Permanyer et al. Population Health Metrics (2022) 20:1 https://doi.org/10.1186/s12963-021-00279-8 RESEARCH Open Access On the measurement of healthy lifespan inequality Iñaki Permanyer1,2* , Jeroen Spijker1 and Amand Blanes1 Abstract Background: Current measures to monitor population health include indicators of (i) average length-of-life (life expectancy), (ii) average length-of-life spent in good health (health expectancy), and (iii) variability in length-of-life (lifespan inequality). What is lacking is an indicator measuring the extent to which healthy lifespans are unequally distributed across individuals (the so-called ‘healthy lifespan inequality’ indicators). Methods: We combine information on age-specific survival with the prevalence of functional limitation or disability in Spain (2014–2017) by sex and level of education to estimate age-at-disability onset distributions. Age-, sex- and education-specific prevalence rates of adult individuals’ daily activities limitations were based on the GALI index derived from Spanish National Health Surveys held in 2014 and 2017. We measured inequality using the Gini index. Results: In contemporary Spain, education differences in health expectancy are substantial and greatly exceed dif- ferences in life expectancy. The female advantage in life expectancy disappears when considering health expectancy indicators, both overall and across education groups. The highly educated exhibit lower levels of lifespan inequality, and lifespan inequality is systematically higher among men. Our new healthy lifespan inequality indicators suggest that the variability in the ages at which physical daily activity limitations start are substantially larger than the vari- ability in the ages at which individuals die. Healthy lifespan inequality tends to decrease with increasing educational attainment, both for women and for men. The variability in ages at which physical limitations start is slightly higher for women than for men. Conclusions: The suggested indicators uncover new layers of health inequality that are not traceable with currently existing approaches. Low-educated individuals tend to not only die earlier and spend a shorter portion of their lives in good health than their highly educated counterparts, but also face greater variation in the eventual time of death and in the age at which they cease enjoying good health—a multiple burden of inequality that should be taken into consideration when evaluating the performance of public health systems and in the elaboration of realistic working- life extension plans and the design of equitable pension reforms. Keywords: Population health measures, Life expectancy, Health expectancy, Lifespan inequality, Socio-economic inequality, Spain, Education Background widely used indicator: its values are regularly reported by When measuring population health across countries, life international institutions and National Statistical Offices expectancy (LE) has become the most well-known and worldwide. Despite its popularity, LE has two impor- tant shortcomings. First, its definition only takes into consideration the mortality of individuals, thus ignor- *Correspondence: ipermanyer@ced.uab.es ing the health status of those who remain alive. Second, 1 Centre d’Estudis Demogràfics, Carrer de Ca n’Altayó, Edifici E‑2, Campus de la UAB, 08193 Cerdanyola del Vallès, Spain LE is simply an average that does not explain how the Full list of author information is available at the end of the article length of life is distributed across the members of a given © The Author(s) 2021. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/. The Creative Commons Public Domain Dedication waiver (http://​creat​iveco​ mmons.​org/​publi​cdoma​in/​zero/1.​0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data. Permanyer et al. Population Health Metrics (2022) 20:1 Page 2 of 9 population. To address these limitations, two impor- complete lifespans, we suggest partitioning the latter in tant research avenues have evolved during the last dec- the periods that individual spend across health states and ades. The first one has promoted the creation of ‘health investigating how these periods are distributed across expectancy indicators’ (HE) that measure the number of individuals. years that individuals are expected to live in ‘good health’ These different ideas are schematically represented under prevailing mortality and morbidity conditions. in Fig. 1, in which the white cells show information and These measures combine not only the quantity but also examples on existing families of population health indi- the quality (in terms of health) of the expected remaining cators, and the shaded cell does so for the new family years of life [1, 2]. The second one urges researchers and of indicators we propose to investigate here. The health policy-makers alike to look beyond