Rural–Urban Disparities in Multimorbidity Associated With Climate Change and Air Pollution: A Longitudinal Analysis Among Chinese Adults Aged 45+

Abstract

Background and Objectives

Chronic conditions and multimorbidity are increasing worldwide. Yet, understanding the relationship between climate change, air pollution, and longitudinal changes in multimorbidity is limited. Here, we examined the effects of sociodemographic and environmental risk factors in multimorbidity among adults aged 45+ and compared the rural–urban disparities in multimorbidity.

Research Design and Methods

Data on the number of chronic conditions (up to 14), sociodemographic, and environmental factors were collected in 4 waves of the China Health and Retirement Longitudinal Study (2011–2018), linked with the full-coverage particulate matter 2.5 (PM_2.5) concentration data set (2000–2018) and temperature records (2000–2018). Air pollution was assessed by the moving average of PM_2.5 concentrations in 1, 2, 3, 4, and 5 years; temperature was measured by 1-, 2-, 3-, 4-, and 5-year moving average and their corresponding coefficients of variation. We used the growth curve modeling approach to examine the relationship between climate change, air pollution, and multimorbidity, and conducted a set of stratified analyses to study the rural–urban disparities in multimorbidity related to temperature and PM_2.5 exposure.

Results

We found the higher PM_2.5 concentrations and rising temperature were associated with higher multimorbidity, especially in the longer period. Stratified analyses further show the rural–urban disparity in multimorbidity: Rural respondents have a higher prevalence of multimorbidity related to rising temperature, whereas PM_2.5-related multimorbidity is more severe among urban ones. We also found temperature is more harmful to multimorbidity than PM_2.5 exposure, but PM_2.5 exposure or temperature is not associated with the rate of multimorbidity increase with age.

Discussion and Implications

Our findings indicate that there is a significant relationship between climate change, air pollution, and multimorbidity, but this relationship is not equally distributed in the rural–urban settings in China. The findings highlight the importance of planning interventions and policies to deal with rising temperature and air pollution, especially for rural individuals.

Translational Significance: This study examines the effects of sociodemographic and environmental risk factors in multimorbidity among Chinese adults aged 45+ and compared the rural–urban disparities in multimorbidity related to temperature and particulate matter 2.5 (PM_2.5) exposure. Our findings show rural respondents have a higher prevalence of multimorbidity related to rising temperature, whereas PM_2.5-related multimorbidity is more severe among urban ones. Our study emphasizes the relationship between climate change, air pollution, and multimorbidity is not equally distributed in the rural–urban settings in China, which highlights the importance of planning interventions and policies to deal with rising temperature and air pollution, especially for rural individuals.

Along with increasing industries worldwide, the burning of fossil fuels leads to substantial changes to the global climate system, including the emissions of greenhouse gases (e.g., CO₂) and air pollutants (e.g., particulate matter 2.5 [PM_2.5], ozone, methane; Gibson, 2015; Kinney, 2018; Yu et al., 2019). The disruption of the ecological system, once a theoretical concern, is now challenging our health safety with increasing mortality and morbidity due to environmental pollution (Brønnum-Hansen et al., 2018; Pope et al., 2019) and changes in temperature, precipitation and occurrence of heat waves, flooding, droughts, and wildfires (Solomon & LaRocque, 2019; Woodward et al., 2014).

Of these changes in climate, the rise in global temperatures is a leading cause of concern for chronic health, such as respiratory and heart diseases (de Sario et al., 2013; Schwartz et al., 2004). Further evidence shows higher deaths due to abnormal temperatures, and the associations between temperature and mortality were higher among older adults and in cities with fewer green spaces (Zanobetti et al., 2012). With the development of industrialization and the growing number of global populations, the surface temperature will rise 1.1 to 6.4°C by the end of the twenty-first century (Redshaw et al., 2013). It is necessary to understand the role of climate/temperature changes on human health to reduce future potential health care expenses.

In parallel, air pollution was also identified as one of the modifiable risk factors that could be prevented to delay potential chronic diseases (Livingston et al., 2020; Weichenthal et al., 2013). For example, living in areas with higher levels of air pollutants is associated with higher prevalence of hypertension (B.-Y. Yang et al., 2018), respiratory (Chen et al., 2020), or cardiovascular diseases (F. Liang et al., 2020). As chronic diseases often cluster due to common risk factors, health effects of air pollution are not likely on a single disease but on various multiple disease clusters (Hu et al., 2022). Thus, it is necessary to examine how air pollution promotes the development of multimorbidity (multiple chronic diseases).

