The DEP-6D, a new preference-based measure to assess health states of dependency

1 The DEP-6D, a new preference-based measure to assess health states of dependency Eva Rodríguez Míguez1 José María Abellán Perpiñan2 José Carlos Álvarez Villamarín1 José Manuel González Martínez1 Antonio Rodríguez Sampayo3 1 Department of Applied Economics, University of Vigo. 2 Department of Applied Economics, University of Murcia. Spain 3 Department of Economics. University of Santiago de Compostela. Spain Abstract In medical literature there are numerous multidimensional scales to measure health states for dependence in activities of daily living. However, these scales are not preference-based and are not able to yield QALYs. On the contrary, the generic preference-based measures are not sensitive enough to measure changes in dependence states. The objective of this paper is to propose a new dependency health state classification system, called DEP-6D, and to estimate its value set in such a way that it can be used in QALY calculations. DEP-6D states are described as a combination of 6 attributes (eat, incontinence, personal care, mobility, housework and cognition problems), with 3-4 levels each. A sample of 312 Spanish citizens was surveyed in 2011 to estimate the DEP-6D preference-scoring algorithm. Each respondent valued six out of the 24 states using time trade-off questions. After excluding those respondents who made two or more inconsistencies (6% out of the sample), each state was valued between 66 and 77 times. The responses present a high internal and external consistency. A random effect model accounting for main effects was the preferred model to estimate the scoring algorithm. The DEP-6D describes, in general, more severe problems than those usually described by means of generic preference-based measures. The minimum score predicted by the DEP-6D algorithm is -0.84, which is considerably lower than the minimum value predicted by the EQ-5D and SF-6D algorithms. The DEP-6D value set is based on community preferences. Therefore it is consistent with the so-called ‘societal perspective’. Moreover, DEP-6D preference weights can be used in QALY calculations and cost-utility analysis. Keywords: Spain, dependency, preference-based measures, QALY, DEP-6D, time trade-off. Acknowledgements: Financial support from Spanish Ministry of Science and Innovation (ECO201125661), Regional government of Galicia (10SEC300038PR and ECOBAS [AGRUP2015/08]), and Caixa Galicia Foundation is gratefully acknowledged. The funding agreement ensured the authors’ independence in designing the study, interpreting the data, writing, and publishing the report. We are also very grateful to researchers Lina Sofia de Matos Lourenço-Gomes (University of Trás-os-Montes and Alto Douro) and María Loureiro García (University of Santiago de Compostela) for their help on the experimental design. 2 1. Introduction The aging population represents one of the most important challenges developed countries must face in the upcoming years. The portion of the population aged over 65 accounted for 15.4% of the OECD average in 2010, but this percentage is likely to nearly double in 2050 (OECD, 2013). One of the most obvious consequences of an aging population is the increasing number of people who will require assistance to carry out their activities of daily living (ADLs). This is one of the greatest concerns of the adult population. It is even more important for them than the loss of their very own health (Finlayson, 2004; Quine and Morrell, 2007). This process involves increasing the burden of care within the family (informal care) and will pose a great challenge to the sustainment of the public purse —if in 2006-2010 the public expenditure on long-term care services in the OECD represented 0.8% of GDP, it is estimated to involve between 1.6% and 2.1% of GDP in 2060 (De La Maisonneuve and Martins, 2013). In any case, forecasting how the prevalence of ADL dependence will evolve is subject to great uncertainty because many factors are likely to affect its future trend. In addition to lifestyle changes that can modify risk factors, new technologies are continuously emerging. They will clearly have an impact on the levels of dependency because they either prevent or delay the onset of dependency or reduce its severity once it actually happens (preventing stroke, delaying both the onset of disease and the progression of Alzheimer’s disease, rehabilitation, technical aids, education programs to prevent falls, etc.). Within this context, it is useful to have instruments capable of measuring the changes in the severity of dependence fairly accurately to both plan social/health services, in general, and evaluate programs aiming to prevent and/or delay dependency states, in particular. Multiple scales are designed to measure the loss of independence in the ADLs within a clinical context 3 (McDowel, 2006; Kane and Kane, 2002), such as the Katz scale (Katz et al., 1963), Barthel Index (Mahoney and Barthel, 1965) or Lawton and Brody scale (Lawton and Brody, 1969). However, despite its widespread use, these instruments are not appropriate for economic evaluations because they are not able to yield quality-adjusted life-years (QALYs), the outcome measure recommended by leading health technology assessment agencies. To estimate the number of QALYs gained from an intervention, life years are weighted by preference weights (or utilities), where zero indicates death and one good health. A common way of assigning utilities to health-related quality of life (HRQoL) states is to use one of the existing preference-based generic instruments. Well-known examples are EQ-5D (Dolan, 1997), SF-6D (Brazier et al., 2002, 2004) and Health Utility Index (Feeney et al., 2002). However, one disadvantage of these instruments is that their health state descriptive system are not sensitive enough for some medical conditions (Brazier et al., 1999), and the effectiveness of interventions may be undervalued. Donaldson et al. (1988) and Chilsholm et al. (1997) conclude that generic measures are, as compared to condition specific measures, less sensitive to changes in older adult health status. In particular, these generic measures were