averages and assess outcome of the indicators included in the first column how ‘unequal’ or ‘disperse’ length of life distributions are, of Fig. 1 is “individuals’ longevity”, while the outcome of i.e., to quantify the amount of ‘lifespan inequality’ (LI) those in the second column is “individuals’ healthy lon- existing in the ages-at-death distributions across popula- gevity” (i.e., the number of years individuals live in good tion members [3, 4]. health). The indicators in the first row measure the mean Surprisingly, key insights and contributions from these of the distribution (i.e., they are an average) while those important research avenues have barely influenced each in the second row measure the variability/inequality in other over the years. While HE indicators have made the distribution. The main contribution of this paper is the important distinction between ‘quantity’ and ‘qual- to highlight that, currently, there are no indicators fitting ity’ of years of life and LI measures have separated ‘effi- in the second row and second column of the table, i.e., ciency’ (i.e., average achievement) from ‘equality’, these measuring the variability in individuals’ healthy lifespans. two important analytical axes have not been consid- This is what the new HLI indicators are aiming at. ered simultaneously. The main proposal of this paper There are many reasons why healthy lifespan inequal- is to bring together these two strands of research into a ity can be considered a fundamental quantity in health coherent whole to obtain a deeper and more comprehen- research that should be reported alongside other well- sive understanding of contemporary population health known mortality and morbidity summary indicators. dynamics. To attain this goal, we introduce the concept Population health means more than simply averting of ‘healthy lifespan inequality’ (HLI), which is designed death, and societies are concerned not only about aver- to investigate the extent to which healthy (and unhealthy) age levels of disease and/or disability, but also in the ways lifespans are unequally distributed across population in which the latter are distributed. Larger levels of HLI members. Rather than looking at the distribution of indicate greater heterogeneity in underlying population Fig. 1 Summary diagram organizing population health indicators along the ‘quantity/quality’ and ‘efficiency/equality’ axes. Source Authors’ own elaboration Permanyer et al. Population Health Metrics (2022) 20:1 Page 3 of 9 health, an issue that can have implications both at the expectancy at birth—and are therefore indistinguishable micro and at the macro level. At the individual level, HLI for any HE indicator—the extent of health inequality in indicators measure uncertainty in the timing of disease, B is much larger than in A. Consider now a third hypo- disability or physical limitation onset—the latter being thetical society C in which all individuals die when they key events with a strong and enduring impact on indi- approach the age ω, but in which one half of its mem- viduals’ well-being—with potentially important effects bers enjoy a perfect health until they die and the other on individuals’ decision-making. At the macro level, half spends half of their lifetime in perfect health and HLI are appealing, simple measures of population health the remaining half in very limiting health conditions (see that might arguably be more meaningful than their LI right panel in Fig. 2). Despite the rampant health inequal- counterparts: while the former look at the distribution ities, current LI measures would conclude that length-of- of a normatively desirable quantity (‘years spent in good life inequality in such society would be close to zero. health’) the latter complicate matters by including quan- These examples suggest the need to complement cur- tities that might not be universally desirable (e.g., ‘years rently existing measures of population health with indi- spent in very bad health’). cators that simultaneously account for the quality (i.e. To illustrate the usefulness of the suggested approach, health status) of life years and the extent to which these in this paper we investigate how LE, HE, LI and the new years are equally distributed across individuals. This is HLI indicators behave across sex and education groups in what the new HLI indicators are meant for. To compute contemporary Spain. them we combine information on mortality and morbid- ity, and proceed in two steps. First, we estimate the distri- Methods bution of age-at-disability onset for the population under Overview study on the basis of the so-called ‘morbidity curve’ (see Consider three hypothetical population health distribu- below), although other more sophisticated and data tions. In the first one (A), individuals start developing demanding methods can be implemented as well (see a chronic