Artículo de revista: Permanyer, Iñaki; Villavicencio, Francisco; Trias-Llimós, Sergi (2023) «Healthy lifespan inequality: morbidity compression from a global perspective». European Journal of Epidemiology, 38: 511-521 (ISSN 1573- 7284) doi: 10.1007/s10654-023-00989-3 European Journal of Epidemiology (2023) 38:511–521 https://doi.org/10.1007/s10654-023-00989-3 MORTALITY Healthy lifespan inequality: morbidity compression from a global perspective Iñaki Permanyer1,2 · Francisco Villavicencio3,4,5 · Sergi Trias‑Llimós1 Received: 5 December 2022 / Accepted: 7 March 2023 / Published online: 7 April 2023 © The Author(s) 2023 Abstract Current measures of population health lack indicators capturing the variability in age-at-morbidity onset, an important marker to assess the timing patterns of individuals’ health deterioration and evaluate the compression of morbidity. We provide global, regional, and national estimates of the variability in morbidity onset from 1990 to 2019 using indicators of healthy lifespan inequality (HLI). Using data from the Global Burden of Disease Study 2019, we reconstruct age-at-death distributions to calculate lifespan inequality (LI), and age-at-morbidity onset distributions to calculate HLI. We measure LI and HLI with the standard deviation. Between 1990 and 2019, global HLI decreased from 24.74 years to 21.92, and has been decreasing in all regions except in high-income countries, where it has remained stable. Countries with high HLI are more present in sub-Saharan Africa and south Asia, whereas low HLI values are predominant in high-income countries and central and eastern Europe. HLI tends to be higher for females than for males, and HLI tends to be higher than LI. Globally, between 1990 and 2019 HLI at age 65 increased from 6.83 years to 7.44 for females, and from 6.23 to 6.96 for males. Improvements in longevity are not necessarily accompanied by further reductions in HLI among longevity vanguard countries. Morbidity is compressing, except in high-income countries, where it stagnates. The variability in the ages at morbidity onset tends to be larger than the variability in lifespans, and such divergence broadens over time. As longevity increases worldwide, the locus of health inequality is moving from death-related inequalities to disease- and disability-centered ones. Keywords Ageing · Health inequalities · Longevity · Morbidity · Population health indicators Introduction average number of years individuals are expected to live – attempts to reduce disease incidence and disability rates The ageing process that is unfolding across countries all over have been less successful [5]. In an ageing world, there are the world is an unprecedented landmark in human history: increasing concerns as to whether the extra years of life are individuals worldwide can now expect to survive to ages spent in ‘good’ or in ‘less-than-good’ health [6], and there is that were deemed unattainable only a few decades ago [1–4]. a growing consensus that interventions to improve popula- However, despite the increases in life expectancy (LE) – the tion health should not only focus attention on the quantity of years that individuals live, but also on their quality [7]. * Iñaki Permanyer Surprisingly, current measures of population health have ipermanyer@ced.uab.es paid insufficient attention to the patterns of individuals’ health deterioration. While there are well-known indica- 1 Centre for Demographic Studies, Centres de Recerca de tors assessing how long are we expected to live in ‘good Catalunya (CERCA), Universitat Autònoma de Barcelona, 08193 Bellaterra, Spain health’ on average – many health-adjusted life expectancy 2 (HALE) indicators have been introduced in the literature ICREA, Passeig Lluís Companys 23, 08010 Barcelona, Spain for that purpose [6] – there is virtually no information about 3 Department of Economic, Financial and Actuarial the variability in the ages at which individuals cease to be Mathematics, University of Barcelona, Barcelona, Spain in good health. For example, a population in which all its 4 Interdisciplinary Centre on Population Dynamics, University members cease to be in good health at exactly age 65 is of Southern Denmark, Odense, Denmark substantially different from a population in which 50% of its 5 Department of International Health, Johns Hopkins members cease to be in good health at age 55 and the other Bloomberg School of Public Health, Baltimore, MD, USA 13 Vol.:(0123456789) 512 I. Permanyer et al. 50% at age 75 – a critical information for policy makers. timing of health deterioration [9, 10]. At the macro level, The seminal work by Fries on the ‘compression of morbid- HLI indicators capture the heterogeneous characteristics of ity’ theory suggests that “morbidity is compressed into the ageing populations, and provide crucial information for the shorter span between the increasing age-at-morbidity onset design of social care and health provision programs. and the fixed occurrence of death” (p 133), and that “[t]he rectangularization of the survival curve may be followed by rectangularization of the morbidity curve and by compres- sion of morbidity” (p 135) [8]. This theory naturally lends Methods to the study of the variability in the ages-at-morbidity onset – a fundamental dimension of health that has been largely Overview overlooked in the literature [9, 10]. We contribute to the ‘compression vs expansion of mor- We provide annual estimates on healthy lifespan inequal- bidity’ debate by using the recently proposed healthy lifes- ity (HLI) for