Educational attainment and male-female health-survival paradox among older adults in China: a nationally representative longitudinal study

Abstract

Background

The male-female health-survival paradox is characterized by the phenomenon where “women get sicker, but men die quicker.” Health expectancy, as a composite metric that encompasses both the quantity and quality of life, serves as a unique tool for analyzing this gender paradox. In this study, we investigate the relationship between educational attainment and the gender paradox among older adults in China.

Methods

Utilizing data from the China Health and Retirement Longitudinal Study (CHARLS), we focused on community-dwelling individuals aged 60 and above. Health was assessed using the Activities of Daily Living (ADLs). Educational attainment was dichotomized into low (primary education and below) and high (secondary education and above). We controlled for demographic, socioeconomic, and health behaviors confounders. Microsimulation techniques were employed to estimate total life expectancy (TLE), disability-free life expectancy (DFLE), and health ratio.

Results

In China, educational attainment was positively associated with TLE and DFLE, with these benefits being more pronounced in females. Among individuals with lower educational attainment, females had significantly greater TLE (female-male difference: 3.82 years, 95% CI: 3.68 to 3.96) and DFLE (2.91 years, 95% CI: 2.78 to 3.04), but a lower health ratio (-2.14%, 95% CI: -2.41% to -1.87%) compared to males. In contrast, females with higher educational attainment not only lived longer but also healthier. Among these individuals, females had significantly greater TLE (5.89 years, 95% CI: 5.71 to 6.08), DFLE (6.02 years, 95% CI: 5.84 to 6.19), and a more favorable health ratio (95% CI: 2.60% to 3.19%) compared to males.

Conclusions

Education plays a crucial role in enabling females to overcome disadvantages associated with the gender paradox in China. Enhancing gender equality in educational opportunities is expected to promote healthy longevity among females in the country.

Background

“Women get sicker, but men die quicker” [1]. Since at least the mid-eighteenth century, empirical evidence has consistently shown that women tend to outlive men [2]. Despite men’s higher mortality rates, women are more likely to experience worse health, spending a larger proportion of their remaining life in poor health [3, 4]. This phenomenon is known as the male-female health-survival paradox [5]. While an extensive body of literature has confirmed the existence of this gender paradox, the underlying mechanisms driving this phenomenon remain unclear [2, 4].

As a composite metric encompassing both the quantity and quality of life, health expectancy is regarded as a unique tool for analyzing the gender paradox [5, 6]. Health expectancy is routinely expressed in terms of life expectancy in a specified state of health [7]. In comparison to individual health or mortality indicators, health expectancy holds two notable advantages. First, it can simultaneously reflect the impact of factors on both health and survival. Second, it allows for the calculation of health ratio (i.e., the proportion of life expectancy in optimal health states in total remaining life expectancy), serving as an important indicator of healthy aging. When measured through health expectancy, existing studies consistently show that, while females enjoy a higher total life expectancy, they experience a smaller proportion of life expectancy in optimal health than males [5, 8, 9]. Moreover, the extent of the gender paradox varies depending on factors such as age, the specific health measure used, and the broader macro context [10, 11].

Although many studies have examined the relationship between education attainment and health expectancy, few have specifically utilized health expectancy to analyze the relationship between educational attainment and the gender paradox. Previous research shows that education is a powerful social determinant, extending both total life expectancy and life expectancy in optimal health states [12]. In most of these studies, gender is used as a grouping variable, but the gender paradox is not their primary focus. As a result, these studies often fail to provide sufficient indicators related to the paradox, such as total life expectancy, health expectancy, and health ratio. Our comprehensive literature review reveals several conflicting findings regarding the potential heterogeneity in the gender paradox across different levels of educational attainment. A few studies suggest that females experience a greater increase in both total life expectancy and health ratio compared to males [12,13,14], while others propose a different pattern [15, 16].