disabling disease around the age of 70, and Discussion section). Second, we measure the amount of the variability in such ages is not very large. In the sec- inequality in such distribution. ond one (B), the average age at which individuals start developing diseases is also 70, but the variability is much larger. While both A and B have 70 years of healthy life Fig. 2 Hypothetical health distributions. Left panel: age at disability onset distributions for populations A and B; Right panel: survival and health-corrected survival curves (in continuous and dashed lines, respectively) for population C. Source: Authors’ own elaboration Permanyer et al. Population Health Metrics (2022) 20:1 Page 4 of 9 Estimating health distributions and HLI indicators is that the former are based on the Similar to the Sullivan approach [5], we combine infor- ′ dx distribution, while the latter are based on the dx   mation on age-specific survival with information on the distribution. prevalence of functional limitation or disability to esti- mate the distribution of ages at which individuals cease Data to be in good health. As a measure of disability we use the We use the mortality microdata files by educational Global Activity Limitation Index (GALI), originally attainment for those who died after age 35, which are designed and subsequently validated for health expec- publically available from the Spanish Statistical Office tancy comparisons across Europe, and capturing long- (INE). INE used a matching algorithm linking regis- term limitations (≥ 6 months) to perform daily activities tered deaths to population databases, including cen- due to health problems [6, 7]. The prevalence levels of suses, municipal population registers, the ministry of disability by age, sex and education are smoothed using education, and the Public State Employment Service three-year moving averages to avoid some noisy fluctua- [10], to obtain the deaths according educational attain- tions arising from the small sample size of some specific ment. The INE also provides the total population bro- subgroups. We use a synthetic cohort approach in which ken down by sex, age, and educational attainment, its members are subject to current mortality and morbid- which are required for the denominators of our mortal- ity conditions along their lifetimes. Multiplying the lx col- ity indicators. With these registers, we can determine umn of the life table (showing the number of survivors at mortality levels across three education groups (Less age x ) by 1 − πx (the percent of population at age x not than Primary, Primary, Secondary or more), and for ′ limited to carry out daily activities) we obtain lx : the women and men separately. number of healthy survivors at age x . This is the so-called The GALI is obtained from the 2014 and 2017 Spanish ′ ‘morbidity curve’. From the lx and lx columns we derive National Health Surveys (SNHS). Specifically, respond- ′ the standard dx and dx distributions (i.e. number of ents are asked if they have for the past six months or deaths and individuals ceasing to be in good health more been limited in ‘carrying out usual activities due between ages x and x + 1) defined as lx − lx+1 and to health problems’ [11]. We consider someone as disa- ′ ′ lx − lx+1, respectively. The ages at which individuals die bled when they respond to be either ‘severely limited’ or or cease to be in good health are thus described by the ‘limited but not severely’. Data for both years are pooled ′ dx and dx distributions.   and prevalence rates are generated by age, education and sex. Ages were bottom-truncated at 35 because of the Inequality measurement low number of respondents in the lowest educational Once the age-at-death and age-at-disability-onset distri- category (< 100) and to allow virtually all individuals butions have been estimated, we proceed to measure the to complete their formal education. Ages were also top corresponding levels of inequality. Many indicators are truncated at 85 because of lacking representative mor- currently available [8], and all of them are highly corre- bidity data at higher ages for all education groups. This lated [4, 9]. To present our findings we use the Gini coef- means that all LE, HE, LI and HLI indicators will be ficient, a measure of inequality taking values between 0 based on the 35–85 age range. (perfect equality) and 1 (maximal inequality). This is a frequently used indicator in studies of income or length- of-life inequality [3, 8, 9]. Other well-known inequality Results measures have been used as robustness checks. Figure 3 plots the estimated survival (top row) and healthy survival (bottom row) curves for ages 35–85 by sex and level of education for Spain in 2014–17. Among Other population health measures women, there are relatively small mortality differences The other population health measures used in