Domain-specific trends in healthy life expectancy among Chinese older adults: Compression, expansion, and dynamic equilibrium

Abstract

Background

Amid rising global life expectancy, the debate over whether population aging entails a “compression” or an “expansion” of morbidity remains unresolved. Previous evidence, drawn largely from Western nations and often reliant on cross-sectional data or single health indicators, has produced mixed findings. This study addresses this gap by investigating the health trajectory of older adults in China, examining whether it reflects morbidity compression or expansion across multiple health domains.

Methods

We analyzed 20 years of longitudinal data (1998–2018) from the Chinese Longitudinal Healthy Longevity Survey (CLHLS), comprising 93,581 observations of birth cohorts from 1895 to 1949. To overcome the limitations of prevalence-based measures, we employed continuous-time multi-state Markov models for interval-censored data. This approach enabled us to estimate cohort-specific transition probabilities (i.e., onset, recovery, and mortality) and compute Healthy Life Expectancy (HLE) across three distinct domains: cognitive function, physical independence (ADL), and major chronic diseases.

Results

After controlling for age and sex, we observed divergent health trajectories across successive birth cohorts. In the cognitive domain, a relative compression of morbidity emerged, characterized by a delayed onset of impairment and a reduced proportion of total life expectancy (TLE) spent with cognitive deficits. The functional (ADL) domain, however, exhibited a pattern of dynamic equilibrium, with the proportion of TLE lived with disability remaining highly stable or showing only a marginal, subtle decline across cohorts. Conversely, the chronic disease domain revealed a pronounced morbidity expansion. While the incidence of major chronic diseases declined only marginally across cohorts, the associated mortality risk decreased substantially. Consequently, later-born cohorts are living more absolute years, and a greater proportion of their TLE, with chronic conditions.

Conclusions

The health transition among older adults in China is highly domain-specific, characterized by a compression of cognitive morbidity, a dynamic equilibrium in physical functioning, and an expansion of chronic disease morbidity. More recent cohorts are maintaining their cognitive clarity and physical autonomy (proportionately) for longer, yet this is increasingly occurring alongside prolonged periods of chronic illness. These findings challenge the conventional compression-versus-expansion dichotomy and highlight the need for public health systems in rapidly aging societies to shift focus from acute treatment toward proactive chronic disease management and integrated long-term care.

Highlights

• •Analyzes 20-year longitudinal data (1998–2018) from Chinese older adults across 1895–1949 birth cohorts.
• •Uses continuous-time multi-state Markov models to estimate domain-specific health transitions.
• •Finds a relative compression of cognitive morbidity and a dynamic equilibrium in ADL functioning across cohorts.
• •Reveals expansion of morbidity for major chronic diseases, with longer survival but higher disease burden.
• •Challenges the compression-versus-expansion dichotomy, highlighting the need for integrated chronic care.

1. Introduction

The substantial global increase in human life expectancy—a major public health achievement—has sparked significant academic and policy debate regarding the “lifespan-healthspan” dynamic (Jugran, 2025; Vaupel, 2010). At the core of this discourse is whether older adults are living longer, healthier lives, or merely enduring extended periods of illness and disability. Three competing paradigms address this question. The “compression of morbidity” hypothesis posits that lifestyle modifications and medical advances delay the onset of chronic illness, compressing the period of infirmity into a shorter interval near the end of life (Fries, 1980, 2005). Conversely, the “expansion of morbidity” (or “failures of success”) perspective argues that medical technologies—particularly those controlling fatal complications—inadvertently prolong the duration of chronic diseases and disability (Gruenberg, 1977). Proposing a middle ground, the “dynamic equilibrium” theory suggests that while the prevalence of chronic conditions may rise with increased survival, their severity diminishes, maintaining a relatively stable overall functional status (Manton, 1982).

Decades of empirical research have yielded mixed results, indicating that these paradigms are not mutually exclusive but rather domain-specific and highly contextualized (Chatterji et al., 2015). Globally, recent comprehensive analyses reveal that gains in life expectancy have outpaced improvements in health-adjusted life expectancy. This suggests a general expansion of morbidity driven by the rising prevalence of chronic conditions and behavioral risk factors (Chen et al., 2023; Xiong et al., 2024). Meanwhile, the overall global disease burden remains highly concentrated in older populations (Prince et al., 2015). However, this overarching trend masks divergent patterns across specific nations and health domains.

In the United States, researchers have documented a pronounced expansion of disease alongside a clear deterioration in mobility functioning (Crimmins & Beltrán-Sánchez, 2011). Although some long-term improvements in disability-free life expectancy at older ages have been observed (Crimmins et al., 2016), recent cohort studies reveal marked reversals. Younger cohorts, such as the Baby Boomers, are experiencing a pronounced expansion of morbidity, while disability trends remain generally stable—aligning with dynamic equilibrium—or only expand among socioeconomically disadvantaged subgroups (Payne, 2022; Shen & Payne, 2023). Furthermore, these recent cohorts exhibit worsening physiological and mental health profiles (Zheng & Echave, 2021). European evidence, by contrast, presents a differing trajectory. Danish studies, for instance, demonstrate a “success of success” effect, where later-born cohorts of nonagenarians exhibit significantly better cognitive functioning and abilities in daily living, though objective physical performance showed no consistent improvement compared to their predecessors born a decade earlier (Christensen et al., 2009, 2013). Similarly, longitudinal data from the Netherlands highlight a distinct domain divergence: a substantial decline in physically healthy life expectancy occurring simultaneously with a marked increase in cognitively healthy life expectancy (Deeg et al., 2018).

Collectively, these findings underscore the multidimensionality of late-life health trajectories. A compression of morbidity in functional or cognitive domains frequently co-occurs with an expansion of physiological disease burden, thereby precluding any simplistic binary categorization of population health trends.

To comprehensively understand these complex health transitions, a cohort perspective is essential. While disentangling age, period, and cohort (APC) effects remains a classic methodological challenge, historical shifts in late-life health—such as the postponement of functional and cognitive impairments—are theoretically best conceptualized through the lens of birth cohorts. Following Ryder's (1965) foundational work, birth cohorts serve as fundamental vehicles of social change. Younger cohorts uniquely embody the progressive, long-term improvements in childhood nutrition, maternal care, and early public health infrastructure that occurred during their formative years. Cumulative advantage theory further posits that these early-life, cohort-specific experiences translate into lifetime accumulations of health capital, systematically shaping and differentiating aging trajectories (Dannefer, 1987; Haas et al., 2017). For example, later-born cohorts benefit from secular gains in education (e.g., the Flynn effect) that confer higher baseline cognitive reserve (Brailean et al., 2018; Finkel et al., 2007, pp. P286–P294), and these accumulated health assets are the primary mechanisms explaining long-term declines in severe disability and chronic disease mortality (Yang, 2008). Therefore, rather than attributing health improvements to immediate temporal shocks (period effects), robust investigations must track continuous, dynamic transitions across successive birth cohorts to capture the lifelong dividends of social and environmental progress.

