Productivity-adjusted life year loss among the late middle-aged adults in China

Abstract

Background

Productivity-adjusted life years (PALYs) measure the impact of health conditions on an individual’s ability to contribute productively to the economy. However, they have not been evaluated for the Chinese population. We aim to estimate PALYs for the Chinese population aged 50 to 64 years.

Methods

We applied the China Aging and Retirement Simulation (CHARISMA) model, a validated microsimulation model, to calculate PALYs for three age cohorts (50-54, 55-59, 60-64). PALYs were calculated by integrating workforce participation, absenteeism, and death. A probabilistic sensitivity analysis was conducted to address parameter uncertainties. PALY losses due to specific conditions were estimated by nullifying each health condition’s risk separately. The results were stratified by sociodemographic factors and adjusted using 2018 population data to estimate total PALYs accrued by 2032.

Results

Here we show that the average PALYs before age 65 are 6.80, 4.35, and 1.50 years for the 50–54, 55–59, and 60–64 age groups, respectively. Significant PALY variations are observed among subgroups, with males, rural residents, and individuals with lower education exhibiting higher PALYs. From 2018 to 2032, total accrued PALYs for the cohort initially aged 50-64 in 2018 are approximately 1.24 billion years. The primary conditions contributing to PALY loss are dyslipidemia, hypertension, heart disease, diabetes, and stroke.

Conclusions

This study reveals substantial sociodemographic variations in PALYs and identifies major health conditions impacting productivity. Our results can guide targeted health interventions and policies to sustain workforce productivity and social security amid China’s aging demographics.

Plain Language Summary

As China’s population ages, understanding how health conditions affect working ability is important for policy planning. We examined how many productive working years people aged 50-64 might lose due to health problems. Using computer simulations of individual health patterns, we calculated productivity-adjusted life years (PALYs). We found that men, rural residents, and those with less education had greater potential working years but some of these groups also faced higher health-related productivity losses. Heart disease, high blood pressure, and high cholesterol caused the largest reductions in working ability across all groups. These findings help policymakers design targeted health programs and retirement policies that support older workers, potentially reducing economic burden while improving workforce productivity and individual wellbeing in China’s aging society.

Jiang et al. estimate productivity-adjusted life years for Chinese adults aged 50-64 using a microsimulation model and identify major health conditions contributing to productivity loss. Dyslipidemia and hypertension are the leading contributors to productivity loss, with substantial variations observed across sociodemographic subgroups.

Introduction

Conventionally, the description of population disease burden and the evaluation of health intervention outputs in the attribute of health-adjusted life years rely on metrics such as disability-adjusted life years (DALYs) and quality-adjusted life years (QALYs)^1–4. While these measures are instrumental in capturing the health-related quality and quantity of life, they fall short in informing policies that require an understanding of the broader economic impacts of health conditions, especially in terms of productivity⁵. This inadequacy becomes increasingly pertinent in the context of aging populations, where sustaining workforce productivity is crucial not only for economic stability but also for maintaining the balance of social security systems. To address these limitations, the concept of productivity-adjusted life years (PALYs) has emerged^5,6. PALY incorporates workforce participation, productivity losses due to absenteeism, and mortality⁵.

Complementing the insights gained from QALYs and DALYs, PALYs provide a nuanced understanding of how health conditions affect an individual’s ability to contribute productively to the economy. The application of PALYs has been explored in various contexts to better understand the productivity losses associated with specific health conditions. For instance, studies have utilized PALYs to evaluate the economic burden of chronic diseases such as diabetes and cardiovascular conditions, the results of which highlighted the sizeable impact on workforce participation and productivity^7,8. In addition to specific diseases, PALYs have been employed to assess the productivity loss of working-age smokers⁹. The rationale for using PALYs in these studies stems from the need to inform policy decisions with a metric that captures the economic consequences of health conditions, thereby enabling more targeted and effective interventions among the working population.