this paper across education groups (see upper left panel in Fig. 3). In are standard LE, HE and LI indicators. LE measures the contrast, the survival curves by educational attainment average time individuals are expected to live under the are further apart from each other among men (upper mortality conditions prevailing in a given year, while HE right panel), thus indicating a steeper mortality gradient measures how much of that time is spent in good health. across them. As regards the healthy survival curves, we The latter is estimated using the Sullivan method [5], observe the expected pattern: they decline with age and with the GALI indicator as a measure of disability preva- they are more favorable for the higher educated groups lence. To measure LI, we apply the Gini index to theage- (see bottom row in Fig. 3). Both for women and for men, at-death distribution of the life table (i.e. the dx the declines of the healthy survival curves are more pro- column). Observe that the only difference between LI nounced than their ‘survival-only’ counterparts. Such Permanyer et al. Population Health Metrics (2022) 20:1 Page 5 of 9 Fig. 3 Synthetic cohort survival (top row) and healthy survival (bottom row) curves for women (first column) and men (second column) across education groups for ages 35–85, Spain 2014–17. Source: Authors’ calculations based on INE’s 2014–17 death files and the 2014 and 2017 SNHS steeper decline suggests that the distributions of ages at to live longer than men: the overall truncated LE are physical limitation onset are more unequally distributed 46.4 and 43.1, respectively. Inspecting the levels of than the corresponding age-at-death distributions – an health expectancy (upper right panel), we observe issue we will now quantify. that higher educated men and women are expected Table 1 presents the values of the mean health and to live in good health for longer than their lower edu- health inequality indicators associated to the survival cated counterparts. Furthermore, the differences in curves shown in Fig. 3. In the upper left panel we show HE across groups are much larger than the differences truncated LE indicators, which should be interpreted in LE. While longevity ranges between 44.5 and 46.9 in relative terms with respect to the maximal upper for women and 40.4 and 44.3 for men, HE ranges from bound of 85–35 = 50 years (i.e., the 46.9 years of trun- 22.4 to 34.0 for women and from 24.0 to 33.5 for men. cated life expectancy for highly educated women mean The female advantage in LE disappears when consid- that this group attained 94% (100*46.9/50) of the maxi- ering HE indicators. Regarding lifespan inequality, it mal longevity potential). As expected, among both decreases with increasing education for both women sexes higher educated groups are more longevous than and men. We also observe that LI is higher among less educated ones. Additionally, women are expected men, overall and across all education groups (see Permanyer et al. Population Health Metrics (2022) 20:1 Page 6 of 9 Table 1 Life expectancy, healthy life expectancy, lifespan inequality and healthy lifespan inequality indicators for women and men aged 35–85 across education groups in Spain 2014–17. Morbidity measures are based on the GALI index and inequality is measured with the Gini coefficient. Shaded cells highlight the new HLI indicators proposed in this paper. Source: Authors’ calculations based on INE’s 2014–17 death files and the 2014 and 2017 SNHS Mean health indicators Life expectancy (LE): Healthy life expectancy (HE): Low Mid High Total Low Mid High Total Women 44.5 46.3 46.9 46.4 22.4 29.3 34.0 29.7 Men 40.4 42.5 44.3 43.1 24.0 30.0 33.5 30.8 Health inequality indicators Lifespan inequality (LI): Healthy lifespan inequality (HLI): Low Mid High Total Low Mid High Total Women 0.081 0.075 0.068 0.069 0.129 0.119 0.106 0.113 Men 0.086 0.085 0.072 0.077 0.121 0.114 0.104 0.112 bottom left panel). Finally, as regards the new healthy of this study remain unaltered. All robustness check lifespan inequality indicators (see bottom right panel), calculations are shown in the Additional File 1. we observe that: (i) HLI are substantially larger than their LI counter- Discussion parts, with the former being, on average, 50% larger This paper suggests incorporating the new family of than the latter. ‘healthy lifespan inequality’ indicators into the list of (ii) HLI decreases with increasing education for both measures that are regularly reported to monitor popula- sexes. tion health. To the extent that one is interested in meas- (iii) Sex differences are not very large, but HLI indica- uring (1) how long individuals are expected to live, (2) tor values tend to be somewhat higher for women. how much of that time is spent in good health, and (3) how heterogeneous the distributions of lifespans can To ensure that our findings are