not designed to measure ADL dependency and although they could be used for this purpose (for instance EQ-5D does this to some extent on the three functioning dimensions), empirical evidence suggests these instruments may not be sensitive enough to detect significant changes in the level of ADL dependency. On the one hand, different studies found a poor relation between the EQ-5D and the Barthel index (Kaambwa et al., 2013; Van Exel et al., 2004; Hickson and Frost, 2004), “underscoring the fact that these outcome measures were designed to capture different aspects of health status” (Kaambwa et al., 2013). For instance, the loss of feeding ability involves an important change at an individual and family level, yet they would rarely produce a change in any of preference-based generic instruments 4 commonly used. On the other hand, the loss of physical capacity is not the only dimension of ADL dependence. Decisional dependence is yet another dimension frequently pointed out in the literature (O’Shea et al., 2007). Cognitive limitations can produce loss of capacity for autonomous decision-making, which makes a person reliant on others regardless of whether or not they have the physical capacity to do the activities. This loss of decisional independence is not included in the generic instruments usually used for economic evaluation. The lower sensitivity of these instruments in detecting changes in the physical or decisional dependence could explain why groups of patients requiring very different hours of help can obtain similar scores in HRQoL generic instruments (Sandberg et al., 2015). Hence, it would be desirable to have a richer descriptive system to optimally evaluate those health care programs that specifically aim to prolong or enhance independence. This is why some researchers (Goldstein et al., 2002; Bravata et al., 2005; Sims et al., 2008) have elicited preference weights for various ADLs dependence states based on the combination of the 6 ADLs included in the Katz scale plus an additional ADL, walking. However, these authors do not use an experimental design (they only obtain the weights of specific states) that allows them to estimate a scoring algorithm capable of predicting all the possible health states for dependence in ADLs. Moreover, the severity of the dependence in each of the 7 ADLs is not graded (it only considers whether or not help is required to perform each ADL). Lastly, preferences were elicited from a convenience sample (older adult members of the Kaiser Permanente Medical Care Program of Northern California) rather than from the general population. Other authors (Ryan et al., 2006; Coast et al., 2008) have developed instruments related to dependence states for the evaluation of health care and social services interventions for older 5 people. In the first case, domains and levels of their instrument (OPUS) were designed to reflect whether needs relevant to social services clients are met (e.g. whether their home is clean and comfortable) rather than capture changes in functional status (e.g. whether dependency lowers with rehabilitation) as, indeed, is our aim. The ICECAP capability index for older people developed by Coast et al. (2008) does not assess preferences but capabilities, and its scores are anchored on an “absence of capability”-“full capability” scale. The instrument introduced in our paper, on the contrary, provides classical QALYs anchored on a death-full health scale as those commonly used in cost-utility analysis. Therefore, both approaches are complementary rather than mutually exclusive This paper aims to propose a new dependency health state classification system, called DEP6D, and estimate its value set so it can be used in QALY calculations. This instrument could be included within the domain-specific measurement scales where the dominance to be evaluated is the level of dependence in the ADL. Although there are different methods to obtain QALY, we followed the guidelines from NICE (Brazier and Rowen, 2011) to design specific instruments, that is “preference-based measures derived from validated measures of HRQoL, with the value set obtained from the general population preferably using techniques similar to the protocol used to obtain the EQ-5D value set”. Following these guidelines, first, we propose a new dependency health state classification system based on items usually reported in the functional disability indexes, plus an additional item, which recalls levels of dependence related to cognitive problems that these traditional disability indexes do not consider. Second, we estimate the DEP-6D scoring algorithm from a set of direct measurements of dependency states performed with the time trade-off (TTO) from a general population sample. 6 2. Methods 2.1. The DEP-6D dependency health states classification system The process to design the descriptive component of the DEP-6D instrument is detailed in Electronic supplementary material [INSERT LINK TO ELECTRONIC SUPPLEMENTARY MATERIAL]. The resulting selection of dimensions/attributes is showed in Table 1. DEP-6D states are described as a combination of 6 attributes (eat, incontinence, personal care, mobility, housework and cognition/mental problems), with 3 or 4 levels each. The first four dimensions assess the level of performance of basic ADLs and they are present in the most common scales used to measure functional disability (Mahoney and Barthel, 1965; Katz et al., 1963). The fourth dimension assesses instrumental ADLs that need to be performed daily. Finally, a remarkable feature of the DEP-6D system is that includes cognitive/mental impairment as one of the attributes that characterizes dependency functional status. This dimension includes issues related to the ability to make decisions about the ADLs and the presence or absence of behavioral disorders that hinder aid. Thus, some individuals may achieve most ADLs but do not act by their own initiative and do not collaborate with their caregivers. According to the interviewed experts, such situations would not be described in a suitable way just by means of the top 5 dimensions shown in Table 1; the sixth attribute (cognitive/mental problems) is added to capture this sort of dependency. These