Therefore, using nationally representative longitudinal data, this study aims to analyze the role of education in the gender paradox in China. Specifically, employing multinomial logistic regression analysis, the study examines the impact of gender, education, and their interaction on health status and mortality, while controlling for confounding factors. Based on the regression results, microsimulation techniques were used to calculate the person-years and healthy person-years for the remaining life across gender and education subgroups. The study further estimates indicators such as gender-specific and education-specific total life expectancy, disability-free life expectancy, and health ratios.

Methods

Study population

We utilized data from the China Health and Retirement Longitudinal Study (CHARLS), a nationally representative longitudinal survey [17]. CHARLS employs a four-stage, stratified, cluster sampling method to recruit community-dwelling residents aged 45 years and above in China. CHARLS collects comprehensive information on respondents’ demographic background, family characteristics, health behavior and status, as well as retirement information. The national baseline survey was conducted in 2011–2012, and follow-up surveys every two to three years. To date, CHARLS has publicly released data in five waves (2011, 2013, 2015, 2018, and 2020) [18]. Given the influence of the COVID-19 pandemic on individuals’ lifestyles and health, the study analyzed the CHARLS 2015 and 2018 data to derive more robust conclusions.

The subjects of our study comprise community-dwelling individuals aged 60 years and above. In CHARLS 2015, a total of 9,867 older individuals were interviewed, among whom 9,142 were re-interviewed in or deceased before CHARLS 2018, and 725 were lost to the follow-up survey. Moreover, 358 samples were excluded due to missing data on gender, education, control variables, and sample weights and 33 samples were dropped due to missing data on the Activities of Daily Living (ADL) scale. Finally, the sample sizes for the analyses is 8,751 (Fig. 1).

Fig. 1

Flow diagram of study subjects. ADL = Activities of Daily Living

Measurement of health states

Health states had three categories: non-disabled, disabled, and deceased. Disability is measured using the ADL scale. Individuals requiring assistance with any of the six tasks (dressing, bathing, eating, indoor mobility, toileting, and continence) are classified as disabled [19, 20].

Measurement of gender and educational attainment

Gender is self-reported and binary (male and female). China implemented nine-years compulsory education policy in 1986, when our study subjects were above 30 years. Therefore, their level of education is generally low. Among the respondents, 52.21% were illiterate or had not finished primary education, 21.97% finished primary education, 19.00% secondary education, and 6.82% tertiary education. The educational attainment of female is even lower. For example, among female individuals aged 75 and above, only 2.60% (27 respondents) finished tertiary school. To avoid estimation bias caused by the small sample size of certain subgroups, we dichotomized educational attainments into two levels: low (i.e., primary education and below) and high (i.e., secondary education and above).

Control variables

According to existing literature [4, 9, 21], this study incorporated three sets of confounding factors. First, demographic variables included five-year age groups (60–64 years, 65–69 years, 70–74 years, and 75 years and older), marital status (unmarried and married), and residence (urban and rural).

Second, socioeconomic variables included current or pre-retirement job (farmer, worker, cadre, and others), per capita household expenditure (PCHE; categorized as low, medium, and high), health insurance (government medical insurance or urban employee medical insurance, and others), and house ownership (yes and no). Consistent with OCED, PCHE was calculated by dividing total household yearly consumption by the square root of household size [22], and a hot deck method was employed for missing value imputation. Subsequently, PCHE was classified into three categories using tertiles.

Third, health behavior variables included drinking (never or rarely, less than once per month, and more than once per month), smoking (never, < 20 cigarettes per day, and ≥ 20 cigarettes per day), normal weight (no [BMI < 18.5 or BMI ≥ 24], yes [18.5 ≤ BMI < 24], and missing), normal sleep (no [sleep duration below 6 h or above 8 h], yes [sleep duration between 6 and 8 h], and missing), and a social activity score. Respondents reported the frequency of eleven social activities in the last month, with scores assigned as follows: “almost daily” (3), “almost every week” (2), “not regularly” (1), and “never” (0). The total social activity score ranged from 0 to 33 and was divided into four categories (0, 1–8, 9 and above, and missing). Exercise was not included in the model, as the related questions were only presented to half of the sampled households.