not contingent upon be, it is only natural to be also interested in the hetero- arbitrary methodological choices, we have performed geneity of the distribution of healthy lifespans. Apply- several robustness checks. First, we have re-calculated ing the traditional LE, HE, LI together with the new HLI all our LI and HLI indicators for the different groups measures across education groups and sexes in contem- using alternative inequality measures, like the Gini porary Spain, several interesting patterns can be identi- mean difference, the Theil index or the Coefficient fied. In line with previous studies [12–14], we observe of Variation. All correlations among pairs of indica- that education differences in HE are substantial and tors are above 0.94. Second, we have used alternative greatly exceed differences in LE, a well-known pattern measures to define less-than-good health, namely ‘self- also observed across countries (i.e. international differ- assessed health’ and ‘ability to climb 12 stairs’. While ences in LE being smaller than international differences different disability measures generate different values in HE [7]). The female advantage in LE disappears when for the HE and HLI indicators, the substantive findings considering HE indicators, overall and across education Permanyer et al. Population Health Metrics (2022) 20:1 Page 7 of 9 groups. The observed social patterning in LI coincides of life irrespective of the health conditions in which that with recent studies suggesting that socio-economically year is lived into standard lifespan inequality measures advantaged groups tend to exhibit lower levels of lifes- can muddy the waters in regard to the interpretation of pan inequality, and that the latter is systematically higher their values. These conceptual problems are sidestepped among men [4, 15–20]. by our HLI measures, which only take into consideration The observed levels of LI dwarf when compared to the variability in healthy lifespans. their HLI counterparts: the new indicators suggest that In this paper we have focused our attention on the dis- the variability in the ages at which daily activity limita- tribution of healthy lifespans, but it could be equally rea- tions start can be substantially larger than the variability sonable to look at the distribution of unhealthy lifespans. in the ages at which individuals die (increases by a fac- Both approaches are interesting in their own right, and it tor of 1.5). Our findings indicate that the levels of HLI is not a priori clear what the relationship between the two are slightly higher for women – a result that coheres with approaches could be. One could easily imagine hypothet- the mortality advantage and morbidity disadvantage of ical scenarios with low levels of healthy lifespan inequal- women vis-à-vis men [21]. Importantly, HLI increases ity (i.e., everyone enjoying the same number of healthy with decreasing educational attainment, both for women years) but high levels of “unhealthy lifespan inequality” and for men. Recently, it has been suggested that because (i.e., large differences in the number of years individuals low-SES individuals tend to live shorter lives and face spend in bad health) or vice versa. These extremely inter- greater uncertainty in the age at which they will die than esting questions will be investigated in future research. high-SES individuals, they are exposed to a ‘double bur- The new indices hold promise to be an important com- den of inequality’ [4, 18]. In light of our findings, where plement to traditional LE, HE and LI measures, which, low-educated individuals are additionally expected to on their own, do not explain the whole story and might have shorter healthy lifespans and face greater uncer- lead to the elaboration of unfair or misinformed policies. tainty in the age at which they will start experiencing Inter alia, HLI indicators can be crucial for the design physical limitations, one could argue they are indeed of equitable pension schemes and retirement policies exposed to a ‘quadruple burden of inequality’. that are sensitive to the underlying heterogeneity in the To our knowledge, this is the first study proposing population, and for the public provision of medical care measures to assess the extent to which the years spent (especially at advanced ages). From a public health policy in good health are (un)equally distributed across indi- perspective, larger HLI might be indicative of a wors- viduals. Previous analyses have investigated differences in ening state of affairs across or within socially relevant health expectancies across subnational populations (e.g. groups – a cause of legitimate ethical concern, especially comparing HE across women and men, socio-economic when social patterning in health is attributable to pre- groups and/or racial groups [12–14]). While somewhat ventable causes. Finally, studying