aspects of dependence are also considered on different scales of physical and cognitive disability (Linn y Linn, 1982; Hébert et al. 2001). [Insert Table 1] 7 2.2. Selection of health states to be valued The estimation of a scoring algorithm for the DEP-6D requires previously the direct valuation of a set of dependency health states by a sample of the general population. It is common in studies to model generic preference-based measures that the selection of the health states to be valued is based on an orthogonal design. However, the focus group conducted revealed that some combinations of DEP-6D dimensions and levels are implausible. In consequence, we opted by applying an optimal design (Fedorov, 1972) in order to exclude unrealistic combinations ensuring, at the same time, the ability of obtaining accurate estimates for the remaining DEP-6D states. SAS Software (version 9.1; OPTEX Procedure) was used to generate a set of 24 combinations divided into four blocks of size six (Table 2). The D-efficiency of the design obtained was 75.5%. This design only captures main effects, so the existence of first and higher-order interaction effects cannot be tested. [Insert Table 2] 2.3. Selection of respondents The sample consisted of 312 citizens drawn from the Galician general population (a region of Spain) recruited at home in 2011 using a stratified random sampling. Face-to-face interviews were conducted by 6 trained interviewers. Each participant valuated only one of the four blocks of DEP-6D states shown in Table 2. Blocks were randomly assigned among all the participants. The order in which the six states were presented was randomized to minimize order effects. The average time per interview was around 20 minutes. 2.4. The questionnaire 8 The questionnaire was structured into five sections. Firstly, the DEP-6D classification system was explained to the respondents. Next, six dependency states were valued using the TTO method (Torrance et al., 1972; Torrance, 1986). In the third part of the questionnaire, participants were asked to rank the six states previously valued. In the fourth section, respondents chose the first and second attribute they regarded as the most severe among the six dimensions of the DEP-6D. Lastly, standard sociodemographic questions (age, sex, income, education level, etc.) were asked to the participants. 2.5. The valuation method Elicitation procedure began with a starting question to identify whether the state to be valued was regarded as a state better than dead (SBD) or a state worse than dead (SWD). Participants were asked to assume that they were seriously ill, in such a way that they would die unless they got a treatment. They had to make a choice between dying in a few days (No treatment) and spending 10 years in the dependency state being valued followed by death (Treatment). If they chose (refused) the treatment this means that the state was regarded as a SBD (SWD). Depending on how the dependency health state was considered, the framing used for the TTO assessment was different (see Figure 1). For SBD, the framing consisted of the comparison between living 10 years in the dependency state (No treatment) and living XBDS years in full health (Treatment). Full health was described as being in 111111 state. Next, an iterative updown procedure based on standard decision analysis for ‘zoomed in’ on the indifference values in preference assessment (e.g. Keeney and Raiffa, 1993) was applied to find the number of years X*BDS at which the respondents were indifferent between the two options. The starting value for XBDS was set equal to 5 years. When respondents chose one of the alternatives, XBDS was adjusted up or down until the indifference point was bounded. For SWD, 9 the choice was between dying and spending XWDS years in full health followed by (10 - XWDS) years in the state being valued. As before, the value for XWDS was initially set in 5 years and moved up or down until the convergence process terminated. The final stage was to obtain the value X*BDS or X*WDS which let the participant indifferent between both options. When indifference was not directly obtained (most cases), the midpoint of the interval is presumed. [Insert Figure 1] Utility for state i, from participant j, yij, was obtained by assuming that the utility of each alternative can be decomposed according to the QALY model for chronic health states, i.e. H(Q)  T, where H(Q) is the utility function over health status and T are life years. In addition, the convention that the utility of full health (state 111111) is 1 and the utility of death is 0 was adopted. If state i is regarded as a SBD then, yij =X*BDS /10, whereas if is regarded as a SWD, yij=-X*SWD /(10X*SWD). However, since negative utilities calculated in this way do not have a lower bound, resulting in distributions very skewed to the left, we applied the transformation suggested by Patrick et al. (1994), yij = - X*WDS /10, bounding negative values at -1. Shaded boxes in Figure 1 contain the value assigned to yij in each of the hypothetical paths. 2.6. Statistical inference We estimated a main effects model by using random effects estimators. The random effects regression model was used because the same individual values six dependency states and so those observations were not independent. The general equation was defined as: yij = α +  dϵD  lϵL βld Xld + uj + eij, [1] where yij denotes the value that respondent i assigns to dependency state j; α is the intercept; Xld represents the fifteen dummy variables, which indicate the presence of either level 2 or 3 16 Although we tested both models with interaction terms similar to LEAST and MOST terms in the SF-6D, none of them was significant, so our algorithms reflect main effects only. In addition we re-estimated model 2 (not shown) including a series of dummy variables to account for the characteristics of the respondents. The coefficients of the variables describing the health states were not affected. From the added variables, only to live in a village with more than 10.000 inhabitants had a positive and significant coefficient. The coefficients for model 2 show that the greatest decrements to health state value associated to the