China is experiencing a significant demographic shift due to its rapidly aging population, which poses critical challenges for the country’s economic sustainability and social security systems¹⁰. This demographic trend intensifies the ongoing debate over China’s relatively young retirement ages, which are set at 55 for women (and in some cases as early as 50) and 60 for men, established during a period when life expectancy and economic conditions differed significantly from today. In the face of increased life expectancy and a declining workforce, policymakers are scrutinizing these retirement norms to extend working lives, thereby counteracting the shrinking labor pool and managing the rising pressures on pension systems^11,12. Understanding the late middle-aged population’s capacity to maintain a healthy and productive working life is crucial in this context. Not only does it provide insights into workforce capabilities, but also it informs strategies that can enhance economic participation among older adults.

Accordingly, we concentrate our analysis on the 50–64-year age bracket. This stage of life is a critical window in which many chronic conditions emerge, workforce participation begins to wane, and the economic consequences of health shocks are most acute. This span aligns with ongoing policy discussions about retirement-age reform and captures the cohort for whom interventions to preserve productivity may yield the greatest return. While extending the model to younger workers would broaden the scope, focusing on ages 50–64 allows us to concentrate on the nexus of health deterioration, productivity loss, and social-security pressures that is central to China’s current demographic challenges.

Equally important is identifying the health conditions that take up the largest share of the impairment of the healthy working life of this demographic group. By pinpointing which conditions are most detrimental, policymakers and healthcare providers can effectively target resources to support healthy aging and sustained workforce involvement.

Akin to other health-adjusted life year metrics, PALYs of a target population are typically estimated using state-transition models or the lifetable approach. As an approach that is commonly used for relatively complex and intricate tasks of DALY and QALY estimation, microsimulation has been ignored in the calculation of PALY by far. Microsimulation offers a distinct methodological advantage over conventional methods such as state-transition models, especially in its capacity of estimating results across diverse subgroups and quantifying PALY impacts attributable to a range of health conditions. While state-transition models are useful for calculating PALY expectancy, their reliance on aggregated data and fixed pathways often limits their ability to accurately reflect the diversity within subpopulations and capture the specific impacts of individual diseases. This limitation becomes apparent when attempting to model the complex interactions and variability inherent in real-world scenarios^13,14. In contrast, microsimulation allows for the detailed representation of individual life histories and characteristics, enabling a more granular analysis of how health conditions variably affect productivity across different demographic subgroups. Such functional features make microsimulation particularly advantageous for our study that delves deeply into the nuanced productivity impacts within China’s aging population.

To date, no studies have estimated PALYs or evaluated the impact of specific diseases on PALYs in the Chinese context. This study aimed to fill this gap by estimating the current PALY expectancy for an average individual across three age brackets (50–54, 55–59, and 60–64) and examining variations among subgroups, including sex, rural versus urban residence, and educational attainment. We also projected the total accrued PALYs for the cohort initially aged 50–64 in 2018 over a 15-year period from 2018 to 2032. Furthermore, we assessed the impact of specific diseases on both individual PALY expectancy and total accrued PALYs. By addressing these questions, this study provides crucial insights into how various health conditions influence economic productivity in China. These findings are essential for informing policy development and resource allocation strategies, as understanding these dynamics is key to designing effective interventions to promote healthy aging and enhance workforce participation. Such insights are particularly important in addressing the socioeconomic challenges posed by China’s demographic transition, thereby informing social policy making and public health strategies.

In this study, we find that the average PALYs before age 65 are 6.80, 4.35, and 1.50 years for individuals aged 50–54, 55–59, and 60–64 years, respectively, with substantial variations across sociodemographic subgroups. From 2018 to 2032, the cohort aged 50–64 years in 2018 is projected to accrue approximately 1.24 billion PALYs. We identify dyslipidemia, hypertension, heart disease, diabetes, and stroke as the primary health conditions contributing to PALY loss. These findings reveal the significant impact of chronic conditions on workforce productivity and highlight the need for targeted interventions to address sociodemographic disparities in China’s aging population.