the overall population and above age 65 from pan inequality (HLI) indicators that measure the variability 1990 to 2019 for 204 countries and territories using data in healthy lifespans across individuals [10]. We estimate, for from the Global Burden of Disease (GBD) Study 2019 [5, the first time, global, regional, and national levels and trends 13]. We also provide global and regional estimates using the in HLI from 1990 to 2019, and compare the new indicator same regional classification as GBD, which comprises seven with currently existing population health measures such as super-regions and 19 regions (Supplementary Information LE, HALE and lifespan inequality (LI) [11]. There are sev- [SI] Appendix 1). If no confusion arises, in the following eral reasons why HLI can be considered a key indicator in we will refer to countries and territories as ‘countries’, and health research that should be reported alongside traditional to super-regions and regions as ‘regions’. To obtain the dif- mortality and morbidity summary measures [10]. Improving ferent population health indicators, we first reconstructed population health goes beyond delaying death, and current ‘mortality curves’ (commonly referred to as ‘survival societies are increasingly concerned not only with the rising curves’) based on life table data (Fig. 1 panel A). Next, we prevalence of disease or disability, but also with rising health reconstructed ‘morbidity curves’ using age-specific remain- inequalities [12]. HLI indicators capture the heterogeneity ing HALE estimates from GBD (Fig. 1 panel B). Morbidity in the underlying distribution of population health and can curves measure survival probabilities in good health. Finally, have important implications at the micro and macro levels. from the mortality and morbidity curves we obtained the From the individuals’ perspective, HLI indicators can be corresponding age-at-death (Fig. 1 panel C) and age-at- used to study the variability in the ages-at-morbidity onset, morbidity onset (Fig. 1 panel D) distributions, from which a fundamental concept that measures the uncertainty in the we calculated LI and the new HLI indicators [10]. Fig. 1  Diagram of key mortal- ity-related (life expectancy and lifespan inequality) and morbid- ity-related (health-adjusted life expectancy and healthy lifespan inequality) indicators 13 Healthy lifespan inequality: morbidity compression from a global perspective 513 Data curves become increasingly rectangular over time. To do so, we followed Wilmoth and Horiuchi [18] and meas- The Global Burden of Disease Study 2019 reports global, ured whether the age-at-morbidity onset distributions regional, and national estimates on age- and sex-specific ( n dx∗) become more or less concentrated (i.e., whether HLI probabilities of dying and age- and sex-specific remaining increases or decreases over time). LE in 5-year age groups from 1950 to 2019 [14]. These data can be used to reconstruct mortality curves for all region- and country-years using standard life table tech- Uncertainty analysis niques. Recently, GBD has also been publishing estimates of age- and sex-specific HALE for all countries from 1990 We assessed the uncertainty of the LI and HLI estimates to 2019 [5, 13]. These estimates are obtained using mod- based on the uncertainty of the input data from GBD. els that incorporate estimates of years lived with disability, Uncertainty was obtained by sampling from the correspond- life tables, and standard demographic methods [5] – but the ing uncertainty intervals (UIs) of LE, death probabilities, underlying morbidity curves remain unknown. and HALE reported by GBD on Monte Carlo simulations (SI Appendix 3.3), adapting an analogous approach used Reconstructing mortality and morbidity curves elsewhere [19]. We report 80% UIs rather than 95% UIs because of the substantial uncertainty inherent in HALE Using data on age-specific probabilities of dying and age- estimates [13, 19]. Intervals based on higher uncertainty specific remaining LE, for each country/region, year, and sex levels would not constitute useful and meaningful summary we calculated the age-specific average person-years lived measures. in each age interval by individuals dying in that interval We followed the GATHER guidelines for global health (the n ax values) and the remaining columns of the life table, estimates and included the GATHER checklist for transpar- including the mortality/survival curves 𝓁x (Fig. 1 panel A). ency and replicability (SI Appendix 4). We carried out all Next, we used the n ax values from each setting, in combina- our analyses using the open-source statistical software R tion with the age-specific HALE estimates, to reconstruct (version 4.1.1) [20]. To reduce computing time we imple- the corresponding morbidity curves 𝓁x∗ (Fig. 1 panel B). mented parallel processing using the R package ‘doParallel’ Additional details on the reconstruction of mortality and (version 1.0.17) [21]. morbidity curves are found in the SI (Appendix 2). Calculating inequality Results For each country/region, year, and sex we took first differ- ences from the mortality and morbidity curves to derive the Levels and trends in healthy lifespan inequality corresponding age-at-death (n dx = 𝓁x − 𝓁x+n, Fig. 1 panel C) (HLI) and age-at-morbidity onset (n dx∗ = 𝓁x∗ − 𝓁x+n ∗ , Fig. 1 panel D) distributions (SI Appendix 2). From these distributions, we In 2019, the regions with the lowest HLI values are southeast measured the corresponding levels of LI and HLI using the Asia, east Asia and