worst level in a dimension concern ‘mental problems’, ‘mobility’, ‘incontinence’, ‘personal care’, ‘eat’, and ‘housework’, in that order. These results are broadly consistent with the relative importance of the dimensions directly provided by the participants. In this way (see Table 3) 79% of respondents place mental impairment in first or second place in order of severity, followed by incontinence (37%), mobility (29%), personal care (26%), eat (25%) and housework (3%). Only mobility and incontinence interchange their positions with regard to the ranking derived from parameter estimates. The level of agreement between actual and predicted valuations for the 24 states used to estimate model 2 is quite high both according to Pearson correlation coefficient (rho = 0.99) and the MAE (0.048). From the coefficients estimated, we may predict the utility weight for any dependence state that was not valued directly. Applying this scoring algorithm, we obtain that the utilities associated to the different dependence profiles generated by the DEP-6D instrument range from -0.837 to 0.704. 17 3.5. Construct validity The sample of dependents used to approach the construct validity of the new instrument is made up predominantly of elderly people (21.5% has less than 75 years and 35.5% has more than 84 years), mainly women (79.9%). These dependents are in very bad condition. Most of them have severe mental disorders, the levels which produce the largest losses of utility (69.6% has level 3 or 4 in the ´cognition problems´ dimension). This instrument does not seem present an important floor effect because only a 6.3% of patients present the worse score (DEP-6D=-0.837). A very strong correlation was achieved between both DEP-6D and Barthel index (Spearman´s rho=0.85). As hypothesized this correlation is higher when the ´housework´ and ´cognition problems´ dimensions are not included in the score (Spearman´s rho=0.94). Although we used a convenience sample, and these results should be confirmed with a larger sample and using several instruments, the result obtained seem support the construct validity of the DEP-6D. 4. Discussion This paper reports the estimation of a preference-based scoring algorithm for a new dependency health state classification system coined as DEP-6D. In the DEP-6D instrument, each possible condition of daily dependence is characterized as a combination of 6 dimensions with 3 or 4 levels each. The model estimated for the DEP-6D allows to generate preference scores or utilities for a wide range of states of dependency. The interpretation given to this is akin to the one given to the scores predicted by other based-preferences generic instruments. 18 Other researchers before us (Bravata et al., 2005; Sims et al., 2005; Sims et al., 2008) have elicited preference scores for combinations of ADL dependencies. There are some remarkable differences between those studies and ours, however. An obvious difference deals with the way dependencies are characterized. Whereas our classification system resembles the combination of dimensions and different severity levels, each of the attributes used by the abovementioned authors contains one single level, reflecting just dependence or absence of it, but not how serious the dependence is in each dimension. In some respect, the 6 Katz ADLs plus an additional ADL of walking first used by Goldstein et al. (2002), and afterwards by Bravata et al. (2005) and by Sims and colleagues, are similar to the lowest levels of the first four dimensions of the DEP-6D. Hence, the DEP-6D is able to describe a wider and richer set of dependence situations. Besides, the DEP-6D includes cognitive impairment as one of its attributes, one critical dimension strongly associated with dependency. Almost 80% of respondents in our sample regarded this dimension as the most severe. Another difference concerns the source of preferences. The DEP-6D is fully consistent with the so-called ‘societal perspective’ (Drummond et al., 2015), according to which economic evaluations should include all potential effects and costs regardless of payer or beneficiary. As Gold et al. (1996) claim, a logical extension of that reasoning suggests that society’s preferences should be gathered from a representative sample of general population. The DEP-6D algorithm is based on community preferences, not on preferences elicited from a specific sample of adults. Lastly, other differences arise from the analysis of the consistency and invariance of responses. Only 33% of participants in Bravata et al.’s (2005) study gave at least one utility less than one and had no order inconsistencies. Moreover, 19% of the respondents gave a utility of 1 for all health states of ADL dependence. In our study, 79% of participants had no order inconsistencies and no participant gave a utility of 1 for all states. Extreme invariance 19 explains the absence of variability in the mean utilities reported by Bravata et al. which range from 0.76 to 0.89 (-0.60 to 0.65 in our study). To explain these findings, the authors argue that it is possible that at least some of the invariant subjects did not understand actually the complexity of the valuation task. Factors such as low educational level or the high age of respondents (average age was 73.2) may play a role here. Notwithstanding, in our opinion, there is another additional explanation for the high percentage of invariant responses. The elicitation procedure used in their study was the standard gamble method. It has been repeatedly shown (Bleichrodt, 2002) that this method suffers from upward biases (particularly loss aversion) that generally will lead to overestimate the utility of a health state (Bleichrodt et al., 2007; Abellán-Perpiñán et al., 2012). This bias may explain that some subjects gave utility of 1 to all health states. On the contrary, the elicitation technique used in our study was the TTO. We opted for the TTO because there is some evidence (Bleichrodt and Johanesson, 1997; Abellán-Perpiñan et al., 2009) suggesting that the biases for TTO (loss aversion and scale compatibility) go in mutually opposite directions and tend to offset each other. This means that we could expect that the TTO lowers error/objection