and polluted environments (Boyle et. al., 2004; Brown et. al, 2017). In addition recent research on the developing impact of health behaviours in terms of physical activity, diet and growing obesity have also had a spatial component (Carroll-Scott et. al., 2013). While there has always been a strong associational relationship between the two measures, this has been less tested at an aggregate areal scale than one might expect (McIntyre, Ellaway and Cummins, 2002). This is in part based on wider theoretical discussions within social and health geographies about the relative importance of context (the place) or composition (the people in a place) in explaining spatial variations (Pickett and Wilkinson, 2015). Which of these factors is more important is still open to debate, but for those people who suggest that context matters more than composition, the focus of their research lies on identifying a fine enough scale of analysis to get as close to neighbourhood level as possible. In addition sensitivity to data protection means that individual level data cannot be used (Rigby et. al, 2017). From a geographical perspective this has been in part because access to data at meaningful spatial scales, especially for mortality, can be difficult (Foley and Kavanagh, 2016). This paper presents newly collected data from Ireland on mortality and, as a proxy for morbidity. self-reported health and limiting long term conditions (illness and disability, or LLTC) status. The aim of the piece is to identify whether any strong associations emerge to prove the relative importance of context. At the heart of the research was a relatively straight-forward piece of geo-spatial modelling. This consisted of the mapping out and comparing of data on the geographical distribution of mortality, captured at a new intermediate scale of reporting, with the geography of proxy morbidity, as measured via Census data on selfreported health. The novelty of the paper is that this is the first time testing out associations of mortality and morbidity data in Ireland over two time periods, as the relevant data has not been available before now. Up to the last decade, getting information on the geography of mortality in Ireland had been more accurate in 1900 than in 2000 (Houghton, 2005). In general, mortality is formally reported through official routes, which includes information on specific causes of death. Other specific cause of death data, such as for cancer, are collected by separate agencies in Ireland such as the Cancer Registry, who produce regular maps of aggregated deaths by specific cancers for specific clusters of years and make these available publicly (https://www.ncri.ie/data/maps). In choosing a mortality over a morbidity route, the research was able to explore a number of key spatial issues, with a primary aim to establish statistical results on the associations between the two indicators at a Revista de Estudios Andaluces, núm. 35 (2018) pp. 179-192. e-ISSN: 2340-2776 http://dx.doi.org/10.12795/rea.2018.i35.07 Esta obra se distribuye con la licencia Creative Commons Reconocimiento-NoComercial- SinObraDerivada 4.0 Internacional 183 Ronan Foley standardised areal scale. A further objective was to provide some additional nuanced results by cause of death to see if there were stronger relationships for these specific indicators. While a number of additional constraints were identified, at the heart of the work was the identification of the value of aggregated mortality data and its capacity to be modelled at an area-level spatial scale, notwithstanding some known modifiable areal unit problems (Schuurman et. al., 2007; Rigby et. al., 2017). Within Ireland, there are a number of administrative geographies against which data is collected, primarily via the 5-yearly Census. The most recent census years have been 2016, 2011, 2006 and 2002 (the delayed 2001 census). These shift and change over time but in 2011, the core collection date for this analysis, there were 34 Local Authorities (LA), the traditional cities and counties of Ireland with an average population of around 130,000 and a range from 32,000 to 527,000. Traditionally the main lower level geography had been the electoral division (ED, n=3409, average population= 1,350, range 75 to 36,000). This administrative unit has been in place since the mid-19th century, though its' significantly skewed population range makes it potentially problematic. In 2011, on the back of a series of national consultation exercises, a new neighbourhood level Small Area (SA) geography was introduced which had 18,488 units, a mean population of 250 and a narrow range of 70 to 1,940. In effect, especially in urban settings, this was a housing estate level geography. In developing this analysis, mortality data (a non-Census indicator) was collected and mapped at a newly created intermediate level geography (IA, n=410, average population=10,900) while the self-reported health and LLTC caring data was collected at the above three administrative scales, LA, ED and SA. The new IA level geography was developed by the team at the NCG, Maynooth University, using an aggregation programme, AZTool (Cockings et. al., 2011) to amalgamate SAs to a target population size of 10,000 based on a number of criteria around number thresholds, shape, social homogeneity and containment constraints (see Rigby et. al., 2017 for fuller details). 