healthy lifespan varia- related, the two approaches are fundamentally differ- tion can enrich the longstanding ‘compression/expansion ent: the latter compares levels of HE across a closed list of morbidity’ debate, which aims at understanding of pre-specified groups, and the former investigates vari- whether prevalence of morbid conditions accord with or ability in healthy lifespan distributions across individuals. diverge from trends in mortality [28, 29]. Since its incep- These approaches echo a two-sided debate of the early tion more than 30 years ago, the contrasting hypotheses 2000s, when the World Health Organization (2000) rec- in this debate have been mostly tested by comparing LE ommended going beyond group-based mean compari- with HE indicators (i.e., inspecting trends in average years sons [22, 23] and incorporate individual-based data in of life vis-à-vis trends in average years in good health [1, the analysis of health inequalities [24, 25]. Since the turn 2, 7]). Yet, the original formulation of the compression of the century, the concept of inter-individual inequal- of morbidity hypothesis [28]—which was stated in terms ity has gained traction, thus favoring the spread of a like ‘compression into a shorter span between the age of bourgeoning literature on lifespan inequality. Yet, these disability onset and death’, or ‘rectangularization of the contributions are based, implicitly or explicitly, on the morbidity curve’ – naturally lends itself into an inspec- assumption that ‘longer lives are normatively preferable tion of HLI trends to test its validity. There are compel- to shorter ones’ – a gross simplification that greatly facili- ling reasons to believe that going beyond such ‘averages tates the construction of the corresponding LI measures comparison’ by taking into consideration the entire dis- but that might be hardly tenable in many circumstances. tribution of ages at which death and diseases occur can Confronted with the choice between ‘a prolonged yet throw considerable light into a long-lasting debate with unhealthy life’ and ‘a shorter but fully healthy life’, it is crucial implications for understanding the development not clear that the former would be universally chosen of human health and the performance of health systems. in favor of the latter [26, 27]. The inclusion of any year Permanyer et al. Population Health Metrics (2022) 20:1 Page 8 of 9 This study has several limitations. First, our method morbid conditions occur above age 85. Despite the limi- to estimate healthy lifespan distributions is based on tations of our dataset, a clear education gradient emerges simplistic and somewhat unrealistic assumptions. Fol- and the values of traditional LE, HE and LI indicators go lowing our approach (in which the individuals of a fic- in the expected direction. Thus, the quality of the dataset titious cohort are subject to the current mortality and should be good enough to illustrate the usefulness of the morbidity conditions throughout their lifetimes), we are new HLI indicators proposed in the paper. implicitly assuming that there is no possibility of recov- ery from disease/disability, and that the risk of mortality Conclusions is the same for healthy and unhealthy individuals. These The previous limitations notwithstanding, the results are exactly the same limitations assailing the Sullivan of this study suggest that ‘healthy lifespan inequality’ is method [5] that is commonly used to estimate HE indi- a substantively meaningful concept, which can be eas- cators. Despite these shortcomings, the method has suc- ily implemented in practice. The suggested approach is ceeded in becoming the workhorse of ‘health expectancy an invitation to go beyond country-level averages and studies’ owing to its simplicity and applicability in a wide explore not only the mean number of years individuals range of geographical, temporal and conceptual settings are expected to live in good health, but also the patterns [1]. Some studies have shown that, under mild regular- in which these healthy years are distributed among them. ity conditions, Sullivan’s method is generally acceptable Future research could determine whether the socio-eco- for monitoring long-term trends in HE [30]. In future nomic health patterning we observe in contemporary research, it would be desirable to apply other methods Spain also extends to other countries around the world. to estimate healthy lifespan distributions more realisti- In addition, exploring whether the factors that drive cally. Some of these methods rely on longitudinal data- changes in HLI indicators are the same as those influenc- sets that allow tracking individuals’ health trajectories ing the behavior of standard LE, HE and LI indicators can over time and calculate transition probabilities across be a fruitful avenue of research to improve our