responses. Other elicitation methods, such as lottery equivalent techniques or discrete choice experiments (DCE), also have advantages that could justify their use (Abellán-Perpiñán et al., 2012; Clark et al., 2014). However, we thought that the TTO is less cognitive demanding than the lottery equivalent and, besides, that as it was implemented in a choice-based manner (respondents make multiple choices until indifference is reached) it was a good alternative to a DCE, keeping the capability to obtain the indifference point at the individual level. Validity analysis show that the preferences that generated the DEP-6D scoring algorithm present good properties in terms of internal and external validity. Concerning the 20 psychometric properties of this new instrument, the preliminary results obtained from a convenience sample provides support for the construct (convergent) validity. However, more research (in progress) is required to confirm this result as well as the content validity, practicality and reliability of the DEP-6D descriptive system. In terms of practicality, the instrument takes very little time to administer with a small respondent burden. However, future research should analyze the more convenient proxy responders given that, very frequently, the dependent can be too ill or cognitively impaired to answer. Caregivers may be the best option, as recommended by NICE (Brazier and Longworth, 2011), but the differences between dependents and carers should be analyzed within this context. Regarding content validity and reliability, the DEP-6D descriptive system has been obtained from highly validated instruments. Yet more research should be conducted in this direction. The study introduced in this article has also limitations. One comes from the fact that our design did not allow us to test for interaction effects. Whereas we are aware that this is a restrictive assumption, it has to be noticed that those models that allow for interactions do not usually improve the predictive ability of main effects models and frequently yield inconsistent estimates (Dolan, 1997; Greiner et al., 2005). In addition, the majority of the algorithms estimated for the EQ-5D and the SF-6D only reflect main effects (Tsuchiya et al., 2002; Lam et al., 2008; Ferreira et al., 2008; Brazier et al., 2009; Abellan-Perpiñan et al., 2012) or, at best, include some extreme level interaction term such as the intercept dummy N3 in the EQ-5D (Dolan, 1997). Another potential limitation related to the design is that the worst DEP-6D state was not included. Although in the design used there is two very severe states (all dimensions are in their worst level except one), perhaps the inclusion of the worst state would have had some effect on the parameters estimated. We were worried about the 21 fatigue and the refusal to participate of the interviewees (they did not receive economic incentive) and only the states obtained from optimal design were valuated. Otherwise, that the TTO utilities can be less biased than those measured with other methods does not mean that our TTO measurements cannot be improved. We used an up-down procedure to reach the indifference between the two alternatives confronted in our TTO assessments. Once the indifference point was bounded between two durations, we recorded it as the midpoint between both. This means that utilities were calculated with an accuracy of 0.05 points of utility. A more accurate way to find the indifference point perhaps had affected our estimations, making, for example, level 2 of ‘housework’ dimension significantly different from level 1. In any case, it is reasonably standard to calculate utility scores at this level of accuracy and maybe the small sample size could explain this lack of significance. Finally, as noted before, our data show clusters of values primarily at both ends of the utility range and around 0. Such distributions of individual values are not unusual in studies addressed to estimate social tariffs. This is the case, for example, in the study conducted to estimate the EQ-5D-3L value set for the UK (Dolan and Roberts, 2002). Recent findings regarding the new EQ-5D-5L value set for England (Devlin et al., 2015), in which a composite TTO method was used, also show clusters of values at -1, 0, 0.5, and 1. This is interesting because the descriptive systems and protocols used for the DEP-6D and EQ-5D valuation studies are different, suggesting that the emergence of clusters in our study can be related to the application of the TTO method (e.g., the iteration procedure). Obviously, the occurrence of clusters affects the scoring algorithm, which reflects, to some extent, that latent distribution. Further methodological research is needed on this issue, in order to avoid or mitigate this limitation when obtaining the set of values. The DEP-6D is designed to be applicable in a general adult population. However, given that almost all of its items are concerned with severe levels of 22 dependence, with a very high prevalence in the elderly people, it can be especially useful for this group of population or institutionalized patients. This instrument is particularly indicated to detect dependence in basic ADLs, such as Barthel or kantz scales do. However it can also measure dependence in instrumental ADLs (although with a lower sensitivity level), which allows to detect dependence at initial stages. The DEP-6D could have both clinical and economic applications. It can be used in clinical studies as a supplemental tool for measuring the clinical course of different diseases that cause dependence, as well as the effects of their treatment. To the best of our knowledge, no index commonly used in clinical context to measure the loss of independence has interval scale properties, a desirable property if we want to add the partial score of dimensions to a final score. Moreover, given that this instrument measures the changes in the dependence level in utility scores anchored to a death-full health scale, QALY gains derived from DEP-6D could be used in the economic evaluation context. It proves especially useful in evaluating programs that produce significant changes in the level of dependence (physical or