Oceania, central Europe, eastern Europe standard deviation (SI Appendix 3.1), an indicator measur- and central Asia, and high-income countries; at the oppo- ing the spread of a distribution that has been widely used site end, the highest HLI is observed in sub-Saharan Africa as a measure of lifespan inequality [15, 16]. Our findings and south Asia (Table 1). The countries with the highest proved robust when compared to those obtained using other HLI values (above 22 years) are mostly concentrated in sub- popular inequality measures, such as the coefficient of vari- Saharan Africa (for instance Mali, Niger, Nigeria, Burkina ation or the Gini index given the strong correlation among Faso, and Chad), but high HLI values are also observed in them (SI Appendix 3.2). When the indicators proposed in Haiti, Pakistan, and Yemen (Fig. 2). The mid-range of the the paper are computed for ages above 65, we add the num- HLI distribution (19−22 years) is mainly represented by ber 65 to the acronyms used to refer to them (i.e., LE65, countries from Latin America and Caribbean like Brazil, HALE65, LI65 and HLI65). from south Asia (India and Bangladesh), from sub-Saharan Africa (Rwanda, Kenya, and Congo), and from north Africa Measuring morbidity compression and Middle East (Sudan and Afghanistan). At the bottom of the distribution, the lowest HLI values (below 19) are To assess whether morbidity is ‘expanding’ [17] or ‘com- mostly found in high-income countries, but also in China pressing’ [8], we adopted the approach proposed by Fries and central and eastern Europe (Fig. 2). [8], who suggested evaluating the extent to which morbidity 13 514 I. Permanyer et al. Fig. 2  World map with national healthy lifespan inequality levels in 2019, both sexes combined. Source: Authors’ elaboration based on data from GBD [5, 13] Between 1990 and 2019, global HLI levels declined intervals, are available on GitHub (see SI Appendix 5 for from 25.09 years (80% uncertainty interval [UI] details). 21.50−28.46) to 22.10 (17.61−26.54) for females, from 24.32 (21.51−27.09) to 21.65 (17.83−25.36) for HLI and other population health measures males, and from 24.74 years (21.59−27.74) to 21.92 (17.79−25.80) for both sexes combined. Steady declines A visual inspection of the association between LI and HLI have also been observed across all world regions except reveals that these two measures are positively correlated, for high-income countries, which have remained stable and their values follow clear temporal and geographical pat- between 1990 and 2019 with HLI levels hovering around terns (Fig. 4). Both HLI and LI tend to decline over time, 19.3 years for females and 18.4 for males. In most regions, even though the latter declines faster. As we approach 2019, HLI levels are larger for females than for males (e.g., sub- for most country-year observations the values of HLI are Saharan Africa, south Asia, and central Europe, eastern larger than those of LI, being sub-Saharan African countries Europe and central Asia), except for Latin America and the only exception. This is especially noticeable when ana- the Caribbean (Fig. 3 panels A and B). lyzing global and regional trends in the ratio between HLI For ages above 65, we find increasing HLI65 trends and LI (SI Appendix 6): increasing trends in all regions are across the board (Fig. 3 panels C and D). Globally, observed but point estimates in sub-Saharan Africa remain HLI65 goes from 6.83 years (80% UI 6.40−7.28) in 1990 below one for the entire 1990–2019 period. In high-income to 7.44 (6.91−8.03) in 2019 for females, and from 6.23 countries (located at the lower end of the cloud of data points (5.93−6.58) to 6.96 (6.56−7.41) for males. Both for in Fig. 4), the HLI/LI ratio reaches 1.36 (80% UI 1.05–1.58) females and males, the highest HLI65 levels are observed for females, meaning that HLI levels are 36% higher than for Latin America and Caribbean and high-income coun- those of LI (SI Appendix 6). Worldwide, in 2019 HLI is tries. At the other end, south Asia and sub-Saharan Africa higher for females than for males in 153 (75.0%) of the 204 stand out as the regions with lowest HLI65 levels. Once countries and territories analyzed, in contrast with LI, that is again, HLI65 levels tend to be higher for females than higher for males in 183 (89.7%) of the cases. HLI is higher for males. All the global, regional, and national LI and than LI in 321 (78.7%) of the 408 country-sex combinations. HLI estimates, and the corresponding 80% uncertainty 13 Healthy lifespan inequality: morbidity compression from a global perspective 515 Table 1  Global and regional estimates of healthy lifespan inequality by sex in 2019. Healthy lifespan inequality measured as the standard devia- tion of the corresponding age-at-morbidity onset distribution (values between parenthesis denote 80% uncertainty intervals) Females Males Both sexes Global 22.10 (17.61−26.54) 21.65 (17.83−25.36) 21.92 (17.79−25.80) Central Europe, eastern Europe and central Asia 18.83 (14.64−22.01) 18.57 (15.28−21.35) 18.97 (15.37−22.17) Central Asia 19.20 (15.17−22.91) 18.87 (15.48−22.26) 19.17 (15.46−22.69) Central Europe 17.70 (13.59−20.65) 17.16 (13.51−19.82) 17.60 (13.88−20.40) Eastern Europe 18.43 (14.43−21.32) 18.25 (15.09−20.87) 18.69 (15.24−21.31) High-income countries 19.35 (14.96−22.46) 18.46 (14.63−21.52) 18.95 (14.86−21.91) Australasia 19.15 (14.64−22.22) 18.54 (14.40−21.75) 18.87 (14.55−21.98) High-income Asia Pacific 18.55 (14.03−21.65) 17.39 (13.68−20.56) 18.05 (14.03−21.23) High-income North America 19.59 (15.27−22.59) 