under- several health states (e.g. multistate life table or Markov standing of contemporary health dynamics. chain techniques [1]). Unfortunately, the relative scarcity The HLI indicators uncover new layers of health ine- of longitudinal datasets across space and time limits the quality that are not observable with currently existing empirical applicability of such sophisticated methods. methods. In Spain, low-educated individuals tend to not Second, our approach to measure less-than-good only die earlier and spend a shorter portion of their lives health could be criticized on grounds of arbitrariness. in good health than their highly educated counterparts, ‘Health’ is a multidimensional and fuzzy concept whose but also face greater variation in the eventual time of measurement can be operationalized using other pos- death and in the age at which they cease enjoying good sible health outcomes. Additionally, there are several health—a multiple burden of inequality that should be techniques to measure how such outcomes contribute to taken into consideration when evaluating the perfor- healthy/unhealthy life-years, using either dichotomous, mance of public health systems and in the elaboration of ordinal or continuous scales [1, 2]. These well-known realistic working-life extension plans and the design of challenges have already been encountered by previous equitable pension reforms. attempts to measure HE and compare its values across countries/over time [1] so they are not exclusive to the measurement of HLI indicators alone. To address them, Abbreviations LE: Life expectancy; HE: Health expectancy; LI: Lifespan inequality; HLI: Healthy it is important to use consistent definitions when mak- lifespan inequality; INE: Spanish statistical office; SNHS: Spanish national health ing comparisons and, if feasible, use different health indi- surveys; GALI: Global activity limitation index. cators to check the robustness of results. In our setting, the use of alternative conceptualizations of the concept of Supplementary Information ‘disability’ besides the standard GALI indicator (i.e. self- The online version contains supplementary material available at https://​doi.​ perceived health or the ability to perform certain physical org/​10.​1186/​s12963-​021-​00279-8. tasks) does not alter the substantive findings of the paper (see Additional File 1: Tables S3 and S4). Additional file 1: Supplementary Material with robustness checks. Third, our analyses are restricted to the ages rang- ing between 35 and 85 because of data limitation con- Acknowledgements straints. In all likelihood, the differences between women Not applicable. and men and across education groups would be altered if Authors’ contributions one considered all the ages above 35. Spain is among the I.P. conceived, designed and directed the project, wrote and reviewed the world’s most longevous countries and many deaths and manuscript, conducted analyses and interpreted the results. J.S. interpreted Permanyer et al. Population Health Metrics (2022) 20:1 Page 9 of 9 the results and revised the manuscript. A.B. collected data and performed 13. Cambois E, Robine J-M, Hayward MD. Social inequalities in disability-free calculations. All authors read and approved the final manuscript. life expectancy in the French male population, 1980–1991. Demography. 2001;38(4):513–24. Funding 14. Matthews RJ, Jagger C, Hancock RM. Does socio-economic advantage The research leading to these results has received funding from the European lead to a longer, healthier old age? Soc Sci Med. 2006;62(10):2489–99. Research Council (ERC-2019-CoG-864616, HEALIN project) and the Spanish 15. Bronnum-Hansen H, Baadsgaard M. Widening social inequality in life Ministry of Economy and Competitiveness “Ramón y Cajal” Research Grant expectancy in Denmark. A register-based study on social composition Program (RYC-2013–14196 and RYC-2013–14851). and mortality trends for the Danish population. BMC Public Health 2012; 12: 994. Availability of data and materials 16. Permanyer I, Spijker J, Blanes A, Renteria E. Longevity and lifespan vari- The data analyzed in this paper is available from the Spanish Statistical Office ation by educational attainment in Spain: 1960–2015. Demography. (INE; see https://​www.​ine.​es). 2018;55(6):2045–70. 17. Sasson I. Trends in life expectancy and lifespan variation by educational attainment: United States, 1990–2010. Demography. 2016;53(2):269–93. Declarations 18. Seaman R, Riffe T, Leyland AH, Popham F, van Raalte A. The increasing lifespan variation gradient by area-level deprivation: a decomposition Ethics approval and consent to participate analysis of Scotland 1981–2011. Soc Sci Med. 2019;230:147–57. The results presented in this paper are based on the analysis of secondary 19. Van Raalte AA, Kunst AE, Lundberg O, et al. The contribution of educa- data (Spanish National Health Surveys), which has been properly anonymized. tional inequalities to lifespan variation. Popul Health Metrics. 