Although China is committed to achieve the climate commitment to the Paris Agreement, there is still a small rise in energy-related CO₂ emissions since 2010 (Qian et al., 2021). Plus, air pollution is another challenge for environmental management, despite gaining remarkable achievements in ecological civilization and sustainable development (X. Liu et al., 2013). Meanwhile, China is experiencing rapid aging, with 18.7% (264 million) population aged over 60 years in 2020 (National Bureau of Statistics, 2021). Compared with younger adults, older people have a higher possibility of living with multiple chronic diseases and are more susceptible to health risks (e.g., extreme climate conditions and severe air pollution; Carnes et al., 2014; Hu et al., 2022; Tian et al., 2019).

Evidence shows that changes in the local environment and climate have serious consequences for human health, especially one of the most severe threats to noncommunicable diseases (Campbell-Lendrum & Prüss-Ustün, 2019; Woodward et al., 2014). These changes could exacerbate existing health problems (Paavola, 2017), and populations who are currently most affected by climate-related illnesses may be at the greatest potential risk of multiple diseases (Woodward et al., 2014). The close reciprocal relationships between climate change, air pollution, and population health make demography a highly relevant discipline in providing evidence-based policy solutions to building sustainable and resilient societies (Muttarak, 2021; H. Yang et al., 2022). Thus, it is necessary to explore the multidimensional relationship between climate change, air pollution, and multimorbidity, especially for older adults.

Although current studies have established the relationship between climate change (rising temperatures), air pollution, and health outcomes (Carnes et al., 2014; Cohen et al., 2017), there remain several gaps in how both rising temperatures and air pollution develop the accumulation of multiple chronic diseases among older adults. First, the mean concentrations of air pollution or average temperatures during a given period measured by most studies could not reflect the dynamic nature of climate change and pollution exposure in the long term. Second, the rural–urban settings in China are quite diverse, which is not only related to the unequal distribution of climate or air pollution but also represents social inequality in health to some extent. Thus, a prerequisite of understanding the rural–urban disparities in health is to estimate health inequalities related to rising temperatures or air pollution in China. In this study, we use a large, prospective, nationally representative survey data set of middle-aged and older adults (aged 45+) in China, linked with PM_2.5 exposure derived from satellite data and temperature records from China Meteorological Bureau. Through longitudinal analyses, we examine the associations between PM_2.5 exposure, rising temperatures, and multimorbidity, as well as compared the rural–urban disparities in multimorbidity related to climate change and air pollution.

Method

Study Population

This study collects four waves of the China Health and Retirement Longitudinal Study (CHARLS 2011–2018), a large, prospective, nationally representative survey data set of middle-aged and older adults. Samples of the CHARLS were obtained from a four-stage stratified sampling approach (province–city–county–community, four administrative levels in China), with an overall response rate of 80.5% at the 2011 baseline survey (Zhao et al., 2014). The CHARLS baseline in 2011 interviewed 17,705 respondents from 28 provinces, 150 cities/counties/districts, 450 communities, and 10,257 households, collecting individual demographic, social, economic, and health circumstances (Zhao et al., 2014). After this baseline survey, there were three follow-up surveys in 2013, 2015, and 2018.

To focus on the middle-aged and older populations, we first excluded respondents aged less than 45 years old in each wave. Second, we dropped those who moved their residential locations from 2001 to baseline or during the period of 2011–2018 because we cannot identify the air pollution exposure history if respondents moved between waves. Third, we employed the listwise deletion process for inclusion and exclusion in Supplementary Figure 1. In total, 21,857 respondents (52,625 observations) from 125 cities were included in this study.

Outcome: Multimorbidity

The CHARLS records 14 self-reported doctor-diagnosed chronic diseases: hypertension; dyslipidemia; diabetes or high blood sugar; cancer or malignant tumor; chronic lung disease; liver disease; heart problems; stroke; asthma; kidney disease; stomach or other digestive diseases; emotional, nervous, or psychiatric problems; memory-related disease; and arthritis or rheumatism (Yao et al., 2020). Based on the diagnosis of a doctor, respondents were asked whether they do or do not have these chronic diseases (0 for no, 1 for yes). In line with previous studies using the CHARLS data (Hu et al., 2022; Yao et al., 2020; R. Zhang et al., 2019), we also used a disease count approach where we summed all 14 binary disease indicators (range 0–14) to capture multimorbidity, with a higher score for more diseases.