18.99 (15.18−21.91) 19.33 (15.22−22.34) Southern Latin America 19.21 (15.01−22.41) 18.70 (15.01−21.96) 19.04 (15.06−22.44) Western Europe 18.89 (14.47−22.03) 17.72 (13.92−20.73) 18.35 (14.27−21.45) Latin America and Caribbean 20.62 (16.13−24.36) 20.87 (17.07−24.45) 20.82 (16.68−24.45) Andean Latin America 20.50 (15.76−24.32) 20.64 (16.08−24.70) 20.59 (15.89−24.37) Caribbean 22.67 (18.26−26.77) 22.50 (18.35−26.26) 22.62 (18.47−26.63) Central Latin America 20.01 (15.57−23.26) 20.51 (16.55−24.00) 20.31 (16.18−23.88) Tropical Latin America 20.94 (16.52−24.95) 21.16 (17.47−24.50) 21.16 (17.14−25.01) North Africa and Middle East 20.32 (15.90−24.11) 20.24 (16.18−24.08) 20.28 (16.09−24.10) South Asia 21.78 (17.30−25.85) 21.24 (17.08−25.09) 21.50 (17.37−25.24) Southeast Asia, east Asia and Oceania 18.64 (14.10−22.16) 18.57 (14.60−22.15) 18.66 (14.53−22.10) East Asia 17.49 (12.97−20.63) 17.33 (13.29−20.70) 17.47 (13.41−20.59) Oceania 21.58 (17.19−25.53) 21.42 (17.40−25.16) 21.49 (17.29−25.43) Southeast Asia 19.69 (15.38−23.39) 19.73 (15.95−23.34) 19.79 (15.78−23.38) Sub-Saharan Africa 23.96 (20.39−27.47) 23.76 (20.77−26.66) 23.89 (20.71−27.12) Central sub-Saharan Africa 22.85 (18.83−26.67) 22.44 (18.68−25.76) 22.70 (18.91−26.54) Eastern sub-Saharan Africa 22.97 (19.26−26.56) 22.70 (19.54−25.85) 22.86 (19.41−26.11) Southern sub-Saharan Africa 22.52 (19.13−25.75) 21.74 (19.02−24.43) 22.24 (19.19−25.26) Western sub-Saharan Africa 25.10 (21.57−28.64) 25.20 (22.05−28.18) 25.16 (21.89−28.34) The overall association between average health (as meas- For ages above 65 years, the association between average ured by LE and HALE) and health inequality (as measured health and health inequality reverses to positive across all by LI and HLI) is negative, showing that higher longevity world regions (Fig. 5 panels C and D). That is, when con- and healthy longevity tend to be associated with lower vari- ditioning upon survival to 65, higher average longevity is ability in lifespans and healthy lifespans within countries, associated with more variability in ages-at-death, and higher respectively (Fig. 5 panels A and B). While the relationship healthy longevity is associated with greater variability in the between LE and LI is strong and negative (Fig. 5 panel A, ages at which morbidity starts. While there is a great deal of the slope of the linear trend is − 0.43 with R2 = 0.747), it is cross-country variability in the relationship between LE65 weaker and flatter between HALE and HLI (Fig. 5 panel B, and LI65 (Fig. 5 panel C), the association between HALE65 slope of − 0.28 and R2 = 0.590). The negative relationship and HLI65 is much tighter (Fig. 5 panel D). In addition, between the later tends to disappear (or even reverse to posi- LI65 levels tend to be larger than those of HLI65, but the tive) when the values of HLI hover around 18, as it is the latter increase at a faster pace. case in high-income countries and central Europe, eastern Europe and central Asia. That is, despite continued improve- ments in HALE, HLI levels seem to have reached a lower bound where no further reductions (or indeed increases) are observed – a pattern that does not occur when comparing LE and LI. 13 516 I. Permanyer et al. Fig. 3  Global and regional time trends in healthy lifespan inequality by sex, for all ages (panels A and B) and for ages above 65 (panels C and D) from 1990 to 2019. Source: Authors’ elaboration based on data from GBD [5, 13] Discussion stagnant – and even increasing slightly since 2000. At ages above 65 years, HLI65 increased over the study Summary period in all seven super-regions. Importantly, the values of HLI tend to be substantially larger than those of LI in Healthy lifespan inequality (HLI) indicators measure the all regions except for sub-Saharan Africa, and such dif- variability in age-at-morbidity onset, so they can be used to ferences increase over time. Higher values of LE tend to assess the time patterns of individuals’ health deterioration. be associated with lower values of LI, but the relationship This study documents, for the first time, the levels and between HALE and HLI is considerably weaker (espe- trends in HLI indicators across regions and countries all cially among longevity vanguard countries). In general, over the world for a period of 30 years, from 1990 to males tend to exhibit higher levels of LI than females, but 2019. Global HLI declined from 24.74 years (80% UI the opposite is observed with HLI. 21.59−27.74) in 1990 to 21.92 (17.79−25.80) in 2019. Generalized declines are observed across regions, except for high-income countries, where HLI has remained 13 Healthy lifespan inequality: morbidity compression from a global perspective 517 Fig. 4  Relationship between healthy lifespan inequality and lifespan inequality by sex between 1990 and 2019. For each region, the lightest color corresponds to 1990 and darkness increases over years up to 2019. Source: Authors’ elaboration based on data from GBD [5, 13, 14] Interpretation wider than the range of ages in which most individuals die – a key finding with fundamental implications for planners Our results suggest that, while the variability in the ages at aiming at improving population health and reducing health which morbidity starts has decreased in most high-mortality inequalities. For instance: retirement or health care policies countries, it has remained constant or has even increased exclusively based on (increasing) trends in life expectancy for an ever-growing number of countries, especially among might miss the mark