The Chinese context provides a compelling setting for applying this multidimensional, cohort-based framework. China's experience of “compressed modernity”—achieving epidemiological and demographic shifts in decades that took over a century in Western nations—has driven a rapid transition from communicable to chronic non-communicable diseases (Yang et al., 2013). This accelerated transformation functions as a unique natural experiment wherein successive birth cohorts have encountered highly asynchronous changes in the social determinants of health across their life courses (Chen et al., 2010).

Unsurprisingly, empirical evidence regarding Chinese health trends remains highly inconsistent. Some studies observe a trend toward morbidity compression, noting declines in chronic disease prevalence and improvements in self-rated health, although improvements in activities of daily living (ADL) disability were not statistically significant (Gu et al., 2009), with recent research suggesting an ongoing transition from expansion to compression (Zhang et al., 2022). Conversely, a prominent comparative study of the oldest-old revealed the “coexistence of benefits and costs of success”: while later cohorts demonstrated lower mortality and reduced ADL disability, their objective cognitive and physical performance deteriorated (Zeng et al., 2017). Other cohort studies have similarly reported non-linear trends in cognitive impairment (Kuang et al., 2020) and persistent age effects that overshadow cohort-level gains in physical performance (Zhang et al., 2020). We contend that these seemingly contradictory findings are not statistical artifacts. Rather, they reflect a multi-domain divergence in China's health transition, driven by the uneven evolution of underlying social determinants.

Much of the inconsistency in the existing literature stems from conceptual and methodological limitations. First, prior studies typically evaluate health expectancies by treating cognition, ADLs, and chronic diseases in isolation. Consequently, these core domains are rarely examined simultaneously within the same historical cohorts, preventing a systematic mapping of domain-specific health transitions. Second, previous research relies heavily on the Sullivan method (Sullivan, 1971), which computes expectancies using cross-sectional prevalence data. While rigorous mathematical proofs have demonstrated that Sullivan's method yields unbiased and consistent estimates under steady-state population dynamics (Imai & Soneji, 2007), this stationarity assumption is frequently violated during periods of rapid epidemiological change. China's “compressed modernity” represents precisely this exception, where mortality rates, disease incidence, and medical interventions have shifted drastically and asynchronously across recent birth cohorts. Furthermore, static prevalence-based approaches collapse incidence, recovery, and mortality into a single stock measure, failing to capture the continuous dynamics of health transitions (Manton et al., 2008). Without modeling bidirectional recovery and state-specific competing mortality risks, it is impossible to unpack the precise mechanisms driving morbidity expansion. Third, empirical evidence utilizing continuous data up to 2018—a period of profound socioeconomic transformation in China—remains scarce.

To address these gaps, this study leverages 20 years of longitudinal data (1998–2018) from the Chinese Longitudinal Healthy Longevity Survey (CLHLS), encompassing a broad range of birth cohorts (1895-1949). Overcoming the limitations of static, prevalence-based designs, we employ continuous-time multi-state Markov models (Van den Hout, 2017) to dynamically estimate transition probabilities—including onset, recovery, and mortality—to compute Healthy Life Expectancy (HLE). By comparing successive birth cohorts, this study addresses two central questions: (1) How do dynamic transition probabilities between healthy, impaired, and deceased states differ across historical birth cohorts of older adults in China? (2) Does the trajectory of health expectancies among these cohorts indicate a compression, expansion, or dynamic equilibrium of morbidity, and how do these patterns vary across cognitive, functional (ADL), and physiological (chronic disease) domains? By analyzing the multidimensional nature of health transitions through a cohort lens, this study aims to reconcile conflicting literature findings and provide critical insights into the epidemiological transition within a rapidly modernizing, non-Western context.

2. Data and methods

2.1. Data and sample

This study uses data from the Chinese Longitudinal Healthy Longevity Survey (CLHLS), a nationally representative longitudinal cohort of older adults in China. Initiated in 1998, the survey spans 23 randomly selected provinces, capturing approximately 85% of the total Chinese population. It initially targeted the oldest-old (aged ≥80 years) before expanding its scope in 2002 to include younger-old adults (aged 65–79 years). For this analysis, we pooled data across eight survey waves conducted between 1998 and 2018 (1998, 2000, 2002, 2005, 2008, 2011, 2014, and 2018). This 20-year observation window provides robust statistical power for analyzing long-term health transitions and cohort dynamics.

The pooled CLHLS sample with complete interview dates comprised 116,854 person-wave observations, restricted to participants aged 65 to 110 years. To ensure data quality and suitability for multi-state modeling, we applied four sequential exclusion criteria (1) inconsistent time-invariant variables (e.g., sex, years of education): across waves; (2) logical errors, such as a subsequent interview date preceding a prior wave or a negative age change over time; (3) observation intervals shorter than six months, to prevent unstable estimates of short-term transition intensities; and (4) single-wave participation, which precludes follow-up transition analysis. Following these procedures, a total of 23,273 person-wave observations were excluded, yielding a final analytical sample of 93,581 person-wave observations from a mixed longitudinal cohort born between 1895 and 1949.

To compensate for panel attrition—driven primarily by high mortality among the oldest-old—and to maintain national representativeness, the CLHLS employed a continuous refreshment sampling design. Consequently, the analytical sample forms an unbalanced panel of overlapping cohorts entering the study at different waves. Specifically, of the 93,581 person-wave observations, 19,665 originated from the initial 1998 baseline cohort. This core sample was supplemented by refreshment cohorts introduced in 2000 (n = 13,768), 2002 (n = 24,525), 2005 (n = 15,171), 2008 (n = 17,610), 2011 (n = 1630), and 2014 (n = 1212). This mixed longitudinal design facilitates robust estimations of health transitions across diverse historical birth cohorts.

Crucially, this extensive 20-year follow-up generated an exceptionally robust number of empirical transitions between health states, ensuring sufficient statistical power to estimate bidirectional dynamics, including recovery probabilities. As detailed in Supplementary Table S1 (Empirical Wave-to-Wave Transition Frequencies), we captured thousands of backward transitions. For instance, we observed 3064 recovery events from mild cognitive impairment back to intact cognition, 2071 instances of regaining ADL independence, and 5166 transitions from a reported diseased state back to a disease-free state (reflecting symptom remission or successful pharmacological control of chronic conditions). These massive case volumes guarantee the precision of our continuous-time multi-state estimations, particularly for the backward transition intensities.

2.2. Measures

To capture the multidimensional nature of health transitions, we operationalized health across three distinct domains. For each domain, respondents were classified into a set of discrete states at each survey wave. Mortality served as the final, absorbing state across all models, with dates of death ascertained via post-mortem interviews with surviving family members.

Cognitive Function: Cognitive function was assessed using the Chinese version of the Mini-Mental State Examination (MMSE), a validated 25-item instrument measuring orientation, memory, attention, calculation, and language (Folstein et al., 1975). Following established clinical cutoffs, participants were classified into three living states: “Cognitively Intact” (scores 24–30; State 1), “Mild Cognitive Impairment” (scores 10–23; State 2), and “Severe Cognitive Impairment” (scores 0–9; State 3). Death constituted the fourth, absorbing state.