Methods

Overview of the model

This study utilized the China Aging and Retirement Simulation (CHARISMA) model, a microsimulation model validated for projecting health and functional dependencies among China’s aging population¹⁵. The technical details of the model have been elaborated elsewhere¹⁵. Briefly, the CHARISMA model was initially designed to simulate individual trajectories related to health status and functional loss. The model was designed to realistically reflect the complex interactions between socio-demographic factors and health changes over time, which are often not captured by conventional state-transition models.

The model’s structure incorporated detailed demographic, health, and economic data, enabling it to simulate a variety of health scenarios and potential policy interventions. At the start of the simulation, the model was populated with a baseline cohort of individuals aged 50, tailored to meet the purpose of the current study. For each individual in the baseline cohort, the model tracked the development of health conditions, functional status, and productivity index (detailed in a later section). The simulation progressed in annual cycles, where the model first calculated the probabilities of developing new chronic conditions and transitioning between different levels of functional impairment based on individual characteristics, health history, and sociodemographic factors. The model then updated each individual’s health conditions and functional status accordingly in each annual cycle.

Eventually, the CHARISMA model was able to generate individual-level life courses and transitions between different health and productivity states, after which the results of individuals could be aggregated to obtain population-wide estimates. The model’s validity has been supported by its ability to reproduce observed data and generate credible forecasts. The model structure is schematized in Fig. 1 and the simulation process is illustrated in Supplementary Fig. 1. The CHARISMA model was chosen for this study due to its capability to simulate the multifaceted interactions between health conditions and economic productivity, particular to diverse demographic groups. It aligned with our research objectives by enabling the estimation of PALYs while accounting for the variations in health impacts across different segments of the population.

Fig. 1. Schematic overview of the CHARISMA microsimulation model structure for estimating productivity-adjusted life years (PALYs).

The model consists of three main components: (1) Data Input: CHARLS data preparation and analysis to derive disease and work limitation prediction equations, which are used to construct the baseline cohort; (2) Simulation: Year-to-year iteration process where each individual’s health and productivity trajectories are simulated; and (3) Output: Generation of overall PALYs estimates, subgroup analyses, disease-specific impacts, and associated uncertainty estimations. CHARLS: China Health and Retirement Longitudinal Study; CHARISMA: China Aging and Retirement Simulation.

Data and cohort

This study utilized data from the China Health and Retirement Longitudinal Study (CHARLS), a nationally representative survey of Chinese residents aged 45 and older and their partners^16,17. The CHARLS dataset provides comprehensive demographic, health, and economic information on China’s aging population, making it an ideal source for simulating health and productivity outcomes. The simulation was set to commence in 2018, which served as the base year for our analysis. When the current study was conducted, the 2018 wave was the most recent data release of CHARLS. As of November 2024, the 2020 wave had also been released. However, the 2020 survey was conducted during the COVID-19 pandemic, which resulted in a substantial omission of previously collected variables in the latest data release.

For the purposes of this study, we randomly sampled 1000 individuals from each of the age brackets (50–54, 55–59, and 60–64) using the 2018 survey data. These samples were pooled together to create a cohort for the simulation. The disease prevalence rates of the entering cohorts are displayed in Supplementary Table 1. The model operated on a yearly cycle to simulate the life course of each individual until the age of 65 within the cohort. Individuals were simulated until 65 because this was a commonly used retirement age in many upper-middle and high-income countries as well as the extended retirement age of discussion in China^11,18,19.

Individuals entered the simulation with their actual disease status as reported in the CHARLS survey, meaning the cohort included both those with and without pre-existing conditions at baseline. During each cycle, the model estimated the risks of developing 14 specific health conditions, as well as the risk of death. For individuals not already diagnosed with a particular condition, the model compared the predicted probability of developing each condition with a randomly generated number between 0 and 1 from a uniform distribution. If the random number was less than the predicted probability, the condition was considered to have developed. The development of a new condition impacted the risks of subsequent diseases and overall health status in the following cycle. This inclusive approach ensures that healthy full-time workers, those with partial work capacity, and non-participating individuals all contribute to the aggregate PALY estimates.