and have deleterious social conse- the longevity vanguards. In those countries, even if age at quences if the corresponding morbidity onset distribution death is retreating and becoming increasingly concentrated widens over time. Indeed, HLI levels are estimated to be up at older ages [11, 15, 22–24], it is less clear whether mor- to 36% higher than LI among women in high-income coun- bidity onset is also being shifted towards older ages [5, 25, tries. The only exception to this general pattern is found in 26]. In general, improvements in reducing mortality rates sub-Saharan Africa, where LI levels are slightly higher than for most causes have not been matched by similar declines those of HLI – even though they become increasingly simi- in disability rates, which have either been stagnant or have lar over time (see Fig. 4 and SI Appendix 6). This might be increased for several causes, like diabetes or mental and sub- attributable to the high levels of child mortality in the region stance use disorders [7]. Our findings cohere with previous [27] (which generate bimodal age-at-death distributions with studies exploring mortality and morbidity dynamics from a high levels of lifespan inequality) together with the fact that global perspective – while expanding and complementing the incidence of morbidity is relatively low at those younger those analyses in several directions. ages. Further research is needed to examine what are the In this paper, we find that the variability in the ages at specific age groups and causes of death, diseases or condi- which morbidity starts (HLI) can be much larger than the tions driving the trends we document. variability in the ages at death (LI), and such difference The strong and negative association between country-spe- broadens over time. When this happens, the range of ages cific LE and LI estimates (Fig. 5 panel A) has been widely in which most individuals’ health deteriorates becomes documented [22–24, 28]. It suggests that the normatively 13 518 I. Permanyer et al. Fig. 5  Relationship between life expectancy - lifespan inequality B) and ages above 65 (panels C and D). Source: Authors’ elaboration and health-adjusted life expectancy - healthy lifespan inequality for based on data from GBD [5, 13, 14] females between 1990 and 2019, considering all ages (panels A and desirable goals of increasing countries’ average longev- countries have made no further progress in reducing the ity and reducing age-at-death inequality can be achieved variability in healthy lifespans. Such stagnation might be the simultaneously. However, a different picture emerges when outcome of forces pushing in opposite directions. On the one inspecting the relationship between HALE and HLI (Fig. 5 hand, improvements in living standards or the promotion of panel B). There are many countries and regions experienc- healthier lifestyles (e.g., better diets, regular exercise, avoid- ing increases in HALE concomitantly with declines in HLI ing alcohol consumption or smoking) [29, 30] can postpone – thus suggesting that a ‘compression’ [8] of morbidity is the deterioration of individuals’ health. On the other hand, occurring in those places – (e.g., sub-Saharan Africa, south the implementation of prevention and screening programs Asia, north Africa and Middle East, or other regions at the that decrease the age at diagnosis of important diseases (e.g., bottom of the corresponding distributions with plenty of cancers or mental disorders) [31, 32] as well as persistent room for improvement). At the same time, there seems to and increasing socioeconomic inequalities [33] can contrib- be a threshold above which further gains in HALE are not ute to widen age-at-morbidity onset distributions. necessarily accompanied with HLI reductions. This is the These troubling trade-offs between health equality and case for a considerable number of countries – particularly efficiency become even more pronounced when inspecting those from central Europe, eastern Europe and central Asia trends at ages above 65 years. Our findings indicate that and, specially, high-income countries – where it is not clear countries’ overall success in increasing LE65 and HALE65 whether morbidity is ‘expanding’ [17] or ‘compressing’ [8]. inevitably comes at a cost: a simultaneous increase in both Thus, while high-mortality countries have been generally LI65 and HLI65, indicating greater heterogeneity among the successful in increasing (healthy) longevity and simultane- elder population. These results cohere with recent studies ously reducing (healthy) lifespan inequality, low-mortality reporting increasing trends in lifespan inequality at age 75 13 Healthy lifespan inequality: morbidity compression from a global perspective 519 and a positive relationship between LE65 and LI65 [23], and in the ‘health-survival paradox’ [6, 38], but that they tend suggest that the variability in morbidity onset at older ages to face greater uncertainty than men regarding the ages at (HLI65) is going in the same direction. Indeed, Seaman et al. morbidity onset. Further research on the determinants of the also find that the variability in the age at morbidity onset in increasing sex-differences in HLI is also needed. Denmark (as measured by the age at first hospital admission among adults aged 60 and above) also tends to increase from Limitations the 1990s onwards [9]. Unfortunately, the approach followed in that study cannot be replicated at a global scale because of Our paper has several limitations. First, our estimates are its reliance