Activities of Daily Living (ADL): Functional status was measured using the Katz Index of Independence in ADL, which assesses performance in six basic activities: bathing, dressing, toileting, transferring, continence, and feeding (Katz, 1983). Participants were categorized as “ADL Independent” (State 1) if they required no assistance with any activity, or as having an “ADL Disability” (State 2) if they required assistance with at least one activity. Death served as the third state.

Major Chronic Diseases: This domain relied on self-reported medical diagnoses of five major chronic conditions: hypertension, diabetes, heart disease, stroke/cerebrovascular disease, and respiratory diseases (e.g., bronchitis, emphysema). Respondents were classified as “Free of Major Chronic Diseases” (State 1) if they reported none of these conditions, or as “Having at Least One Major Chronic Disease” (State 2) if they reported one or more. Death constituted the third state.

The primary independent variable was birth cohort, measured continuously as the respondent's birth year. Given the 20-year observation period and the broad age range of the participants, the analytical sample encompasses birth cohorts strictly spanning from 1895 to 1949. This specific range captures a unique generation of Chinese older adults who were born and spent their formative years prior to the founding of the People's Republic of China in 1949. Educational attainment was operationalized as a categorical variable to capture potential non-linear effects, reflecting years of formal schooling: 0 years (no formal education), 1–5 years (primary education), 6–9 years (middle school education), and ≥10 years (high school and above).

2.3. Analytical strategy

Traditional methods for calculating health expectancy, such as the Sullivan method (Sullivan, 1971), rely on cross-sectional prevalence data. Although Sullivan's estimator is highly robust when mortality and morbidity rates are stationary (Imai & Soneji, 2007), it inherently lags behind underlying dynamics during periods of rapid cohort-driven epidemiological transitions. Consequently, prevalence-based methods cannot capture bidirectional health transitions (e.g., recovery from impairment) and cannot isolate whether changes in health expectancies are driven by altered incidence or altered mortality among the diseased. Furthermore, because longitudinal surveys like the CLHLS utilize panel data, observations are inherently interval-censored (i.e., the exact timing of a transition between waves is unobserved), and follow-up intervals vary among individuals.

To address these methodological challenges, we employed continuous-time multi-state Markov models (Van den Hout, 2017). For each health domain, let $S\left(t\right)$ denote an individual's health state at age $t$. The process dynamics are governed by transition intensities, $q_{rs}\left(t,z\right)$, which represent the instantaneous risk of moving from state $r$ to state $s$. These intensities are modeled as a function of age ($t$) and a vector of individual-specific covariates ($z$), including birth cohort, sex, residence, and education. The transition intensity is formally defined as:

$$q_{rs}\left(t,z\right)=\underset{\delta t\to 0}{\lim }\frac{P\left(S\left(t+\delta t\right)=s|S\left(t\right)=r\right)}{\delta t}$$

For ADL functioning and chronic diseases, we constructed a three-state “illness-death” model with a transition intensity matrix, $Q$. This model permits transitions from a healthy to an impaired state ($q_{12}$), from either state to death ($q_{13}$, $q_{23}$), and recovery from the impaired to the healthy state ($q_{21}$). For cognitive function, we specified a four-state progressive model. This model allows for transitions from intact to mild impairment ($q_{12}$), from mild to severe impairment ($q_{23}$), and from any living state to death ($q_{14}$, $q_{24}$, $q_{34}$). It also accommodates recovery from mild to intact cognition ($q_{21}$) and from severe to mild impairment ($q_{32}$). For all models, the diagonal elements of the matrices are constrained such that the row sums equal zero: $q_{rr}=-\sum _{s\ne r}{q}_{rs}$. To clarify the structure of these analytical designs, a conceptual diagram illustrating the distinct state spaces and allowable transition pathways for each of the three domains is provided in Supplementary Fig. S1. Importantly, given the computational complexities of continuous-time optimization, the multi-state Markov models for cognitive function, ADL, and chronic diseases were estimated as three separate parallel processes rather than as a single joint model.

We specified the transition intensities using a parametric Gompertz proportional hazards model, wherein the baseline risk of transition changes exponentially with age. Following standard survival analysis formulation, the hazard function for the transition from state $r$ to state $s$ at physiological age $t$ is specified as:

$$h_{rs}\left(t\right)=\exp \left({\beta }_{rs}+{\xi }_{rs}t+{\gamma }_{rs}^{T}z\right)$$

Here, $t$ represents age, and ${\beta }_{rs}$ represents the log-baseline transition hazard (corresponding to the logged scale parameter of the standard Gompertz distribution). The shape parameter ${\xi }_{rs}$ captures the exponential rate of change in the hazard with age. Finally, ${\gamma }_{rs}$ is the vector of log hazard ratios (coefficients) corresponding to the vector of covariates z, which act additively within the log-scale exponent. Models were estimated via maximum likelihood for interval-censored data using the msm package in R (Jackson, 2011).

Using parameters derived from the fitted multi-state models, we computed state-specific expected remaining years. For an individual in state $r$ at age $t$, the expected number of years to be lived in state $s$ (where $s$ is not the absorbing death state, $D$) is calculated by integrating the transition probabilities over their remaining lifespan:

$$e_{rs}\left(t|z\right)={\int }_0^{\infty }P\left(S\left(t+u\right)=s\mid S\left(t\right)=r,z\right)du$$

These integrals were computed numerically using the ‘elect’ package in R (van den Hout et al., 2019). The remaining lifespan was discretized using a fine grid parameter (h = 0.5 years), and the integral of the transition probability matrix $P\left(t\right)=\mathit{exp}\left(tQ\right)$ was evaluated via a piecewise-constant step approximation up to an assumed mathematical maximum age of 115, beyond which survival probability is negligible.

To obtain the marginal life expectancy in state $s$ independent of the initial state, we weighted these state-specific expectancies by the estimated prevalence of each non-death state at age $t$:

$$e_{.s}\left(t|z\right)=\sum _{r\ne D}P\left(S\left(t\right)=r|z\right)e_{rs}\left(t|z\right)$$

Here, the prevalence distribution of living states, $P\left(S\left(t\right)=r|z\right)$, was estimated from the baseline data corresponding to the specific age and sex profiles. Finally, total life expectancy (TLE) is the sum of the marginal life expectancies across all non-death states ($TLE=\sum _{s\ne D}{e}_{.s}\left(t|z\right)$).

To rigorously test hypotheses regarding morbidity compression or expansion, we employed a dual-model analytical strategy. First, to identify the multidimensional predictors of transition intensities (e.g., estimating Hazard Ratios), we utilized a full model adjusting for all covariates including downstream socioeconomic factors (education and residence). Second, to compute and project health expectancies (HLE and TLE) across discrete birth cohorts without introducing over-adjustment bias, we utilized a base demographic model controlling only for physiological age, sex, and birth cohort. Because cohort succession fundamentally encompasses secular gains in educational attainment and living standards, holding downstream socioeconomic covariates constant at arbitrary sample means would artificially strip away the very mechanisms driving the historical cohort effect.