For the objectives of the current study, the model also accounted for PALYs by assigning each individual a productivity index based on their health status during each cycle. This index was calculated annually, with a value of 0 indicating not working, a value of 1 representing full productivity, and values between 0 and 1 reflecting reduced productivity due to missed workdays caused by health issues. This approach allowed the model to quantify the impact of health on economic productivity over time.

Productivity index prediction equation

To estimate the productivity index, which reflected both participation in working and the extent of work engagement, we first developed prediction models using regression analyses. The prediction models utilized all available observations from CHARLS waves 2011–2018 after excluding those with missing values in the dependent variable or any of the predictor variables. The dependent variable for these models was defined based on responses to specific questions in the CHARLS dataset. Participants were asked about the number of workdays missed in the past year due to health problems across three contexts: household agricultural work, employed work, and self-employed non-agricultural work. Health conditions were included as main effects without interaction terms. Testing all pairwise interactions would require 91 additional parameters (14 × 13/2), many of which would be imprecisely estimated or unidentifiable given their relatively low prevalence rates. This additive modeling approach is consistent with established microsimulation frameworks and prioritizes model stability and interpretability^20–22.

A categorical variable for working-limitation was created by combining responses to these questions, with categories defined as: no limitation in working, limitation in working, and not working. To predict these categories, we employed a multinomial logit regression model. For individuals with any limitation in working, a generalized linear model with a gamma distribution was used to predict the number of workdays missed due to health issues. In all regressions, the initial predictors included sociodemographic characteristics, conditions in the previous wave, and functional status in the last wave. These covariates were commonly used in population health status microsimulation studies^20,21,23. Sociodemographic characteristics were age, sex, marriage status, body mass index (BMI), education, smoking, and drinking. We created linear splines of age with knots at 45, 60, and 75 and linear splines of BMI with a knot at 24. We also created an indicator of sex being male. In addition, education was categorized as below-high school, high school, and above high school. More, a dummy variable was created to differentiate participants who drank at least once a day or more frequently from those who drank less often. Of note, BMI was treated as a time-varying variable. Specifically, a linear regression of BMI on the linear splines of age, the indicator of sex, and the lag value of BMI was conducted. Stepwise regressions with backward selection using a significance level of 0.1 for variable removal were conducted to identify the model specification for all multivariable analyses. The definitions and risk prediction models for diseases and functional status are also detailed in the previously referenced manuscript on the CHARISMA model¹⁵.

The productivity index was calculated using the formula ${\mathrm{I}}_{\mathrm{had}\mathrm{limitation}}\times \left(250-\mathrm{work}\mathrm{days}\mathrm{missed}\right)/250$, where 250 represents the standard number of working days in China. This index provides a quantitative measure of productivity, ranging from 0 (not working) to 1 (fully productive), with intermediate values indicating partial productivity due to health-related work absences. During each cycle of the simulation, the current cycle PALY was calculated by first predicting the working-limitation category and then calculating the productivity index. For an individual in the limitation in working category, the productivity index was equivalent to the PALY of that cycle. For individuals who died during the simulation, their productivity index was set to 0 from the time of death through age 65 to capture the complete loss of potential productivity due to premature mortality.

Analysis of PALY expectancy and population total PALYs over 2018–2032

The study estimates PALY expectancy before age 65 for three age cohorts: 50–54, 55–59, and 60–64. Individual life trajectories were simulated from the starting age until age 65 or death, whichever occurred first. For each simulated individual, the PALYs accrued over their life courses were calculated. The PALY expectancy for each cohort was then computed as the average cumulative PALYs across all simulated individuals within that cohort. The corresponding estimates by sociodemographic subgroups were also obtained. The subgroups were defined based on sex, rural and urban residence, education (no high school, high school, and college and above), and marital status (married vs. not married, divorced, or widowed).

To estimate the total PALYs for the population aged 50–64 over the period 2018–2032, the average PALY of each age cohort accrued during 2018–2032 was weighted by the population size of the age cohort. The cumulative total PALYs over the 15-year period were then computed. The projection period 2018–2032 was used because the youngest people in the baseline cohort in 2018 (age 50) would reach age 65 by 2032.