decisional) in such a way that their effectiveness may be undervalued if a generic instrument were used. The proposed instrument provides complementary utilities to generic measures with a wider focus and should be used alongside these to provide a more accurate measurement. The way in which both types of measures can be combined to be used in policy making is an open debate. In any case, we consider that if the DEP-6D produce significant differences with regard to other generic based-preferences measures, these values should be incorporated in the economic evaluation (at least in a subsequent sensitivity analysis of the QALYs gained) because it is more sensitive to measure changes in the dependence levels. For instance, the SF-6D descriptive system does not allow identify dependence states (only provide information 23 about the severity of health states but does not reveal whether this condition cause dependence on others). In turn, EQ-5D-5l only can identify if a person is or not unable to walk, wash/dress or do usual activities but it does not provide information about the severity level or any other dependence dimension. DEP-6D describe, in general, more severe conditions than these generic measures. In consequence, most mean values are negative (states considered worse than death) and the minimum value for the worst DEP-6D condition is - 0.837 whereas the lowest utility for the TTO Spanish EQ-5D (Badía et al., 2001) is -0.654. This gap is larger if the comparison is done with respect to the Spanish SF-6D (Abellan-Perpiñan et al., 2012) with a minimum utility of -0.357. In summary, this study proposes a new instrument to measure the level of dependence on others for ADLs. The descriptive system covers both physical and decisional dependence and can be used within a clinical and economic context. The answers we obtained are highly consistent internally and externally. Although the instrument appears to show good properties in terms of convergent validity and its descriptive system has been obtained from highly validated instruments, further research is required in terms of its psychometric properties. The impact of the use this instrument on the results of the economic evaluation should be compared to those derived from generic preference-based measures. 24 5. References Abellan-Perpiñan, J. M., Bleichrodt, H., & Pinto-Prades, J. L. (2009). The predictive validity of prospect theory versus expected utility in health utility measurement. Journal of health economics, 28(6), 1039-1047. Abellán Perpiñán, J. M., Sánchez Martínez, F. I., Martínez Pérez, J. E., & Méndez, I. (2012). Lowering The ‘Floor’Of The Sf‐6d Scoring Algorithm Using A Lottery Equivalent Method. Health economics, 21(11), 1271-1285. Badia, X., Roset, M., Herdman, M., & Kind, P. (2001). A comparison of United Kingdom and Spanish general population time trade-off values for EQ-5D health states. Medical Decision Making, 21(1), 7-16. Badia, X., Roset, M., & Herdman, M. (1999). Inconsistent responses in three preferenceelicitation methods for health states. Social science & medicine, 49(7), 943-950. Bleichrodt, H., Abellan-Perpiñan, J. M., Pinto-Prades, J. L., & Mendez-Martinez, I. (2007). Resolving inconsistencies in utility measurement under risk: tests of generalizations of expected utility. Management Science, 53(3), 469-482. Bleichrodt, H., & Johannesson, M. (1997). The Validity of QALYs An Experimental Test of Constant Proportional Tradeoff and Utility Independence. Medical Decision Making, 17(1), 2132. Bleichrodt, H. (2002). A new explanation for the difference between time trade‐off utilities and standard gamble utilities. Health economics, 11(5), 447-456. 25 Bravata, D. M., Nelson, L. M., Garber, A. M., & Goldstein, M. K. (2005). Invariance and inconsistency in utility ratings. Medical decision making, 25(2), 158-167. Brazier, J., & Deverill, M. (1999). A checklist for judging preference‐based measures of health related quality of life: Learning from psychometrics. Health economics, 8(1), 41-51. Brazier, J. E., Fukuhara, S., Roberts, J., Kharroubi, S., Yamamoto, Y., Ikeda, S., ... & Kurokawa, K. (2009). Estimating a preference-based index from the Japanese SF-36. Journal of clinical epidemiology, 62(12), 1323-1331. Brazier, J., Roberts, J., & Deverill, M. (2002). The estimation of a preference-based measure of health from the SF-36. Journal of health economics, 21(2), 271-292. Brazier, J. E., & Roberts, J. (2004). The estimation of a preference-based measure of health from the SF-12. Medical care, 42(9), 851-859. Brazier, J., & Longworth, L. (2011). NICE DSU technical support document 8: an introduction to the measurement and valuation of health for NICE submissions. NICE Decision Support Unit, London. Brazier, J. E., & Rowen, D. (2012). NICE DSU Technical Support Document 11: Alternatives to EQ-5D for generating health state utility values. 2011. Chisholm, D., Healey, A., & Knapp, M. (1997). QALYs and mental health care. Social psychiatry and psychiatric epidemiology, 32(2), 68-75. Clark, M. D., Determann, D., Petrou, S., Moro, D., & de Bekker-Grob, E. W. (2014). Discrete choice experiments in health economics: a review of the literature. Pharmacoeconomics, 32(9), 883-902. 32 Table 3: Characteristics of respondents by type of questionnaire (%) Sample (n=312) Population Sex (female) * 47.4 51.6 Age* Mean (range) 41.5 (18-75) 45.4 Education† Primary studies or less 37.5 54.0 Secondary 39.4 27.9 University 23.1 18.1 Habitat* Rural 31.4 31.0 Intermediate 31.1 33.3 Urban 37.5 35.8 Live Alone† 13.5 19.6 Labour status‡ Employed 59.8 46.9 Pensioner/retired 10.9 23.8 Unemployed 16.0 8.2 Student 5.1 6.1 Domestic tasks 8.3 9.9 Home income† (€ monthly) <=500 5.9 3.7 500-1000 13.2 19.1 1000-1500 30.5 18.5 1500-2000 25.7 16.4 2000-3000 16.9 24.5 >3000 7.7 17.9 Good health (EQ-5D=11111) 76.3 Close dependent Any close dependent 53.2 Close dep. (not live together) 40.1 Close dep. (live together) 6.7 Duration of interview (minutes) 22.5 % Participants who placed this attribute in first or second place Eat 24,99 Incontinence 37,23 Personal care 26,25 Mobility 28,89 Housework 3,18 Cognitive/mental 79,46 The population data were obtained from: * Census record (2011) † Living conditions of