on hospital admission data – which is difficult to exclusively based on period life tables. Unfortunately, lon- access and is not easily comparable across countries. Like- gitudinal methods can only be applied in a highly reduced wise, our findings on increasing health variability among the number of data-rich countries. Hence, we rely upon the elder are in line with recent studies documenting an increase synthetic cohort approach in which individuals are subject in the diversification of the causes from which individuals to period-specific mortality and morbidity conditions along die in low-mortality countries – especially among individu- their lifetimes, something that is customarily used in the esti- als dying at ages above 50 [34]. Put together, these studies mation of LE, LI and HALE indicators. Second, the meth- posit that, as mortality is pushed towards increasingly higher ods used to generate age-at-morbidity onset distributions ages, the health profiles of the survivors are increasingly implicitly rely on the assumption that individuals cannot diverse (i.e., with an increasingly heterogeneous mix of recover from their ‘less-then-healthy’ status. While some- robust and frail individuals). what unrealistic, this is the simplifying assumption underly- The findings presented here consistently point towards a ing the Sullivan method [39], which under mild regularity compositional shift in health inequalities within countries: conditions is generally acceptable for monitoring long-term as longevity increases, morbidity-related inequalities are trends in HALE [40], and has been widely used for estimat- gaining prominence with respect to mortality-related ones. ing HALE indicators [6], also by GBD [5]. In addition, the That is, generalized improvements in living standards, medi- definition of HALE is exclusively based on the prevalence cal innovations, the spread of technological breakthroughs and severity of diseases and health conditions, but fails to and public health policies might have contributed to reduce take into consideration other, more holistic, dimensions of mortality-related inequalities across world countries. This health (like self-reported health, or pain and discomfort lev- is reflected in the decreasing inequalities in basic survival els) typically included in other approaches, like the EQ-5D indicators like life expectancy at birth, mostly driven by the measure of the EuroQol Group [41]. Third, the quality of reduction of infant and child mortality [5, 27]. However, the mortality and, specially, morbidity data varies consider- these improvements have in turn contributed to the emer- ably across countries (an issue that is partially attributable gence of new layers of morbidity-related inequalities among to the different sampling strategies followed to obtain health adults at older ages, often to the advantage of privileged information), which is reflected in the uncertainty intervals countries or socio-economic groups [10, 35]. Stated other- of our HLI estimates. All our analyses are based on levels wise: the same structural improvements that have contrib- and trends of the point estimates, but conclusions should be uted to increase the survival chances of the worse-off have formulated with caution given their uncertainty [19]. Lastly, in turn delayed the emergence of health-related inequalities the outbreak of Covid-19 pandemic has not been included in to older ages. This is yet another instance of the successive our analyses. A comprehensive study on the potential impact waves of convergence-divergence cycles in health stipulated that the pandemic might have on the population health indi- by the health transition theory suggested by Frenk et al. [36] cators investigated here is extremely important, but should and later adopted by Vallin and Meslé [37]. probably await better data. Furthermore, our results shed new light on the male- female health and mortality differences. Women tend to live longer, and the length of their lives are less unequally Conclusion distributed than men (Fig. 4) [6]. However, the female advantage is less obvious, or even disappears, when look- Previous studies investigating the ‘compression vs expan- ing at the length of healthy life. It is well known that the sex sion of morbidity’ debate have almost exclusively relied on gaps in HALE are considerably smaller than those in LE the comparison of indicators measuring average longevity [5, 6]. As regards healthy lifespan inequality, we find that (i.e., LE) against indicators measuring the average number in approximately three quarters of our country-year obser- of years spent in good health (HALE) [6, 25]. This ‘aver- vations female HLI levels are higher than those of males. age-based approach’ has many limitations, as it disregards These findings suggest that, not only women spend longer the patterns of health deterioration among individuals. As fractions of their lives in less-than-good health as postulated emphasized by Fries [8] more than four decades ago, the 13 520 I. Permanyer et al. analysis of variation (not of average values) is fundamental to all the data in the study and the corresponding author had final to investigate whether, and to what extent, the emergence of responsibility for the decision to submit for publication. diseases is postponed to increasingly older ages. Notwith- Data Availability The source code to replicate the analyses, input data, standing the limitations of our approach, this paper sheds and results are publicly available for research purposes on the GitHub new light into a longstanding debate with fundamental repository https://​github.