To quantify uncertainty in our estimates, we conducted a probabilistic sensitivity analysis (PSA). Unlike conventional PSA where parameter values are directly sampled, our approach involved resampling the coefficients of the underlying regression models used in the microsimulation. For each PSA iteration, coefficients for all covariates in the regression models were resampled from their respective distributions using the uncertainty estimates from the original regression analyses. The microsimulation was run with the resampled coefficients to generate new estimates of PALY expectancy and total PALYs. This process was repeated 1000 times to create distributions of the outcome measures. The distributions of the results are visualized using violin plots.

To estimate the burden attributable to specific health conditions, we employed a counterfactual approach. For each condition of interest in this analysis, the risk of developing the condition was set to zero in the simulation model, while the risks of all other conditions were unchanged. The microsimulation was run with this modified risk profile to estimate PALY expectancy and total PALYs under the counterfactual scenario. The burden attributable to the condition in question was calculated as the difference between the baseline estimates and the counterfactual estimates. This process was performed for each of the 14 health conditions. The loss of PALY expectancy for each of the age cohort overall and by sociodemographic subgroups were also obtained. Additionally, we estimated the total PALY loss for the population aged 50–64 over the period 2018–2032.

Statistics and reproducibility

The analyses were conducted using Stata 15 (Stata Corp., College Station, Texas, USA) and Python 3.11. The details of statistical analyses have been described above. Reproducibility was verified by re-running the code to confirm that results remained consistent.

Human Ethics

This study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Biomedical Research, School of Public Health (Shenzhen), Sun Yat-sen University, with reference number No. 039 [2022]. The requirement for informed consent was waived by the same committee as the study utilized secondary, de-identified data. The original CHARLS study was approved by the Ethics Committee of Peking University (IRB00001052-11015), and informed consent was obtained from all participants and kept by the CHARLS office at Peking University.

Results

PALY expectancy and total PALYs accrued

The overall and subgroup estimates of PALY expectancy before age 65 for the age cohorts are listed in Table 1 and the distributions of the results are plotted in Fig. 2. The estimated PALY expectancy before age 65 decreased with advancing age among the cohorts studied (Table 1). For the overall population, individuals aged 50–54 had an average of 6.80 PALYs remaining before age 65. This expectancy reduced to 4.35 PALYs for those aged 55–59 and further to 1.50 PALYs for the 60–64 age group.

Table 1.

Overall and subgroup PALYs for each age bracket

	50–54	55–59	60–64
Overall	6.80	4.35	1.50
By sex
female	6.13	3.90	1.25
male	7.45	4.80	1.75
By rural and urban
urban	5.47	3.38	1.09
rural	7.78	5.11	1.80
By education level
No high school	7.05	4.56	1.51
High school and vocational education	5.62	3.77	1.53
College and above	5.96	3.90	0.87
By married or partnered
Not married and partnered	6.09	4.28	1.29
Married or partnered	6.95	4.37	1.55

Fig. 2. Distribution of the mean productivity-adjusted life year (PALY) expectancy before age 65 across sociodemographic subgroups for three age cohorts (50–54, 55–59, and 60–64 years).

The distribution of PALY expectancy estimates from probabilistic sensitivity analysis for different sociodemographic subgroups.

Nontrivial variations in PALY expectancy were observed across different sociodemographic subgroups, including sex, residence (urban vs. rural), educational attainment, and marital status. These differences were consistent across all age cohorts. Males consistently exhibited higher PALY expectancy than females across all age groups. For the 50–54 age cohort, males had an average PALY of 7.45 years, while females had 6.13 years. This trend continued in the 55–59 cohort (males: 4.80 years; females: 3.90 years) and the 60–64 cohort (males: 1.75 years; females: 1.25 years). Rural residents had higher PALY expectancy compared to urban residents in each age cohort. Among those aged 50–54, rural individuals had an average PALY of 7.78 years, significantly higher than the 5.47 years for urban residents. In the 55–59 cohort, rural residents had 5.11 years versus 3.38 years for urban residents. The 60–64 cohort showed a similar pattern (rural: 1.80 years; urban: 1.09 years).