Galician families survey (2007) ‡ Active population survey (2010). Office for National Statistics 33 Table 4 : Mean utilities for 24 states directly evaluated All participants Consistent participants State n Mean utility Std. Dev. n Mean utility Std. Dev. 122222 82 0.17 0.54 77 0.18 0.53 133334 82 -0.45 0.47 77 -0.49 0.45 211121 82 0.58 0.47 77 0.60 0.46 214232 82 0.06 0.59 77 0.04 0.60 313331 82 0.17 0.60 77 0.17 0.61 323433 82 -0.49 0.51 77 -0.53 0.48 111221 78 0.65 0.42 77 0.66 0.42 112132 78 0.35 0.57 77 0.35 0.57 112211 78 0.60 0.45 77 0.60 0.45 223234 78 -0.47 0.54 77 -0.48 0.53 234333 78 -0.42 0.51 77 -0.44 0.50 333122 78 -0.23 0.62 77 -0.24 0.61 111112 75 0.40 0.57 66 0.50 0.49 113233 75 -0.12 0.66 66 -0.12 0.66 213322 75 -0.07 0.64 66 -0.07 0.64 222131 75 0.24 0.60 66 0.26 0.60 234431 75 -0.37 0.62 66 -0.37 0.62 334234 75 -0.54 0.51 66 -0.55 0.51 123121 77 0.30 0.57 73 0.32 0.58 212223 77 -0.16 0.67 73 -0.15 0.68 233432 77 -0.45 0.58 73 -0.48 0.57 314434 77 -0.60 0.51 73 -0.62 0.50 324332 77 -0.32 0.61 73 -0.32 0.61 333231 77 -0.19 0.64 73 -0.21 0.65 34 Table 5: DEP-6D models model 1 model 2 Coefficient p-value Coefficient p-value Constant 0.776 0.000 0.773 0.000 EAT2 -0.152 0.000 -0.146 0.000 EAT3 -0.195 0.000 -0.188 0.000 INC2 -0.130 0.000 -0.131 0.000 INC3 -0.263 0.000 -0.263 0.000 PER2 -0.129 0.001 -0.133 0.001 PER3 -0.256 0.000 PER4 -0.230 0.000 PER3+4 -0.258 0.000 MOB2 -0.090 0.002 -0.086 0.003 MOB3 -0.133 0.000 -0.126 0.000 MOB4 -0.294 0.000 -0.289 0.000 HOU2 -0.066 0.140 -0.069 0.126 HOU3 -0.093 0.055 -0.089 0.066 MEN2 -0.228 0.000 -0.224 0.000 MEN3 -0.403 0.000 -0.403 0.000 MEN4 -0.527 0.000 -0.523 0.000 R2 Within Between Overall Wald chi2(p-value) 0.522 0.070 0.344 1604.65 (0.001) 0.522 0.071 0.344 1603.93 (0.001) N 293 293 Observations 1758 1758 35 Figure 1: TTO protocol Figure 2: Histogram for dependency state valuations 0 50 100 150 200 250 300 350 SBD SWD All states 36 ELECTRONIC SUPPLEMENTARY MATERIAL Selection of dimensions and levels of the DEP-6D classification system The DEP-6D classification system is an instrument used to characterize the level of daily dependence on others. To define the severity of dependence it is necessary to select the most relevant dimensions/attributes of the dependence and the levels of severity that these dimensions can present. Given that we want to estimate the severity weights in such a way that they can be interpreted as a utility score, and then be used in QALY calculations, we need a manageable number of dimensions. To this end, firstly we made an initial selection of attributes/dimensions based on the most widely-used generic disability measures. Next, we used the Spanish National Health Survey to identify those dimensions showing a high correlation and they were grouped into a common dimension in order to lower the number of potential attributes to be included in the classification system. Finally, a focus group with five experts in evaluation of ADL dependencies for the regional government of Galicia (north-western Spanish region) were conducted. From focus group we made the final selection of dimensions and levels and identified potential implausible combinations among them. The following is a more detailed explanation of these steps. First step: Identify the common attributes used in the most widely-used generic disability measures. The Barthel index (Mahoney and Barthel, 1965) and the Kantz index (Kantz et al., 1963) are the most cited indicators of dependence in activities of daily living (ADLs). The Barthel index measures functional disability by quantifying patient performance in 10 ADLs (feeding, grooming, dressing, bathing, bowel and bladder care, toilet use, ambulation, transfers, and stair climbing). The Kantz index includes 6 activities (feeding, dressing, bathing, continence, toilet use, transfers). These activities are present in most indicators used to measure performance in ADLs (MacDowell, 2006). Second step: Merge activities to reduce the number of dimensions. Starting from these indicators, we selected some activities separately and merged those which could be grouped into a more general dimension either by grouping activities that present a correlation greater than 75% or considering them as a specific level of a dimension. The information contained in the Spanish National Health Survey (SNHS) was used for this purpose. The dimension feeding was considered in the DEP-6D as an independent dimension, distinguishing between needing help and being unable. 37 Grooming and dressing present a correlation higher than 75% in the SNHS, therefore both were unified in a more general concept called personal care. As usual, in the Barthel index we have distinguished two levels of severity, needing help and being unable. Within the personal care activities, the ability to bathe without help is usually the first to be lost. So the SNHS shows that 50% of people who cannot bathe themselves can perform other self-care activities, however, only 9% of those who cannot dress themselves can bathe without assistance (even in this case they could not perform the activity completely because they would need help to dress). Therefore we incorporate the dependence in the bath as a first level of dependency in the personal care dimension. The dimension referred to incontinence is collected in the Barthel index and Kantz index (Barthel distinguishing between urinary and fecal). Given that our objective is to measure the level of dependence, we incorporate a dimension that takes into account whether the individual has urinary or fecal incontinence, or both, but only if the person needs help derived from this condition, as incontinence does not necessarily imply dependence. Barthel index and Kantz index recall items related to mobility (ability to move from bed to chair and back, to walk and to climb stairs in Barthel index and ability to move from bed to chair and back in Kantz index).We designed a mobility dimension in which the worst level corresponds to inability to change position or bed-ridden or chair-ridden. However, the research team had doubts about the design of the rest of levels of this dimension, and postponed the decision to focus group results. Third step: Discussion about