Detailed definitions of these variables can be found in the appendix (Supplemental Table 1).

Statistical analysis

This study employed a microsimulation technique to estimate multistate-life table (MSLT) functions, encompassing total life expectancy and health expectancy [23].

First, we conducted multinomial logistic regressions with the health states in 2018 (non-disabled, disabled, and deceased) as the dependent variable. The independent variables included health states in 2015 (non-disabled and disabled), gender, education, and control variables encompassing demographic, socioeconomic, and health behavior factors. Using the model results, we performed a post-estimation analysis to derive the three-year transition probability matrices for each age-gender-education-specific group. Based on the model specification, these matrices accounted six types of state transitions from 2015 to 2018: non-disabled to non-disabled, non-disabled to disabled, non-disabled to deceased, disabled to non-disabled, disabled to disabled, and disabled to deceased. We then applied transition intensity to convert the three-year transition probability matrices into one-year transition probability matrices [24, 25].Considering the issue of under-reported mortality in longitudinal surveys, we adjusted the one-year matrices using Chinese death probability data from the World Population Prospects 2022 (see Supplemental Part 1 Technical Appendix).

Second, we performed a microsimulation analysis to simulate the health trajectories of a cohort of 100,000 individuals aged 60 years. To ensure the stability of the results, the gender- and education-specific distribution of health states at age 60 was derived from observed prevalence data among individuals aged 55 to 65 from the CHARLS 2015 wave. Each individual’s health trajectory from age 60 until death (or age 100) was determined by the annual transition probability matrices. The microsimulation generated complete records of health trajectories for the entire cohort. By summing the per capita person years and per capita person years without disability between ages 60 and 100, we calculated total life expectancy and disability-free life expectancy at age 60, both stratified by gender and educational attainment. Finally, for this simulated cohort of 100,000 individuals, 95% confidence intervals were estimated using a bootstrap method with 1,000 replications.

The transformation from three-year to one-year probability matrices was executed using Matlab (version 9.13), while all other analyses were performed using Stata 16.0. Sampling weights, incorporating household and individual response adjustments, were applied in the analysis.

As this study exclusively utilized public datasets, no additional institutional review board approval was deemed necessary. Ethical approval for CHARLS waves was previously obtained from the Biomedical Ethics Review Committee of Peking University (Beijing, China; IRB00001052–11015).

Results

The baseline proportions for disability was 10.2% (Table 1). Approximately 50% of the participants were male, and 80% had an educational attainment of primary school and below. All baseline characteristics exhibited significant bivariate associations with health outcomes in 2018, with the exception of house ownership and drinking (see Supplemental Table 2).

Table 1 Participant characteristics (n = 8,751)

We conducted four multinomial logistic regression models (Table 2 and Supplemental Table 3). In Model 1, we included baseline health states, age groups, and gender. The results indicated that females had a significantly lower probability of death (RRR: 0.67, 95% CI: 0.52–0.88) but tended to have a higher probability of disability (RRR: 1.10, 95% CI: 0.91–1.31). Given that health expectancy is a combination of both disability and mortality, these findings suggest that females live longer but with more years spent in disability. In Model 2, education emerged as a significant protective factor, notably reducing the likelihood of both disability (RRR: 0.73, 95% CI: 0.56–0.96) and death (RRR: 0.54, 95% CI: 0.41–0.72). To explore whether the health benefits of education differ between genders, Model 3 incorporated an interaction term between gender and education. The results revealed that females derived significantly greater protection from education against disability, while this pattern was not observed for mortality. These findings remained consistent after adjusting for demographic, socioeconomic, and health behavior factors in Model 4.