PALY expectancy also varied notably with education level. Individuals without a high school education generally had higher PALYs. In the 50–54 age group, those without a high school diploma had an average PALY of 7.05 years, compared to 5.62 years for those with high school but not more advanced education, and 5.96 years for those with college or higher education. In the 55–59 cohort, the PALYs were 4.56 years (no high school), 3.77 years (high school), and 3.90 years (college and above). In the 60–64 age group, individuals with college education had the lowest PALY expectancy (0.87 years) whereas those with high school but not more advanced education and those without high school education had PALYs of 1.53 and 1.51, respectively.

In terms of marital status, being married or partnered was associated with slightly higher PALY expectancies. In the 50–54 age cohort, married or partnered individuals had an average PALY of 6.95 years, compared to 6.09 years for those not married or partnered. The difference was minimal in the 55–59 cohort (married/partnered: 4.37 years; not married/partnered: 4.28 years), and slightly more pronounced in the 60–64 cohort (married/partnered: 1.55 years; not married/partnered: 1.29 years).

The PSA results showed that the uncertainty ranges were relatively narrow, and there were few overlaps between the distributions of the PALY estimates of different age cohorts and sociodemographic subgroups. This indicates that the observed differences in PALY expectancies are consistent even when accounting for the inherent uncertainties in the model parameters. Importantly, the variations were not large enough to alter the overall patterns observed.

The total PALYs accrued by the cohort initially aged 50–64 in 2018 over the period from 2018 to 2032 are presented in Fig. 3. The cohort accrued approximately 1.24 billion PALYs in the base case. Based on the PSA analyses of total PALYs accrued, the lower bound of the 95% CI exceeded 1 billion PALYs, whereas the upper bound could reach as high as 1.4 billion PALYs.

Fig. 3. Distribution of total productivity-adjusted life years (PALYs) accrued by the cohort aged 50–64 years from 2018 to 2032.

The shaded area represents the distribution of total PALYs from probabilistic sensitivity analysis (PSA). The blue dashed line indicates the mean estimate (1.24 billion PALYs), while the red and green dash-dot lines show the lower and upper bounds of the 95% confidence interval (1.01–1.47 billion PALYs), respectively.

PALY loss

Figure 4 shows the average loss of PALY expectancy attributable to the top five conditions for each age group (50–54, 55–59, and 60–64). The underlying distributions of individual PALY values are presented in Supplementary Fig. 2. In the 50–54 age group, dyslipidemia resulted in the highest average PALY expectancy loss at approximately −0.69 years, followed by hypertension (−0.61 years), heart disease (−0.55 years), diabetes (−0.39 years), and kidney disease (−0.31 years). For individuals aged 55–59, heart disease caused an average PALY expectancy loss of about −0.21 years, with stroke (−0.21 years), dyslipidemia (−0.19 years), hypertension (−0.19 years), and kidney disease (−0.12 years) also ranking among the top five PALY-impairing conditions. Among those aged 60–64, heart disease was again the leading cause of PALY expectancy loss at −0.08 years, followed by hypertension, diabetes, and dyslipidemia, all of which reduced the expected PALY for this age cohort by around −0.05. The fifth most PALY-decreasing condition among this age cohort was asthma (−0.02 years).

Fig. 4. Average productivity-adjusted life year (PALY) loss attributable to the top 5 conditions by age cohort.

Horizontal bars represent the mean PALY loss for the five conditions causing the greatest productivity impact within each age group (50–54, 55–59, and 60–64 years). Error bars indicate the uncertainty around each estimate (±standard error), reflecting uncertainty in the simulated counterfactual estimates rather than sample variability (sample size is not applicable as estimates are derived from microsimulation comparing scenarios with and without each condition across the full cohort).