​com/​panch​oVG/​HLI). implications for our understanding of contemporary health dynamics around the globe. Declarations Most attempts at explaining the drivers of population Conflict of interest The authors declare no conflict of interest. ageing are framed within the epidemiological transition theory [42] or some of its variations – like the fourth stage Ethics approval Not required. added to Omran’s initial theory [43–45] (the so-called Consent to participate Not required. “Age of delayed degenerative diseases”). Among the major critiques directed against it, scholars have highlighted its Consent to publish Not required. overemphasis on mortality (e.g., causes of death), thus giving insufficient attention to morbidity and quality of Open Access This article is licensed under a Creative Commons Attri- life [46]. The findings reported in this paper not only lend bution 4.0 International License, which permits use, sharing, adapta- support to the idea that prospective theories on population tion, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, ageing should give a more prominent role to morbidity, but provide a link to the Creative Commons licence, and indicate if changes also highlight the importance of moving beyond averages were made. The images or other third party material in this article are to study the age and sex patterns in which morbidity onset included in the article's Creative Commons licence, unless indicated affect individuals across populations. Whether further 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 delays in the ages at which individuals die pose a threat permitted by statutory regulation or exceeds the permitted use, you will to the sustainability of the health, pension, and welfare need to obtain permission directly from the copyright holder. To view a systems not only depends on the extent to which these copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/. extra years of life are lived free of disease, disability and/ or physical and mental impairment, but also on the timing patterns of individuals’ health deterioration. References The results reported in this study uncover a surprisingly overlooked layer of inequality that cannot be observed 1. Riley JC. Rising life expectancy: a global history. 1st ed. Cam- bridge University Press; 2001. with traditional population health measures such as LE, 2. Oeppen J, Vaupel JW. Broken limits to life expectancy. Science. HALE or LI, and have substantive implications. Inter alia, 2002;296:1029–31. they reveal that mortality inequalities within countries are 3. Christensen K, Doblhammer G, Rau R, et al. Ageing populations: becoming less prominent than its morbidity counterparts. the challenges ahead. Lancet. 2009;374:1196–208. 4. Vaupel JW, Villavicencio F, Bergeron-Boucher M-P. Demo- As longevity increases worldwide, the locus of health ine- graphic perspectives on the rise of longevity. Proc Natl Acad Sci. quality over the lifespan is gradually moving from death- 2021;118:e10254. related inequalities to disease- and disability-centered ones 5. Wang H, Abbas KM, Abbasifard M, et al. Global age-sex-specific – a compositional shift in health inequality that should be fertility, mortality, healthy life expectancy (HALE), and popu- lation estimates in 204 countries and territories, 1950–2019: a taken into consideration in the elaboration of prospective comprehensive demographic analysis for the Global Burden of public health policies. Disease Study 2019. The Lancet. 2020;396:1160–203. 6. Jagger C, Crimmins EM, Saitō Y, et al editors. International Supplementary Information The online version contains supplemen- handbook of health expectancies. Cham: Springer International tary material available at https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 10654-0​ 23-0​ 0989-3. Publishing; 2020. 7. Murray CJ, Barber RM, Foreman KJ, et al. Global, regional, and Author contributors IP conceptualized and designed the study. FV national disability-adjusted life years (DALYs) for 306 diseases analysed data and STL contributed to data analysis. FV and STL pro- and injuries and healthy life expectancy (HALE) for 188 coun- duced results and created the figures with input from IP. IP, FV and tries, 1990–2013: quantifying the epidemiological transition. The STL wrote the paper. Lancet. 2015;386:2145–91. 8. Fries JF. Aging, natural death, and the compression of morbidity. Funding This study is supported by the European Research Council N Engl J Med. 1980;303:130–5. (grant 2019-CoG-864616) and the Spanish Ministry of Science and 9. Seaman R, Höhn A, Lindahl-Jacobsen R, et al. Rethinking mor- Innovation R + D LONGHEALTH project (grant PID2021-128892OB- bidity compression. Eur J Epidemiol. 2020;35:381–8. I00). FV and STL acknowledge funding from the Spanish State 10. Permanyer I, Spijker J, Blanes A. On the measurement of healthy Research Agency under the Ramón y Cajal Programme (grants lifespan inequality. Popul Health Metr. 2022;20:1. RYC2021-033979‐I and RYC2021-033123-I, respectively). The funder 11. van Raalte AA, Sasson I, Martikainen P. The case for monitoring of the study had no role in study design, data collection, data analysis, life-span inequality. Science. 2018;362:1002–4. data interpretation, or writing of the report. All authors had full access 13 Healthy lifespan inequality: morbidity compression from a global perspective 521 12. Marmot MG. The health gap: the challenge of an unequal world. 