Here, morbidity compression (or expansion) is defined as a decrease (or increase) in the proportion of total life expectancy (%TLE) spent in an impaired health state. Crucially, to strictly prevent the risk of unconvincing parametric extrapolation (e.g., projecting younger cohorts to oldest-old ages they have not yet historically attained), we constrained our health expectancy calculations exclusively to the exact age-sex-cohort combinations empirically observed within our 20-year CLHLS dataset. This data-driven constraint ruled out all hypothetical profiles, yielding a robust set of 1208 unique, historically real covariate profiles. By comparing the resulting health expectancies—both in absolute years and as a proportion of TLE—across sequential cohorts at equivalent observed ages, we effectively isolated historically authentic cohort dynamics from biological aging (see Table 1).

Table 1.

Longitudinal tracking and refreshment of the CLHLS analytical sample (1998–2018).

Survey Year	Entered in 1998	Entered in 2000	Entered in 2002	Entered in 2005	Entered in 2008	Entered in 2011	Entered in 2014	Total
1998	6637	0	0	0	0	0	0	6637
2000	6626	4603	0	0	0	0	0	11,229
2002	3424	4597	7580	0	0	0	0	15,601
2005	1937	2764	7566	5462	0	0	0	17,729
2008	703	1161	4189	5457	6627	0	0	18,137
2011	226	421	2516	2320	6469	575	0	12,527
2014	92	176	1739	1297	3095	622	606	7627
2018	20	46	935	635	1419	433	606	4094
Total	19,665	13,768	24,525	15,171	17,610	1630	1212	93,581

Note: The table presents the number of person-wave observations after data cleaning based on the study's inclusion criteria. The columns represent the year individuals were first enrolled in the survey (refreshment cohorts), and rows represent the cross-sectional survey waves.

3. Results

3.1. Predictors of health transitions: the role of birth cohort

We first estimated continuous-time multi-state Markov models to identify the determinants of transition intensities across the three health domains, with results summarized in Table 2. As expected, advancing age was a significant predictor of health deterioration across the cognitive and functional domains, associated with an increased risk of cognitive decline and ADL disability onset, as well as a higher risk of death from most health states, with the exception of those with cognitive impairments. Interestingly, advancing age did not significantly predict the onset of new major chronic diseases (HR = 1.00, 95% CI [0.99, 1.01]), likely reflecting a mortality selection effect among the oldest-old survivors. Conversely, age was associated with a significantly lower probability of recovery across all domains.

Table 2.

Multidimensional predictors of health transitions: Hazard ratios (HR) and 95% confidence intervals from continuous-time Markov models.

Transition Pathway	Age (per year)	Male (vs. Female)	Birth Cohort	Urban (vs. Rural)	Education: 1–5 yrs	Education: 6–9 yrs	Education: ≥ 10 yrs
Panel A: Cognitive Function (4-State Model)
Intact → Mild Impairment (Onset)	1.03 (1.02, 1.04)	0.84 (0.77, 0.92)	0.97 (0.96, 0.98)	0.99 (0.91, 1.07)	0.83 (0.75, 0.92)	0.61 (0.52, 0.71)	0.49 (0.39, 0.62)
Intact → Death (Mortality)	1.08 (1.06, 1.09)	1.70 (1.49, 1.95)	1.01 (0.99, 1.02)	0.84 (0.74, 0.95)	0.90 (0.78, 1.04)	0.98 (0.83, 1.16)	0.85 (0.67, 1.09)
Mild Impairment → Intact (Recovery)	0.95 (0.94, 0.97)	1.28 (1.13, 1.44)	0.98 (0.97, 0.99)	1.02 (0.91, 1.13)	1.21 (1.05, 1.40)	1.25 (1.00, 1.56)	1.42 (1.03, 1.95)
Mild → Severe Impairment (Progression)	1.05 (1.04, 1.07)	1.37 (1.22, 1.54)	0.99 (0.98, 1.00)	1.01 (0.93, 1.11)	1.23 (1.07, 1.42)	1.14 (0.92, 1.42)	1.14 (0.86, 1.51)
Mild Impairment → Death (Mortality)	0.74 (0.61, 0.90)	1.78 (0.81, 3.91)	0.79 (0.67, 0.92)	2.73 (1.17, 6.33)	1.11 (0.45, 2.74)	1.43 (0.52, 3.90)	1.00 (0.18, 5.69)
Severe → Mild Impairment (Recovery)	0.95 (0.93, 0.98)	1.50 (1.20, 1.89)	0.96 (0.94, 0.98)	0.89 (0.75, 1.07)	1.19 (0.91, 1.56)	0.93 (0.60, 1.43)	0.74 (0.42, 1.31)
Severe Impairment → Death (Mortality)	0.99 (0.98, 0.99)	1.23 (1.16, 1.29)	0.98 (0.98, 0.99)	0.98 (0.94, 1.02)	1.09 (1.02, 1.16)	0.99 (0.89, 1.11)	1.00 (0.85, 1.17)
Panel B: Physical Independence (ADL, 3-State Model)
Independent → Disabled (Onset)	1.07 (1.06, 1.08)	1.00 (0.93, 1.08)	0.99 (0.98, 0.99)	1.02 (0.95, 1.09)	0.92 (0.84, 1.01)	0.90 (0.79, 1.02)	0.79 (0.67, 0.93)
Independent → Death (Mortality)	1.02 (1.00, 1.04)	1.71 (1.40, 2.09)	0.97 (0.95, 0.98)	1.05 (0.86, 1.28)	0.84 (0.67, 1.07)	0.88 (0.66, 1.17)	0.63 (0.41, 0.98)
Disabled → Independent (Recovery)	0.91 (0.90, 0.92)	1.07 (0.94, 1.20)	0.95 (0.94, 0.96)	0.68 (0.62, 0.76)	0.86 (0.75, 0.99)	0.87 (0.72, 1.07)	0.68 (0.52, 0.88)
Disabled → Death (Mortality)	1.03 (1.02, 1.03)	1.30 (1.24, 1.37)	1.00 (0.99, 1.01)	0.81 (0.78, 0.85)	1.02 (0.96, 1.08)	0.88 (0.80, 0.97)	0.80 (0.71, 0.91)
Panel C: Major Chronic Diseases (3-State Model)
Disease-free → Diseased (Onset)	1.00 (0.99, 1.01)	1.14 (1.06, 1.23)	0.99 (0.98, 0.99)	1.06 (0.99, 1.13)	1.00 (0.93, 1.09)	1.07 (0.97, 1.19)	0.96 (0.83, 1.11)
Disease-free → Death (Mortality)	1.07 (1.06, 1.08)	1.16 (1.09, 1.25)	0.99 (0.99, 1.00)	0.97 (0.92, 1.03)	0.97 (0.90, 1.06)	0.84 (0.73, 0.96)	0.74 (0.60, 0.89)
Diseased → Disease-free (Recovery)	0.95 (0.94, 0.95)	1.08 (0.99, 1.16)	0.92 (0.92, 0.93)	0.75 (0.70, 0.81)	0.94 (0.85, 1.02)	0.91 (0.81, 1.03)	0.71 (0.61, 0.84)
Diseased → Death (Mortality)	1.07 (1.06, 1.07)	1.35 (1.26, 1.44)	0.99 (0.98, 0.99)	0.90 (0.85, 0.97)	0.91 (0.84, 0.99)	0.89 (0.79, 1.00)	0.76 (0.65, 0.88)

Notes: Values represent estimated Hazard Ratios (HR) with 95% Confidence Intervals presented in parentheses. Results are estimated using maximum likelihood from continuous-time multi-state Markov models adjusted for all listed covariates simultaneously.