The loss of PALY expectancy for each age cohort by sociodemographic subgroups are plotted in Supplementary Figs. 3–6. Male population, rural residents, and people with college or higher education had relatively high PALY expectancy loss due to diseases, with the conditions causing the most PALY loss in each age cohort resembling those in the main analysis.

The estimates of disease-specific loss of total PALYs accrued over 2018 to 2032 among the initial cohort are depicted in Fig. 5, along with the distributions of the estimates. There were substantial variations across the diseases in terms of their damage to the productivity of the late middle-aged population. The figures ranged from over 80 million PALYs for dyslipidemia and hypertension to less than five million PALY loss for arthritis.

Fig. 5. Distribution of condition-specific productivity-adjusted life year (PALY) losses accrued by the cohort aged 50–64 years from 2018 to 2032.

The distributions of total PALY losses for 14 health conditions, ordered by mean loss magnitude. Dyslipidemia and hypertension show the largest mean losses (>80 million PALYs).

Discussion

In this study, we estimated the PALYs for the aging Chinese workforce using a microsimulation model and analyzed the impacts of specific health conditions across different age brackets. Our findings reveal consequential productivity losses due to health conditions, with heart disease, hypertension, and dyslipidemia being the most impactful across the age groups studied. These results underscore the critical role of health in determining economic productivity and highlight the utility of using PALYs to quantify the economic burden of disease in China.

Understanding the productivity potential of China’s aging workforce is a critical social issue not only public health-wise but also economically in the context of the country’s demographic transition¹⁰. By identifying the specific conditions that most substantially impair productivity, the PALY estimates from this study provide a foundation for developing targeted health interventions to sustain economic productivity. For example, workplace health screening programs could prioritize early detection of these conditions. Although our study measured PALYs while previous research focused on other metrics of healthy life years, both approaches identified similar leading causes of health-related losses²⁴.

Beyond condition-specific patterns, our analysis also revealed substantial sociodemographic variations in PALYs across sex, residence, and educational attainment. It is important to note that we applied uniform disease-specific coefficients of productivity-loss weights across all demographic groups in the productivity index prediction equation. Therefore, these subgroup differences in PALYs reflect variations in disease incidence, mortality rates, and baseline workforce participation rather than differential per-case productivity impacts by subgroup. These disparities suggest that a one-size-fits-all approach to supporting workforce productivity may be inadequate, and policymakers should consider developing differentiated strategies for specific population subgroups to effectively promote healthy aging and sustained workforce participation, especially from the perspective of disease incidence reduction. Of note, the longer PALY expectancy of several sociodemographic subgroups may have only reflected the lower probabilities of transitioning out of workforces, which are themselves indicators of disadvantageous economic conditions²⁵. For example, rural residents may be less likely to retire early than urban residents due to a relatively low level of social security²⁶. However, this should not be interpreted as the case that rural residents are subject to PALY loss attributable to diseases to a lesser extent. In fact, rural residents experience higher PALY expectancy losses compared to their urban counterparts based on the corresponding subgroup analysis results, which call out the importance of understanding the underlying causes of PALY disparities and formulating policies accordingly.

Since the current study represents the first effort to estimate both individual-level PALY expectancy and population-level PALY accrual in China, the results cannot be directly compared with existing evidence in the literature. As such, this study provides unique insights specific to China, which is undergoing rapid demographic changes. While similar studies in other regions have highlighted the productivity impacts of diseases and their risk factors among aging populations, they predominantly engaged a one-disease-per-study approach^7,8,27,28. Perhaps the most closely related prior work is a Finnish study that estimated the PALY loss of eight conditions using a comprehensive approach²⁹. Expanding on their analysis, the current study employed a microsimulation model to not only analyze the impacts of multiple diseases simultaneously, but also to estimate individual PALY expectancy beyond just population-level PALY measures. By maintaining the same setting and parameters, the estimates of PALY loss attributable to diseases are relatively coherent.