Table 2 Adjusted, weighted relative-risk ratios (RRRs) for health states in 2018 (n = 8,751, RRR [95%CI])

Following regression analysis, microsimulation was conducted to derive health trajectories for simulated individuals. Figure 2 illustrates person years and healthy person years expected at each age based on education level and physical health. The enclosed area formed by the person years curve, x-axis, and y-axis represents total life expectancy. Likewise, the area enclosed by the healthy person years curve, x-axis, and y-axis signifies disability-free life expectancy. Educational attainment improvements resulted in outward shifts for all curves, indicating a positive impact on health and survival. Notably, the extent of curve movement was more pronounced for females than males in physical health.

Fig. 2

The person years (PY) and healthy person years by educational attainment. The estimates are based on data obtained from microsimulation.edu = education

At the age of 60, females exhibited a higher total life expectancy (female: 23.56 [95% CI: 23.49 to 23.64]; male: 19.49 [19.42 to 19.57]) and disability-free life expectancy (female: 20.77 [20.70 to 20.85]; male: 17.20 [17.13 to 17.27]) compared to males (Fig. 3; Table 3). The difference between females and males in health ratio was − 0.40% (95% CI: −0.60% to −0.19%) for physical health. This indicates that females lived a significantly lower proportion of their remaining life in optimal health states compared to males.

Fig. 3

The total life expectancy (TLE) and disability-free life expectancy (DFLE) by gender and education. The estimates are based on data obtained from microsimulation. The 95% confidence intervals are calculated by bootstrap method with 1000 replications. edu = education

Table 3 Total life expectancy (TLE), disability-free life expectancy (DFLE) and health ratio by gender and education (estimates, 95% CI)

Educational attainment was positively associated with total life expectancy, disability-free life expectancy (Fig. 3; Table 3). The health benefits derived from enhanced education were more pronounced in females compared to males. Among those with high educational attainment, females exhibited significantly higher total life expectancy (females-males differences: 5.89 [95% CI: 5.71 to 6.08), disability-free life expectancy (6.02 [5.84 to 6.19]) than males. In this subgroup, the female-male difference in health ratio increased to 2.89% (95% CI: 2.60–3.19%) for physical health.

Discussion

Utilizing nationally representative longitudinal data and employing microsimulation techniques, this study extends the existing literature by exploring the relationship between education and the gender paradox among older adults in China [8, 9, 11]. While studies focusing on educational disparities in total life expectancy and health expectancy routinely present gender-stratified results that suggest potential heterogeneity in the gender paradox related to education, the gender paradox is not their primary focus. As a result, the evidence is usually insufficient and often goes unnoticed [12, 15, 16]. This study shows that, due to the greater health benefits that females derive from education, the gender paradox disappears among those with higher educational attainment, with females exhibiting a clear advantage in healthy longevity, as indicated by higher total life expectancy, disability-free life expectancy, and health ratio. This study enhances our understanding of the mechanisms shaping the gender paradox.

Scholars have posited explanations for the gender paradox, which are rooted in biological, social, and psychological interpretations, alongside methodological considerations [3]. Empirical studies indicate that gender disparities in health reporting behavior can only explain a marginal proportion of the observed paradox [2]. The biological mechanism is a prominent explanation for the gender paradox, suggesting that women may be more predisposed to less fatal but more disabling conditions, such as arthritis, whereas men may be more susceptible to diseases strongly associated with mortality. Education, likely in conjunction with biological mechanisms, contributes to the formation of the gender paradox within the general population. This study shows that education is beneficial in promoting health and preventing mortality. The educational attainment of females tends to be lower than that of males in China. In our study sample, the proportion of males with a junior high school education or above is 33.10%, more than twice the corresponding proportion for females (16.00%). Consequently, within the overall population, the mortality disadvantage for males under the influence of biological mechanisms is mitigated, while the advantage of better health status is amplified [3, 4].