After controlling for age, sex, residence, and educational attainment, birth cohort emerged as a pronounced and independent predictor of health transition dynamics. For cognitive function (Table 2, Panel A), individuals in later-born cohorts had a significantly lower risk of transitioning from a cognitively intact state to mild impairment (Intact to Mild, Hazard Ratio HR = 0.97, 95% CI [0.96, 0.98]). However, these cohorts also exhibited a reduced likelihood of recovery, both from mild to intact status (Mild to Intact, HR = 0.98, 95% CI [0.97, 0.99]) and from severe to mild impairment (Severe to Mild, HR = 0.96, 95% CI [0.94, 0.98]). Furthermore, belonging to a later cohort was associated with a significantly lower mortality risk from both mild (Mild to Death, HR = 0.79, 95% CI [0.67, 0.92]) and severe cognitive impairment (Severe to Death, HR = 0.98, 95% CI [0.98, 0.99]).

A similar pattern was observed for functional independence (Table 2, Panel B). Later-born cohorts showed a significantly lower risk of mortality from an ADL-independent state (Independent to Death, HR = 0.97, 95% CI [0.95, 0.98]). Consistent with the findings for cognition, however, their probability of recovering from disability to an independent state was also significantly lower (Disabled to Independent, HR = 0.95, 95% CI [0.94, 0.96]).

The cohort dynamics for major chronic diseases presented a parallel, yet consequentially distinct pattern (Table 2, Panel C). Although later-born cohorts had a modest but statistically significant reduction in the risk of disease onset (Disease-free to Diseased, HR = 0.99, 95% CI [0.98, 0.99]), they faced a markedly lower probability of recovery from a diagnosed condition (Diseased to Disease-free, HR = 0.92, 95% CI [0.92, 0.93]). Notably, the mortality risk for individuals already living with a major chronic disease was also significantly lower for later cohorts (Diseased to Death, HR = 0.99, 95% CI [0.98, 0.99]).

Taken together, the transition matrix reveals a complex dynamic: later-born older adults in China are experiencing a delayed onset of cognitive and functional impairments and face lower mortality risks across most health states. However, this survival advantage is coupled with a diminished likelihood of recovery once these conditions manifest, suggesting a potential for prolonged periods of survival in states of compromised health.

3.2. Compression of morbidity: cognitive function and ADL independence

To translate transition intensities into interpretable population health metrics, we calculated state-specific life expectancies for sequential birth cohorts using the base demographic model. To illustrate these cohort dynamics without engaging in mathematical extrapolation beyond our empirical Lexis surface, we selected exact ages 65 and 80 as representative benchmarks. Age 65 represents the standard threshold for older adulthood, while age 80 marks the transition into the oldest-old population. We present the health expectancies for the major birth cohorts that were actually observed at these exact ages during our study period. (The complete estimations across all empirical age-cohort profiles are provided in Supplementary Table S4–S6). The benchmark results are presented in Table 3.

Table 3.

Cohort differences in multidimensional health expectancies at exact ages 65 and 80.

Health Domain & Demographic	Birth Cohort	Total LE (Years)	Healthy State Years (% of TLE)	Impaired State 1 Years (% of TLE)	Impaired State 2 Years (% of TLE)
Panel A: Cognitive Function			Cognitively Intact	Mild Impairment	Severe Impairment
Men at Age 65	1937	16.02	13.72 (85.6%)	1.56 (9.7%)	0.74 (4.6%)
	1940	16.13	13.88 (86.0%)	1.50 (9.3%)	0.75 (4.6%)
	1943	16.17	13.98 (86.5%)	1.44 (8.9%)	0.76 (4.7%)
Men at Age 80	1917	7.49	5.25 (70.1%)	1.46 (19.5%)	0.78 (10.4%)
	1918	7.59	5.32 (70.1%)	1.46 (19.3%)	0.80 (10.6%)
	1920	7.74	5.45 (70.4%)	1.46 (18.8%)	0.84 (10.8%)
	1922	7.85	5.55 (70.7%)	1.44 (18.3%)	0.86 (11.0%)
	1924	7.92	5.63 (71.1%)	1.41 (17.8%)	0.88 (11.1%)
Women at Age 65	1937	18.37	13.72 (74.7%)	3.17 (17.3%)	1.48 (8.1%)
	1940	18.67	14.04 (75.2%)	3.10 (16.6%)	1.53 (8.2%)
	1943	18.88	14.31 (75.8%)	3.00 (15.9%)	1.58 (8.3%)
Women at Age 80	1917	8.39	4.50 (53.6%)	2.57 (30.7%)	1.32 (15.7%)
	1918	8.54	4.58 (53.7%)	2.60 (30.4%)	1.36 (15.9%)
	1920	8.78	4.74 (54.0%)	2.61 (29.8%)	1.43 (16.3%)
	1921	8.87	4.81 (54.2%)	2.61 (29.4%)	1.46 (16.4%)
	1922	8.96	4.87 (54.4%)	2.60 (29.0%)	1.48 (16.6%)
Panel B: ADL Independence			ADL Independent	ADL Disabled	—
Men at Age 65	1937	15.66	13.72 (87.6%)	1.94 (12.4%)	—
	1940	15.85	13.88 (87.6%)	1.97 (12.4%)	—
	1943	16.04	14.05 (87.6%)	2.00 (12.4%)	—
Men at Age 80	1917	7.44	5.69 (76.4%)	1.75 (23.6%)	—
	1918	7.47	5.71 (76.4%)	1.76 (23.6%)	—
	1920	7.54	5.76 (76.4%)	1.78 (23.6%)	—
	1922	7.61	5.81 (76.4%)	1.80 (23.6%)	—
	1924	7.67	5.86 (76.4%)	1.81 (23.6%)	—
Women at Age 65	1937	17.4	14.74 (84.7%)	2.65 (15.3%)	—
	1940	17.47	14.79 (84.7%)	2.68 (15.3%)	—
	1943	17.55	14.85 (84.6%)	2.70 (15.4%)	—
Women at Age 80	1917	8.57	6.15 (71.8%)	2.42 (28.2%)	—
	1918	8.58	6.16 (71.7%)	2.43 (28.3%)	—
	1920	8.6	6.16 (71.6%)	2.44 (28.4%)	—
	1921	8.61	6.16 (71.6%)	2.45 (28.4%)	—
	1922	8.62	6.17 (71.6%)	2.45 (28.4%)	—
Panel C: Major Chronic Diseases			Disease-Free	Diseased	—
Men at Age 65	1937	15.57	6.60 (42.4%)	8.97 (57.6%)	—
	1940	15.72	6.16 (39.2%)	9.56 (60.8%)	—
	1943	15.89	5.76 (36.3%)	10.13 (63.7%)	—
Men at Age 80	1917	7.59	4.58 (60.4%)	3.01 (39.6%)	—
	1918	7.61	4.50 (59.2%)	3.10 (40.8%)	—
	1920	7.64	4.35 (57.0%)	3.29 (43.0%)	—
	1922	7.69	4.21 (54.8%)	3.48 (45.2%)	—
	1924	7.74	4.08 (52.7%)	3.66 (47.3%)	—
Women at Age 65	1937	17.36	7.64 (44.0%)	9.73 (56.0%)	—
	1940	17.55	7.14 (40.7%)	10.41 (59.3%)	—
	1943	17.76	6.69 (37.7%)	11.07 (62.3%)	—
Women at Age 80	1917	8.64	5.34 (61.9%)	3.29 (38.1%)	—
	1918	8.66	5.26 (60.7%)	3.40 (39.3%)	—
	1920	8.71	5.09 (58.4%)	3.63 (41.6%)	—
	1921	8.74	5.01 (57.3%)	3.74 (42.7%)	—
	1922	8.77	4.93 (56.2%)	3.85 (43.8%)	—