These findings point to explicit policy implications, particularly in light of ongoing debates about retirement age and workforce participation in China^11,18. By demonstrating the productivity burden of specific health conditions, this study highlights the importance of implementing health interventions that target the most impactful diseases in the context of delaying retirement age. In a related aspect, our findings reiterate the priority of disease prevention among late-middle age population. Specifically, the remaining healthy working years of the 50-54 and 55-59 age cohorts are outnumbered by the remaining years they are required to work if they are to retire at the age of 63, which is the new standard retirement age that will be implemented in the near future³⁰. Accordingly, preventing these cohorts from productivity loss due to health issues is vital for the society. Finally, rather than implementing a blanket increase in retirement age, policymakers should consider a flexible system that accounts for individual health status and productivity potential. For example, the social security and healthcare systems can create incentives for healthier individuals to remain in the workforce longer, while providing support and alternative options for those with health conditions that significantly impact their productivity.

We adopted an associative modeling framework in this analysis, which was chosen for several reasons. First, in the absence of exogenous shocks, causal-inference methods such as propensity-score matching or inverse-probability weighting generally produce estimates very similar to fully-adjusted regression models. Second, leading microsimulations in the areas of health economics and policy frequently rely on associative models to inform policy^20–22. Due to data restrictions, this is not necessarily the optimal course to take. However, this approach represents one of the most rigorous and feasible methods available for our study objectives of quantifying population-level PALY burden and informing policy priorities.

Several limitations of the current study must be noted. For example, an assumption inherent in the microsimulation model was that the statistical relationship between productivity and predictors remained unchanged over the study course. Future research could extend the model to provide a more comprehensive understanding of the factors affecting productivity in aging populations. Another important limitation is that health conditions in CHARLS were self-reported, which may lead to potential measurement error due to recall bias, which could result in either under- or over-estimation of health conditions’ prevalence and their associated impacts on productivity. Similarly, the number of work days missed due to health issues was self-reported and may be subject to recall bias. Respondents may underreport short absences or overreport longer periods, potentially leading to measurement error in our productivity index. Additionally, the current analysis focused on quantifying the projected population-level status of PALYs. While this offers valuable insights, extending the model to explore the long-term impacts of specific health interventions on PALYs could further inform the development of targeted policies and guide the strategic allocation of resources. More, while our productivity index incorporated disease-attributable absenteeism, we did not capture presenteeism (reduced on-the-job performance), which may lead us to underestimate the total work-related productivity losses associated with chronic conditions. Furthermore, as mentioned earlier, it is important to note that the current study was not designed to make causal inferences. As such, the results based on subgroup analyses should be interpreted as descriptive patterns rather than definitive causal effects. Future research employing quasi-experimental study designs could establish a stronger evidence base for policy decisions. Finally, the wide uncertainty in our modeled PALYs, the lower bounds of which occasionally crossing zero, reflect imprecision in input parameters and underscoring the need for additional data to refine these estimates.

Conclusions

This study reveals substantial sociodemographic variations in PALYs and identifies prevalent health conditions that are associated with substantial losses of PALY. The findings underscore the urgent need for targeted interventions to support healthy aging and enhance workforce participation, while highlighting the importance of flexible retirement planning.

Author contributions

Y.J. led the study conception and design, secured funding, conducted data analysis, developed the simulation model, performed visualization, managed project administration, and wrote the original draft. Y.G. contributed to study conception, manuscript revision, and supervision. L.S. participated in study design, manuscript revision, and supervision. All other authors (B.H., H.G., W.H., and S.J.) contributed to data collection, validation, and manuscript revision. All authors reviewed and approved the final version of the manuscript.

Peer review

Peer review information

Communications Medicine thanks the anonymous reviewers for their contribution to the peer review of this work.

Data availability

This study used data from CHARLS, which is publicly available to registered users from the website of CHARLS. Source data underlying main Figs. 2–5 can be accessed at https://data.mendeley.com/datasets/f7tvry2425/1.

Code availability

The custom computer code of the main program used in this study has been provided to editors and reviewers for the peer review process. The code files of data cleaning and simulation can be accessed at https://data.mendeley.com/datasets/mzsxv67czt/1³¹.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s43856-025-01294-8.