30. Wensink M, Alvarez J-A, Rizzi S, et al. Progression of the smok- London, Sydney: Bloomsbury; 2015. ing epidemic in high-income regions and its effects on male- 13. Institute for Health Metrics and Evaluation. GBD results tool. female survival differences: a cohort-by-age analysis of 17 coun- Global Health Data Exchange. Available from: http://​ghdx.​healt​ tries. BMC Public Health. 2020;20:1–8. hdata.​org/​gbd-​resul​ts-​tool. Accessed 11 July 2022. 31. Rizzi S, Wensink M, Ahrenfeldt LJ, et al. Age-specific cancer 14. Global Burden of Disease Collaborative Network. Global Burden rates: a bird’s-eye view on progress. Ann Epidemiol. 2020;48:51– of Disease Study 2019 (GBD 2019) Life tables 1950–2019. Seat- 54.e1. tle, United States of America: Institute for Health Metrics and 32. Plana-Ripoll O, Momen NC, McGrath JJ, et al. Temporal changes Evaluation (IHME), 2020. Available from: http://g​ hdx.h​ ealth​ data.​ in sex- and age-specific incidence profiles of mental disorders— org/​record/​ihme-​data/​gbd-​2019-​life-​tables-​1950-​2019. Accessed A nationwide study from 1970 to 2016. Acta Psychiatr Scand. 11 July 2022. 2022;145:604–14. 15. Edwards RD, Tuljapurkar S. Inequality in life spans and a new 33. Mackenbach JP. Health inequalities: Persistence and change in perspective on mortality convergence across industrialized coun- modern welfare states. USA: Oxford University Press; 2019. tries. Popul Dev Rev. 2005;31:645–74. 34. Bergeron-Boucher M-P, Aburto JM, van Raalte A. Diversification 16. Engelman M, Canudas-Romo V, Agree EM. The implications of in causes of death in low-mortality countries: emerging patterns increased survivorship for mortality variation in aging popula- and implications. BMJ Glob Health. 2020;5:e002414. tions. Popul Dev Rev. 2010;36:511–39. 35. Conceiçao P (dir.). Human Development Report 2019: Beyond 17. Gruenberg EM. The failures of success. Milbank Mem Fund Q. income, beyond averages, beyond today: Inequalities in human 1977;55:3–24. development in the 21st century. New York: United Nations 18. Wilmoth JR, Horiuchi S. Rectangularization revisited: vari- Development Programme (UNDP); 2019. https://​hdr.​undp.​org/​ ability of age at death within human populations. Demography. conte​nt/​human-​devel​opment-​report-​2019 1999;36:475. 36. Frenk J, Bobadilla JL, Stern C, et al. Elements for a theory of the 19. Villavicencio F, Bergeron-Boucher M-P, Vaupel JW. Reply to health transition. Health Transit Rev. 1991;1:21–38. Permanyer et al.: The uncertainty surrounding healthy life expec- 37. Vallin J, Meslé F. Convergences and divergences in mortality: tancy indicators. Proc Natl Acad Sci. 2021;118:e2115544118. A new approach of health transition. Demogr Res. 2004;Special 20. R Core Team. R: A language and environment for statistical com- Collection 2:11–44. puting. R Foundation for Statistical Computing, 2021. https://​ 38. Oksuzyan A, Juel K, Vaupel JW, et al. Men: good health and high www.R-​proje​ct.​org/. mortality. Sex differences in health and aging. Aging Clin Exp 21. Microsoft Corporation and Steve Weston (2022). doParallel: Res. 2008;20:91–102. Foreach parallel adaptor for the ‘parallel’ package. R package 39. Sullivan DF. A single index of mortality and morbidity. HSMHA version 1.0.17. https://​CRAN.R-​proje​ct.​org/​packa​ge=​doPar​allel. Health Rep. 1971;86:347. 22. Smits J, Monden C. Length of life inequality around the globe. 40. Mathers CD, Robine J-M. How good is Sullivan’s method for Soc Sci Med. 2009;68:1114–23. monitoring changes in population health expectancies? J Epide- 23. Permanyer I, Scholl N. Global trends in lifespan inequality: 1950– miol Community Health. 1997;51:80–6. 2015. PLoS ONE. 2019;14:e0215742. 41. Rabin R, de Charro F. EQ-5D: a measure of health status from the 24. Vaupel JW, Zhang Z, van Raalte AA. Life expectancy and dispar- EuroQol Group. Ann Med. 2001;33:337–43. ity: an international comparison of life table data. BMJ Open. 42. Omran A. The epidemiologic transition: a theory of the epi- 2011;1:e000128. demiology of population change. Milbank Mem Fund Q. 25. Crimmins EM, Beltrán-Sánchez H. Mortality and morbidity 1971;49:509–38. trends: is there compression of morbidity? J Gerontol B Psychol 43. Olshansky SJ, Ault AB. The fourth stage of the epidemiologic Sci Soc Sci. 2011;66:75–86. transition: the age of delayed degenerative diseases. Milbank Q. 26. Permanyer I, Trias-Llimós S, Spijker JJA. Best-practice healthy 1986;64:355–91. life expectancy vs life expectancy: Catching up or lagging behind? 44. Rogers RG, Hackenberg R. Extending epidemiologic transition Proc Natl Acad Sci. 2021;118:e2115273118. theory: A new stage. Soc Biol. 1987;34:234–43. 27. Perin J, Mulick A, Yeung D, et al. Global, regional, and national 45. Omran AR. The epidemiologic transition theory revisited thirty causes of under-5 mortality in 2000–19: an updated systematic years later. World Health Stat Q. 1998;53:99–119. analysis with implications for the Sustainable Development Goals. 46. McKeown RE. The epidemiologic transition: Changing pat- Lancet Child Adolesc Health. 2022;6:106–15. terns of mortality and population dynamics. Am J Lifestyle Med. 28. Edwards RD. Changes in world inequality in length of life: 1970– 2009;3:19S–26S. 2000. Popul Dev Rev. 2011;37:499–528. 29. Janssen F, Trias-Llimós S, Kunst AE. The combined impact of Publisher’s Note Springer Nature remains neutral with regard to smoking, obesity and alcohol on life-expectancy trends in Europe. jurisdictional claims in published maps and institutional affiliations. Int J Epidemiol. 2021;50:931–41. 13