Notes: Total LE = Total Life Expectancy (in years). Values outside parentheses represent the absolute years of life expectancy; values in parentheses indicate the proportion (% of TLE) spent in that specific health state. Estimates are derived from the demographic baseline models parameterized with exact age, sex, and actual empirical birth cohorts. Note that total life expectancy (TLE) estimates may vary slightly across Panels A, B, and C (e.g., at age 65 for the 1937 cohort) because they are derived from separate multi-state Markov models (a 4-state model for cognitive function, and 3-state models for ADL and chronic diseases) which adapt to different state-specific transition intensities and sample configurations.).

As shown in Table 3, total life expectancy (TLE) at identical ages consistently increased across more recent cohorts. For example, the TLE for 80-year-old women increased from 8.39 years for the 1917 birth cohort to 8.96 years for the 1922 cohort—an increase of 0.57 years. Men aged 80 experienced a similar gain, with TLE rising from 7.49 years (1917 cohort) to 7.92 years (1924 cohort). Substantial longevity gains were also evident at age 65, where TLE for women rose from 18.37 years (1937 cohort) to 18.88 years (1943 cohort).

In the domain of cognitive function, this gain in longevity was primarily characterized by a relative compression of morbidity. Life expectancy in a cognitively intact state improved markedly across cohorts. At age 80, women born in 1922 could expect 4.87 years of cognitively intact life, 0.37 years longer than their counterparts born in 1917 (4.50 years). Men aged 80 demonstrated a similar trend, with intact years increasing from 5.25 to 5.63 years. This compression trend is equally evident at age 65, where the cognitively intact life expectancy for the 1943 cohort was longer than that of the 1937 cohort. Consequently, the proportion of remaining life spent in a cognitively intact state increased for more recent cohorts. Although prolonged survival resulted in a slight increase in the absolute time spent with cognitive impairment among the oldest-old at age 80, a dual reduction in both absolute and relative impairment years was observed at age 65, suggesting an absolute compression of cognitive morbidity at younger-old ages. This relative compression is visually confirmed in Fig. 1 (Panel A), where the impairment curves for the more recent cohorts (e.g., the 1930s and 1940s) are consistently positioned at the lower end of the distribution, indicating a lower proportion of life lived with cognitive deficits at any given age.

Fig. 1.

Cohort-specific life-course trajectories of the proportion of remaining life expectancy spent with cognitive impairment, functional disability, and major chronic diseases.

A parallel pattern of dynamic equilibrium, accompanied by a subtle shift toward relative compression, was observed for ADL functioning. Absolute ADL-independent life expectancy consistently expanded across cohorts: at age 80, independent years for women rose from 6.15 years (1917 cohort) to 6.17 years (1922 cohort), while for men it increased from 5.69 to 5.86 years. At age 65, men's independent life expectancy increased from 13.72 to 14.05 years. Although the absolute years lived with ADL disability rose slightly due to prolonged overall survival, the proportion of remaining life spent in a disabled state remained highly stable, with exact-age estimates in Table 3 showing near-stagnation. However, taking a life-course perspective across the entire age span, Fig. 1 (Panel B) reveals a consistent, albeit marginal, downward shift of the disability curves for more recent cohorts at overlapping ages. This tightly bundled yet sequentially stratified pattern suggests that the extended lifespan is indeed favoring functional independence, aligning with the hypothesis of a dynamic equilibrium with a minor leaning toward relative compression.

3.3. Expansion of morbidity: the burden of major chronic diseases

In contrast to the trends observed in the functional and cognitive domains, data on major chronic diseases indicate a profound expansion of morbidity. As detailed in Table 3 (Panel C), while total life expectancy has increased, the expected years lived without major chronic diseases have consistently declined across successive cohorts.

For instance, when comparing 65-year-olds from the 1937 and 1943 birth cohorts, disease-free life expectancy declined markedly from 7.64 to 6.69 years for women, and from 6.60 to 5.76 years for men. Consequently, the expected years lived with at least one major chronic disease rose substantially. For 65-year-old women, morbid life expectancy increased by 1.34 years (from 9.73 to 11.07 years), while for men, it increased by 1.16 years (from 8.97 to 10.13 years). A similar, substantial expansion of morbid life expectancy is uniquely evident among the oldest-old at age 80.

This expansion of morbidity is further illustrated by the proportion of life spent with chronic conditions. In stark contrast to the tightly bundled curves for functional and cognitive health, Fig. 1 (Panel C) reveals a pronounced “step-ladder” pattern for major chronic diseases. The cohort curves are distinctly stratified, with the proportion of total life expectancy spent with a disease increasing substantially in more recent cohorts across all age groups. For 65-year-old women, the proportion of total life expectancy (TLE) spent managing a chronic condition rose from 56.0% in the 1937 cohort to 62.3% in the 1943 cohort (represented by the uppermost blue dash-dotted curve). Similarly, at age 80, this morbid proportion rose from 38.1% (1917 cohort) to 43.8% (1922 cohort) for women. This marked upward trend is universally evident among both men and women.

Collectively, these findings suggest a critical paradox: although recent cohorts of older Chinese adults are preserving their functional independence and cognitive clarity for longer periods, they are doing so while simultaneously managing a substantially greater and more protracted burden of chronic disease.

4. Discussion

Using 20 years of longitudinal data and a continuous-time multi-state modeling framework, this study elucidates the historical health transitions of Chinese older adults across the 1895-1949 birth cohorts. A central finding is that the epidemiological transition in a rapidly modernizing society does not adhere to a singular paradigm; rather, it is highly domain-specific. This observation challenges the linear, stage-wise assumptions of classic epidemiologic transition theory (Mackenbach, 2022) and suggests a more complex “cross-continuum” of health and disease trajectories (Defo, 2014). Specifically, while we observed a relative compression of morbidity in cognitive functioning, physical independence (ADLs) exhibited a robust pattern of dynamic equilibrium; both trends were accompanied by a sustained expansion of morbidity for major chronic diseases. By capturing the underlying transition dynamics—namely, the delayed onset of impairments alongside prolonged survival with chronic conditions—our study contributes a nuanced, multidimensional perspective that theoretically reconciles the ongoing debate characterized by seemingly contradictory findings in the Chinese context.