2012;10(1):3. Thus, the ethics approval is not necessary. 20. Van Raalte AA, Martikainen P, Myrskylä M. Lifespan variation by occu- pational class: compression or stagnation over time? Demography. Consent for publication 2014;51(1):73–95. Not applicable. 21. Van Oyen H, Nusselder W, Jagger C, Kolip P, Cambois E, Robine J-M. Gender differences in healthy life years within the EU: an exploration of Competing interests the “health–survival” paradox. Int J Public Health. 2013;58(1):143–55. The authors declare no competing interests. 22. Braveman P, Krieger N, Lynch J. Health inequalities and social inequalities in health. Bull World Health Organ. 2000;78(2):232–5. Author details 23. Braveman P, Starfield B, Geiger HJ. World Health Report 2000: how it 1 Centre d’Estudis Demogràfics, Carrer de Ca n’Altayó, Edifici E‑2, Campus de removes equity from the agenda for public health monitoring and policy. la UAB, 08193 Cerdanyola del Vallès, Spain. 2 ICREA Research Professor. ICREA, BMJ Br Med J. 2001;323(7314):678. Passeig Lluis Companys 23, 08010 Barcelona del Vallès, Spain. 24. Gakidou EE, Murray CJ, Frenk J. Defining and measuring health inequality: an approach based on the distribution of health expectancy. Bull World Received: 24 March 2020 Accepted: 16 December 2021 Health Organ. 2000;78:42–54. 25. Murray CJ. Commentary: comprehensive approaches are needed for full understanding. BMJ. 2001;323(7314):680–1. 26. Gerstorf D, Ram N, Röcke C, Lindenberger U, Smith J. Decline in life sat- isfaction in old age: longitudinal evidence for links to distance-to-death. References Psychol Aging. 2008;23(1):154. 1. Robine JM, Jagger C, Mathers CD, Crimmins EM, Suzman RM. Determin- 27. Lawton MP, Moss M, Hoffman C, Grant R, Have TT, Kleban MH. Health, ing health expectancies. West Sussex: Wiley; 2003. valuation of life, and the wish to live. Gerontologist. 1999;39(4):406–16. 2. Salomon JA, Wang H, Freeman MK, et al. Healthy life expectancy for 187 28. Fries JF. Aging, natural death, and the compression of morbidity. N Engl J countries, 1990–2010: a systematic analysis for the Global Burden Disease Med. 1980;303(3):130–5. Study 2010. Lancet. 2012;380(9859):2144–62. 29. Manton KG. Changing concepts of morbidity and mortality in the elderly 3. Smits J, Monden C. Length of life inequality around the globe. Soc Sci population. Milbank Mem Fund Quart Health Soc. 1982;60:183–244. Med. 2009;68(6):1114–23. 30. Mathers CD, Robine J-M. How good is Sullivan’s method for monitor- 4. Van Raalte AA, Sasson I, Martikainen P. The case for monitoring life-span ing changes in population health expectancies? J Epidemiol Commun inequality. Science. 2018;362(6418):1002–4. Health. 1997;51(1):80–6. 5. Sullivan DF. A single index of mortality and morbidity. HSMHA Health Rep. 1971;86(4):347. 6. Jagger C, Gillies C, Cambois E, et al. The Global Activity Limitation Index Publisher’s Note measured function and disability similarly across European countries. J Springer Nature remains neutral with regard to jurisdictional claims in pub- Clin Epidemiol. 2010;63(8):892–9. lished maps and institutional affiliations. 7. Jagger C, Gillies C, Moscone F, et al. Inequalities in healthy life years in the 25 countries of the European Union in 2005: a cross-national meta- regression analysis. Lancet. 2008;372(9656):2124–31. 8. Wrycza TF, Missov TI, Baudisch A. Quantifying the shape of aging. PloS one 2015; 10(3): e0119163. 9. Wilmoth JR, Horiuchi S. Rectangularization revisited: Variability of age at Ready to submit your research ? Choose BMC and benefit from: death within human populations. Demography. 1999;36(4):475–95. 10. INE. Estadística de Defunciones. Asignación de nivel educativo a ficheros • fast, convenient online submission de defunciones de 2017. Método de obtención y advertencias a usuarios • thorough peer review by experienced researchers in your field [Death statistics. Assignment of educational level to the 2017 death files. • rapid publication on acceptance Obtaining method and warnings to users]. Madrid: INE (Instituto Nacional de Estadistica), 2018. http://​www.​ine.​es/​metod​ologia/​t20/​t2030​306_​ • support for research data, including large and complex data types nived​uc.​pdf. • gold Open Access which fosters wider collaboration and increased citations 11. INE (Instituto Nacional de Estadistica). National Health Survey 2017. • maximum visibility for your research: over 100M website views per year Madrid: INE, 2017. https://​www.​ine.​es/​en/​metod​ologia/​t15/​t1530​41917_​ en.​pdf. At BMC, research is always in progress. 12. Crimmins EM, Saito Y. Trends in healthy life expectancy in the United States, 1970–1990: gender, racial, and educational differences. Soc Sci Learn more biomedcentral.com/submissions Med. 2001;52(11):1629–41.