another dimensions. Given that the DEP-6D aims to characterize daily dependence, we considered it relevant to take into account the instrumental activities that need to be performed daily, mainly making food and cleaning crockery. These activities, present in most indicators used to measure performance in instrumental ADLs (Lawton y Brody, 1969), have a correlation higher than 75% (based on SNHS). So both were grouped in a single dimension, housework, distinguishing between needing help and being unable. We also considered relevant to include the cognitive problems dimension. Obviously the above dimensions can be caused by mental or physical problems. In principle this is not relevant to measure the level of daily dependence. However there are aspects of cognitive problems, related to the ability to make decisions and take responsibility about the ADLs, which are not considered in the above dimensions. So it is possible for a person to perform the abovementioned basic activities but present a significant dependence in decision-making (for instance, a person with Down syndrome or with an incipient Alzheimer). This condition can cause the patients not to be able to live alone (indicating a 38 high level of dependence) but their dependence on basic activities to be very low. Another important factor when measuring the level of dependence, also related to cognitive problems, is the presence or absence of behavioral disorders that hinder aid. These aspects of dependence are considered on different scales of physical and cognitive disability (Linn y Linn, 1982; Hébert et al. 2001; etc.) and were mentioned in informal meetings of the research team with health professionals and dependent relatives. The final design of this dimension was postponed to focus group results. Fourth step: Conducting a focus group. We conduct a focus group with five experts in the assessment of severity of dependency, valuation required to access to benefits derived from the Dependency Law. The Dependency Law began to be applied in 2007 in Spain. It regulates the public support for people who cannot lead independent lives for reasons of illness, disability or age. An Official Dependence Index is created to measure the level of dependency. Experts selected to participate in the focus group are people whose work is to assess the loss of independence in ADLs of the people applying to these public subsidies. From this assessment the dependence score is established. So these experts have extensive experience in assessing the level of dependence and caregiver burden. The meeting proceeded as follows: 1. Discussion about the ability to go to the bathroom (dimension present in the Kantz index and Barthel index) and the levels of mobility dimension. The experts consider that a good description of mobility dimension is to distinguish between ability to move in and out of home (in addition to the ability to move from bed to chair and back). They consider that the ability to go to the bathroom without help is related to the ability to move within the home. 2. Regarding cognitive/mental problems, there is consensus that the above attributes capture some aspects of mental dependence but not all of them. Thus, if a patient has severe cognitive problems, such limitations are evident in the ability to eat, to move, etc. However, there are situations that are not included in the above attributes and are very relevant when measuring dependency. First, they consider that there are conditions associated to cognitive problems that require the supervision of a third party but these conditions may or may not affect, the ability to perform basic ADLs. So the person may have the ability to bathe alone but needs someone to order him. The level of dependence would be strongly underestimated if these aspects were not taken into account. Second, all participants emphasized that it is important to not only quantify the level of dependence in activities that the patient cannot perform but also to quantify their level of collaboration with the caregiver. Any activity becomes especially toilsome if the dependent is violent. The Spanish Official Dependence Index caters for both aspects (ability to make decisions 39 and behavioral disorders). The final version of the cognition problems dimension was obtaining from this discussion. 3. We check the rest of dimensions. They consider that the most relevant dimensions of dependence has been recall and the levels of each dimension are not ambiguous and present a clear upward progression of dependence. 4. In the second part of the meeting, we identified unrealistic combinations in order to exclude such combinations from the set of dependency health states to be evaluated in the survey. For this the experts checked, one by one, all the possible combinations of pairs of attributes (a total of 183 combinations of two levels were checked) and 31 of them were considered implausible. For instance, the experts noted that level 4 in mobility is incompatible with level 1 in eat, level 1 and 2 in personal care, and level 1 and 2 in housework. These restrictions —which produce a total of 314 plausible dependency states— were introduced in the OPTEX Procedure to generate the 24 states evaluated in the survey. References Hébert, R., Guilbault, J., Desrosiers, J., & Dubuc, N. (2001). The functional autonomy measurement system (SMAF): a clinical-based instrument for measuring disabilities and handicaps in older people. Geriatrics Today, 4, 141-158. Katz, S., Ford, A. B., Moskowitz, R. W., Jackson, B. A., & Jaffe, M. W. (1963). Studies of illness in the aged: the index of ADL: a standardized measure of biological and psychosocial function. Jama, 185(12), 914-919. Mahoney, F. I., & Barthel, D. (1965). Functional evaluation: the Barthel index. Maryland state medical journal, 14, 56-61. Lawton, M.P., & Brody, E.M. (1969). Assessment of older people: Self-maintaining and instrumental activities of daily living. The Gerontologist, 9(3), 179-186. Linn, M. W., & Linn, B. S. (1982). The rapid disability rating scale—2. Journal of the American Geriatrics Society, 30(6), 378-382. McDowell, I. (2006). Measuring health: a guide to rating scales and questionnaires. Oxford University Press.