2.DATA COLLATION AND METHODS In calculating the mortality data, Jan Rigby and colleagues were able to access individual level death records annually for the period 2006 to 2011 and map this to the new aggregated IA geography described above. Mortality data, from the Irish Central Statistics Office (CSO) and the General Record Office (GRO). There were a number of problems with the raw data that were tidied up at data preparation stage. Firstly, the data was cleaned as, rather surprisingly, several people were recorded as dying more than once. Secondly, the home address of the individual was geo-referenced, but the success rate for this process was relatively low (on average around 80%). This was due to a mix of poor recording but also the presence of a considerable number of duplicate addresses, especially in rural areas. With the arrival of new individual household postcodes in 2016, this process should be more speedy and accurate in future years. Finally, small number problems, even when aggregated to IA scale, meant that data Revista de Estudios Andaluces, núm. 35 (2018) pp. 179-192. e-ISSN: 2340-2776 http://dx.doi.org/10.12795/rea.2018.i35.07 Modelling mortality from self-reported health and disability data: a multi-scale report from Ireland, 2011-2016 Revista de Estudios Andaluces, núm. 35 (2018) pp. 179-192. e-ISSN: 2340-2776 http://dx.doi.org/10.12795/rea.2018.i35.07 Esta obra se distribuye con la licencia Creative Commons Reconocimiento-NoComercial- SinObraDerivada 4.0 Internacional 184 had to be aggregated across a five-year time period, which meant that longitudinal analysis was affected. Data was developed for all deaths (SMR) and premature deaths (ASR), as well as for four different causes of death, essentially the main ‘killers’ in Ireland, cancer, heart, cardiovascular and respiratory (Rigby et. al., 2017). Figure 1 below shows one representative map of this data based on the Age-Standardised Mortality rates for Under 75s – effectively premature mortality – for 2011 at the IA scale. The map shows a mixed pattern in rural areas, with lower levels in suburban areas, while the inset map for the Greater Dublin Area shows a sort of northeast to south-west divide, with the higher levels associated with a generally more deprived geography north of that line. Figure 1. Age-Standardised Mortality (Under 75 years) in Ireland at IA Scale, 2011. Esta obra se distribuye con la licencia Creative Commons Reconocimiento-NoComercial- SinObraDerivada 4.0 Internacional 185 Ronan Foley The other core data set used to indicate morbidity was taken from the 2011 Census, and used one newly revised variable, self-reported health, and from that data, a derived broad index, the Kavanagh-Foley Index of Wellbeing (KFIW) along with data on LLTC. For the purposes of this paper, the 2011 SA data was not used, but the other data was available at ED level for both indicators. The health question specifically asked for respondents to report on their perceived level of health. In 2011, respondents were asked to select from one of five responses - in effect a Likert scale - in addition to a not-stated class in the final data. The answers were listed as; very good –good – fair – bad – very bad. The amounts and levels of the former two classes were very high (around 85%), while those of the latter two classes were very low (around 2%), compared to other parts of the British Isles (Foley and Kavanagh, 2016). This was accounted for in part by the generally much younger average population age of the Republic of Ireland. Figure 2 below shows the ED level geography for an aggregated percentage of poor health (those who classified themselves as in fair, bad or very bad health) for 2011. The KFIW was calculated as follows; a weighting of 1 was added to the percentage responding ‘very good’ health, a weighting of 2 to the percentage responding ‘good’ and so on, down to a weighting of 5 to those listing their health as ‘very poor’. The effect was to produce a score with a mean of 145.5 with a fairly narrow range (128 to 171) while the score was effectively an inverse health score, i.e. those EDs with the highest score were the areas with the highest levels of poor health. The final LLTC variable, based on specific census questions 16 and 17, was an amalgamated count of all people who recorded a range of disabilities and illness conditions including hearing and visual impairments, mental and physical health problems and chronic illness, as well as the extent to which these conditions impaired them in carrying out basic daily tasks. For statistical purposes, these were all aggregated into a count of individuals who had recorded any of these conditions and the resultant count has been used in Ireland as a proxy for limiting-long term illness. This data for 2016 averaged around 13% and was a significant increase from 9% recorded in 2011. It is reasonable to surmise that this increased was down to a shift in the percentage of the population living to older ages, but within which identifiable health needs were visibly increasing. A number of national organisations, including the Department of Health, identify this pattern, but also the value of this disability/illness variable as a good predictor for service demand (DoH, 2017). The primary data preparation task was to take the health, KFIW and LLTC data – broadly acting as morbidity measures - recorded at ED scale and match it to the IA scale mortality data for the measurement of associations. Given there was a nested relationship between the two scales, GIS redistricting techniques were used to enable direct comparisons. This was a two-step process whereby a union was created which recoded the smaller level geographical codes into the higher-level geographies to which they were spatially associated. What it did effectively was to assign the IA code Revista de Estudios Andaluces, núm. 35 (2018) pp. 179-192. e-ISSN: 2340-2776 http://dx.doi.org/10.12795/rea.2018.i35.07 Modelling mortality from self-reported health and disability data: a multi-scale report from Ireland, 2011-2016 Esta obra se distribuye con la licencia Creative Commons Reconocimiento-NoComercial- SinObraDerivada 4.0 Internacional 186 Figure 2. Geography of Self-Reported ‘Poor’ Health at ED Scale, 2011. (into a new field) against each of the individual EDs that fell inside the associated IA boundary. These were non-overlapping in all cases, as the original IAs had been developed from clusters of SAs (Cockings et. al., 2011; Rigby et. al., 2017). The second step was then to run a dissolve (redistrict) command within the GIS, using the newly created IA field in the ED layer as a look-up, to match the lower-level geography to the upper-level one and to calculate appropriate aggregate statistics. We then had a set of comparable data for mortality and morbidity at the same scale for two different time periods. We were not able to create a full set as the mortality data was an aggregation across the time period from 2006 to 2011, but we did use that aggregated data for modelling at either end of that period, to at least begin to explore the relationships Revista de Estudios Andaluces, núm. 35 (2018) pp. 179-192. e-ISSN: 2340-2776 http://dx.doi.org/10.12795/rea.2018.i35.07