Among individuals with higher educational attainment, females in China show a complete advantage in healthy longevity, enjoying not only higher total and disability-free life expectancy but also higher health ratio. These shifts are attributed to the greater health and survival benefits that females derive from education. The pronounced health and survival advantages accruing to females with higher educational attainment, compared to their male counterparts, align closely with findings from prior studies on educational disparities in total life expectancy and health expectancy [12,13,14]. For example, Guralnik et al. showed that, in the United States, females experienced greater increases in both total and active life expectancy from education compared to males [12]. Andrade et al. found consistent results in Brazil [14]. Recognized as a modifiable social determinant, education is frequently advocated as a global policy objective, exemplified by initiatives such as the Sustainable Development Goals. This study underscores the profound significance of advancing gender equality in education for fostering healthy longevity and controlling healthcare and long-term care expenditures among female population.

The theory of resource substitution provides insight into understanding this phenomenon. According to this theory, education holds paramount importance for the health and survival of individuals in a disadvantaged position. Females generally face relative disadvantages compared to males. With improved education, they gain greater access to resources and opportunities, thereby increasing their ability to adopt a healthy lifestyle [26]. Nevertheless, it is noteworthy that certain studies have reported contradictory outcomes [15, 16].In addition to methodological differences—such as variations in data sources, study populations, measures of health outcomes, and analytical techniques—macro social and institutional factors may also contribute to these inconsistencies [27]. A comparative analysis demonstrated that educational inequalities in disability-free life expectancy, favoring those with higher educational attainment, are more pronounced among women than men in Central and Eastern European countries. In contrast, findings for Western European countries remain inconclusive, emphasizing the influential role of macro-level determinants in either alleviating or exacerbating educational disparities in health and survival [28]. Therefore, future investigations are warranted to extrapolate and validate our findings in diverse international contexts [29].

The study has two strengths. First, health expectancy was estimated using microsimulation method, enabling the modeling of transitions from non-healthy to healthy states [23]. Within our sample, 38.89% of older individuals with disabilities experienced functional recovery after a three-year period. This microsimulation approach yields more precise estimations of health expectancy in comparison to Sullivan’s method. Second, most health expectancy studies typically incorporate only a limited set of variables, such as age, gender, and education, while more recent studies additionally control for several confounding factors [8, 9, 11, 30].This study included a broader range of demographic, socioeconomic, and health behavior factors in the estimation model, further improving the precision in identifying the relationship between education and the gender paradox [2, 4, 21].

This study has several limitations. First, uncontrolled confounding factors may have introduced biases into the research findings. For instance, individuals with strong self-management skills may have an advantage in both education and health outcomes. Unfortunately, due to data limitations, we were unable to control for these confounding factors. Second, the study cohort consists of individuals born before 1955, who experienced significant historical events, such as the Chinese famine of 1959–1961. The selective mortality resulting from these events may have influenced our findings. Third, we adjusted the death probabilities only by age and gender, without considering education, which may introduce bias into our results. The UN World Population Prospects does not provide death probabilities stratified by education level. Furthermore, existing literature lacks data on the probability of under-reporting deaths by education among older adults in China. Without such information, we were unable to adjust death probabilities based on education level. Previous studies have often overlooked the under-reporting of deaths, leading to overestimations of both total and disability-free life expectancy [31, 32]. Although our adjustment is not perfect, it does enhance the accuracy of the estimates. For instance, after adjusting for death probabilities, life expectancy at age 60 for females changed from 28.88 to 23.56 years, which closely aligns with the UN figure of 23.58 years.

Conclusions

There exists a male-female health-survival paradox among community-dwelling older adults in China. Education promotes both health and survival, with females deriving greater benefits from higher educational attainment compared to their male counterparts. Among individuals with secondary education and above, females exhibit a clear advantage in healthy longevity, enjoying not only higher total and disability-free life expectancy but also a higher health ratio than males. These findings underscore the importance of promoting gender equality in educational opportunities, which is expected to improve healthy longevity among females in the country. Furthermore, the conclusions drawn from this study necessitate additional validation in diverse institutional and cultural contexts.