For instance, based on cross-sectional prevalence reductions between 1992 and 2002, Gu et al. (2009) suggested a general compression of morbidity, including chronic diseases. Conversely, examining overlapping cohorts of the oldest-old, Zeng et al. (2017) documented deteriorating objective cognitive and physical performance, suggesting morbidity expansion (the “costs of success”). Our dynamic multi-state approach unifies these disjointed observations by demonstrating that both phenomena—compression and expansion—are occurring simultaneously, but strictly within different physiological domains. We support the functional dynamic equilibrium suspected by Gu et al. as recent cohorts maintain ADL independence for a highly stable proportion of their lifespan while expanding absolute independent years. However, our continuous-time transition matrix avoids the limitations of static prevalence measures, revealing that chronic diseases are indeed expanding. By capturing bidirectional transitions and competing mortality risks, we isolate the exact mechanisms driving this expansion: not an increased disease incidence, but a massive reduction in mortality among the diseased coupled with reduced disease reversibility. Consequently, the ‘benefits of success' (Gu et al., 2009; Zeng et al., 2017) manifest in delayed cognitive decline and the proportional maintenance of physical autonomy, whereas the “costs of success” manifest in prolonged, medically-managed survival with chronic conditions.

An apparent paradox emerges from our multi-state transition intensities: although successive cohorts demonstrate a uniform directional pattern across all domains—namely, delayed onset ($q_{12}$), diminished recovery ($q_{21}$), and reduced mortality ($q_{13},q_{23},q_{34}$)—these consistent micro-level shifts translate into markedly divergent macro-level expectancies, specifically a relative compression of functional impairment, a dynamic equilibrium of ADL disability, versus a substantial expansion of chronic disease.

This divergence is not a statistical artifact but rather reflects the asymmetrical nature of epidemiological competing risks. Medical interventions effectively postpone mortality, thereby substantially increasing the denominator of Total Life Expectancy (TLE). Because severe cognitive declines typically manifest as steep, non-linear trajectories confined to the terminal phase of life, delaying their onset even marginally restricts the absolute duration of impairment to a relatively constant terminal window, yielding a relative compression effect when evaluated against a lengthened TLE. In contrast, trajectories of physical independence (ADLs) display a remarkable dynamic equilibrium, where the postponement of functional decline aligns proportionally with mortality reduction, keeping the relative share of disabled life highly stable. Conversely, chronic diseases emerge much earlier in the life course, presenting a flatter, more progressive trajectory. As these conditions have largely transitioned into irreversible, managed states—evidenced by the pronounced decline in the recovery rate—the additional years of TLE consist almost entirely of morbid life, thereby driving a marked expansion of morbidity.

The observed compression of cognitive impairment and the relative stability of functional capacity should be interpreted through a life-course and social epidemiology framework. China's “compressed modernity” exposed successive birth cohorts to markedly different environmental and socioeconomic conditions across their lifespans. For instance, later cohorts in our study (e.g., those born in the 1940s versus the 1900s) experienced early-life stages that coincided with the founding of the People's Republic of China, which introduced widespread improvements in public health and literacy. Consequently, the significant increase in cognitively intact life expectancy among these later cohorts can be understood through the theoretical lens of cognitive reserve (Stern, 2009). The persistent cohort effects we observe suggest that unmeasured improvements in the early-life environment benefit later-life cognitive trajectories, a finding consistent with the protective effects of early-life health investments on physical function (Haas et al., 2017). This aligns with recent evidence demonstrating that more recent cohorts of older adults in China and England are entering old age with higher overall intrinsic capacity (Beard et al., 2025). Furthermore, cross-national studies in middle-income countries confirm that early-life literacy and education exert a strong protective effect against dementia onset, providing supportive evidence for the cognitive reserve hypothesis (Prince et al., 2012). This cohort-driven improvement mirrors the “Flynn effect” observed in Western cognitive aging studies, where later-born cohorts exhibit better initial cognitive performance due to higher educational attainment. However, evidence from Western studies suggests this initial cohort advantage may not necessarily translate into slower rates of cognitive decline with age (Brailean et al., 2018).

Similarly, the preservation of physical independence trajectories reflects the lifelong accumulation of health capital, displaying a remarkable dynamic equilibrium across recent cohorts. Evidence indicates that aging processes are modifiable; continuous increases in life expectancy have been associated with a parallel postponement of severe functional decline, enabling older populations to maintain physical autonomy for a highly stable proportion of their longer lives (Christensen et al., 2009). Recent empirical data from China further confirm this trend, showing that improved age-friendly home environments and greater healthcare access are significantly associated with stabilizing the rates of severe functional dependency amidst rapid population aging (Gong et al., 2022).

In contrast to the observed functional gains, our results indicate a marked expansion of morbidity concerning major chronic diseases. This expansion is elucidated by the multi-state transition probabilities, which reveal two asynchronous demographic forces: while the incidence of chronic diseases ($q_{12}$) has declined only slightly across cohorts, the mortality risk among those with these diseases ($q_{23}$) has decreased substantially.

As our transition matrix reveals, the decline in chronic disease incidence across successive cohorts is exceedingly modest (HR = 0.99). This suggests that while improvements in early-life conditions and primary prevention may have had some success in delaying disease onset, this protective effect has been almost entirely neutralized by the increasing prevalence of modern behavioral and metabolic risk factors (Adolph & Tilg, 2024). The rise of sedentary lifestyles, Westernized diets, and obesity accompanying China's rapid socioeconomic transition has been widely documented as a primary driver of the growing burden of chronic conditions (Yang et al., 2008, 2013). Because the prerequisite for morbidity compression—a substantial delay in disease onset (Fries, 1980)—has not been met, this persistently high disease incidence presents a fundamental challenge to curbing the expansion of chronic morbidity.

Conversely, the substantial decline in mortality risk among individuals with chronic diseases exemplifies the classic “failures of success” paradigm (Gruenberg, 1977) and the “longer life but worsening health” phenomenon (Verbrugge, 1984). Over the past several decades, China's healthcare system has increasingly prioritized secondary and tertiary prevention. This shift responds to the country's broader epidemiological transition, which is characterized by a declining burden of communicable diseases alongside rising rates of chronic non-communicable diseases (NCDs) (Yang et al., 2008, 2013). The implementation of universal health insurance—a critical population-level intervention that mitigates health inequalities by reducing reliance on individual socioeconomic resources (Phelan et al., 2010)—coupled with the routine clinical use of anti-hypertensive drugs, statins, and advanced cardiovascular interventions, has substantially reduced case-fatality rates (Prince et al., 2015). As observed in contemporary U.S. cohorts, widespread medical management lowers biological risk profiles and slows the physiological pace of aging and death (Levine & Crimmins, 2018). Consequently, older adults are less likely to die from acute cardiovascular events; instead, medical advancements enable the survival of individuals who previously would have succumbed to these conditions.

The net outcome of these two diverging forces is a substantial expansion of morbidity (Crimmins & Beltrán-Sánchez, 2011). Because advances in life-saving medical treatments (reducing mortality) have outpaced gains in disease prevention (reducing incidence), a growing pool of survivors accumulates in the diseased state. As a result, later-born cohorts spend not only more years in absolute terms but also a significantly larger proportion of their total life expectancy with major chronic conditions (Jivraj et al., 2020; Shen & Payne, 2023). This structural shift provides compelling empirical support for recent global demographic analyses, which demonstrate that in low-mortality eras, further gains in overall life expectancy are increasingly and disproportionately driven by the expansion of unhealthy life years (Permanyer & Bramajo, 2023).

Synthesizing these divergent trends, our findings suggest that China is entering an era of “functioning with diseases.” More recent cohorts of older adults are better equipped to preserve their physical independence and cognitive clarity not because they are free of pathology, but because chronic conditions are now more effectively managed through medical technology. This dynamic weakens the link between morbidity and disability, a mechanism consistent with the “dynamic equilibrium” hypothesis observed in high-income nations such as the United States (Shen & Payne, 2023). However, our finding that recovery rates from chronic diseases ($q_{21}$) have significantly decreased in later cohorts warrants cautious interpretation. While this likely reflects the irreversible nature of many managed chronic conditions, it may also indicate that individuals categorized as having “mild” impairments in recent cohorts possess a different, more persistent etiological profile than their predecessors. From a policy perspective, as medical advances continue to reduce mortality, the central challenge for future care systems will be to maintain the functional capacity of a growing population living with chronic diseases (Beard & Bloom, 2015; Feng et al., 2025).

Our analysis also reaffirms the persistent gender health-survival paradox (Case & Paxson, 2005; Jiao et al., 2021). Across all cohorts, women had longer total life expectancies than men but spent a disproportionately greater share of their lives with ADL disability, severe cognitive impairment, and multimorbidity. Men, conversely, exhibited higher mortality rates from both healthy and impaired states, resulting in shorter but more functionally healthy final years. These findings highlight that the expansion of morbidity is a gendered burden, concentrated among older women who require extended periods of long-term care. Although historical trends in women's health expectancy have varied by region (Oksuzyan et al., 2010), recent forecasts align with our results, suggesting that due to increased overall longevity, later-born populations—particularly women—may spend a greater proportion of their lives in frail states, even as the hazard of transitioning into severe frailty declines (Tazzeo et al., 2024).

This study has several limitations. First, regarding the classic age-period-cohort (APC) identification problem, our analytical framework predominantly interprets historical changes through the dimension of birth cohorts rather than period effects. We argue this is substantively justified, as improvements in late-life cognitive and functional health are fundamentally driven by life-course cumulative advantages—such as enhanced early-life nutrition and education—that strictly align with cohort succession. Methodologically, while recent advancements like the bounding analysis framework (Fosse & Winship, 2019) offer robust solutions for linear additive models, such linear-nonlinear decompositions are mathematically incompatible with the continuous-time multi-state Markov model (msm) used in our study. Our transition probabilities are derived via non-additive matrix exponentials capturing complex, interval-censored dynamics, precluding the standard additive separation of period and cohort slopes. Second, our measure of major chronic diseases relies on self-reports. Improved education and medical access in more recent cohorts may introduce a diagnostic bias, whereby individuals are more likely to be screened and diagnosed than in previous cohorts, potentially inflating the estimated expansion of morbidity. Third, as highlighted in our methodology, the multi-state models for cognitive function, ADL, and major chronic diseases were estimated separately. While a joint multi-state model would theoretically capture the dynamic correlation between these state spaces (e.g., the co-occurrence of cognitive decline and physical disability), combining a 4-state cognitive process with a 3-state ADL process yields an exponentially expanded, highly complex joint transition matrix. Fitting such a joint continuous-time model with multiple covariates typically leads to severe overparameterization and convergence failures in maximum likelihood estimation. Consequently, separate estimation remains a necessary pragmatic compromise in current multidimensional life expectancy research. Future studies employing larger datasets or Bayesian MCMC estimation techniques may be able to fully model these correlated, joint health trajectories.

5. Conclusion

In conclusion, the health transition among older adults in China is multifaceted, characterized by a compression of cognitive impairment, a dynamic equilibrium in physical functioning, alongside an expansion of chronic morbidity. The observed cohort advantages are not driven by inherent personal traits but rather reflect the substantial achievements of China's socioeconomic and public health development. As the nation's demographic structure undergoes rapid shifts, these findings have significant policy implications. Increasing longevity coupled with a high burden of chronic disease necessitates a reorientation of the healthcare system: from a hospital-centric, acute-care model to a comprehensive, function-centric public health framework. Such a framework must prioritize maintaining intrinsic capacity and promoting healthy aging over merely treating isolated diseases (Beard & Bloom, 2015; Beard et al., 2016; Gianfredi et al., 2025). Furthermore, clinical practice must evolve to manage complex multimorbidity, reflecting the reality that individuals are increasingly living longer with multiple interconnected conditions (Barnett et al., 2012; Nicholson et al., 2019).

Crucially, while the relative proportion of life spent with severe disability may be compressing, the sheer demographic scale of China's aging population means the absolute number of individuals requiring assistance will increase markedly. Recent forecasts project that the number of older adults with complex care needs will grow substantially by 2030 (Gong et al., 2022). Long-term projections further indicate that the disabled older population will follow an inverted U-shaped curve, peaking around 2070, while total person-years lived with disability will continue to rise, imposing structural pressures on society (Feng et al., 2025). It is therefore imperative to build an integrated social care system that addresses both the medical requirements of surviving with disease and the long-term supportive care necessary to ensure dignity in later life.

Availability of data and material

The data and material that support the findings of this study are available on request from the first author, Dr. Kaishan Jiao.

Ethical statement

The study protocol and data collection procedures of the Chinese Longitudinal Healthy Longevity Survey (CLHLS) were approved by the Biomedical Ethics Committee of Peking University (Approval Number: IRB00001052-13074). All procedures performed in this study involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki declaration and its later amendments.

Informed, written consent was obtained from all individual participants (or their legal representatives in the case of cognitive impairment) during each interview wave. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines for the analysis of longitudinal cohort data. All participant data were pseudonymized prior to analysis to ensure confidentiality and protect participants’ anonymity.

Funding

Funding for this study is from the National Social Sciences Research Funds [grant number 24BRK013] for "The Impact of Long-Term Care Insurance on the Equalization of Elderly Care Services in China" and from the National Social Science Fund of China (grant number 19BRK013) for "The research on Health Transition and Health Needs of the Elderly".

CRediT authorship contribution statement

Kaishan Jiao: Conceptualization, Project administration, Resources, Supervision, Writing – original draft, Writing – review & editing. Yujin Song: Data curation, Formal analysis, Methodology, Writing – original draft, Writing – review & editing. Wen Hu: Conceptualization, Funding acquisition, Project administration, Supervision, Writing – review & editing.

Declaration of competing interest

The authors (Kaishan Jiao, Yujin Song, and Wen Hu) declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article.

Data availability

Data will be made available on request.