Climate Change: Temperature

Empirical research on the relationship between climate change and health has paid attention on climatic or weather extreme events, mainly focusing on temperature rising (Carnes et al., 2014; Obradovich et al., 2018). In this study, we collected daily temperature (at approximately 10 km) from the China Meteorological Administration (https://data.cma.cn/) as the main indicator for climate change. To match with the CHARLS, we first aggregated daily temperature to monthly city-level records because the CHARLS only discloses the interview dates (only including month, year) and city information in the Primary Sampling Units (PSU). Second, we calculated two statistics to measure the change of climate for each respondent. One is the average monthly temperature in a moving period (1, 2, 3, 4, and 5 years, the period prior to the interview date). The other is the coefficient of variation (CV; the ratio of the standard deviation to the mean) in temperature as the extreme heat events, reflecting the extent of variability in relation to the mean of temperature in various moving periods. The higher CVs represent the greater dispersion of temperature. In addition, we also used the quartile to divide temperature and extreme heat events into categorical variables (with four groups).

Air Pollution: PM_2.5

To collect PM_2.5 levels from 2000 to 2018, we used ground-level PM_2.5 estimates extracted from a long-term PM_2.5 data set with high spatiotemporal resolution (daily, 1 km) over China. The detailed model development and validation process can be found in our previous research (He et al., 2020, 2021) and were briefly summarized here. An advanced spatiotemporal model incorporating the geographically and temporally weighted regression model and the adaptive modeling method was employed to predict the daily 1-km PM_2.5 concentrations in the past 18 years (2000–2018). Ground-level PM_2.5 observations were employed as the dependent variable. Explanatory variables for modeling consist of multiple data sources, including the Multi-Angle Implementation of Atmospheric Correction high-resolution aerosol optical depth data, meteorological parameters, and land-related variables. We conducted a strict leave-one-year-out cross-validation technique to fully evaluate the accuracy and uncertainty of the daily PM_2.5 estimates outside the modeling period (He et al., 2020, 2021). The validation results indicate that our spatiotemporal modeling strategy achieved the state-of-the-art model performance, with monthly leave-one-year-out cross-validation R² and root-mean-square-deviation values of 0.74 and 15.75 µg/m³ (He et al., 2021).

Following previous studies on air pollution exposure (Ranft et al., 2009; Tallon et al., 2017; X. Zhang et al., 2018), we used 1, 2, 3, 4, and 5 years as the period of long-term exposure to PM_2.5. To capture long-term exposure, we measured the moving average PM_2.5 concentrations during the exposure period (based on respondents’ addresses [from the PSU in the CHARLS] to link with the historical satellite PM_2.5 concentrations). In addition, we also divided PM_2.5 exposure into four categorical variables for robustness testing. Considering the annual mean PM_2.5 limits in the National Ambient Air Quality Standard (35 µg/m³ for Level 1 and 75 µg/m³ for Level 2) and the median of PM_2.5 averages during the period between 2000 and 2018 (50 µg/m³), we categorized PM_2.5 exposure into four groups using the following cut-points: 1 (0–35 µg/m³), 2 (36–50 µg/m³), 3 (51–75 µg/m³), and 4 (76+ µg/m³).

Control Variables

To reduce the bias from confounders, we controlled a set of covariates, including demographic, socioeconomic, and regional factors. The demographic information contains gender (woman or man), age, and marital status (single or partnered). We also included age squared due to the potential nonlinear association of multiple chronic diseases with aging (Hu et al., 2022). The covariates of socioeconomic status (SES) include education, occupation, and household expenditure. Education includes three categories: no schooling, primary, and middle or higher. Occupational status also consists of three categories (agricultural, nonagricultural, and retired) from respondents’ occupational history. Higher wealth is associated with better health status, so we controlled household expenditure as the proxy for wealth (Luo et al., 2019). In the Chinese context, HuKou (rural and urban) is a special household registration system, not only related to housing address but also personal SES (Hou et al., 2019). Finally, annual regional gross domestic product (GDP) and population density at the city level are controlled to reflect the regional characteristics of cities to adjust for urbanization and industrialization (Samoli et al., 2019; Sun & Gu, 2008).

Analytical Strategies

We employed the growth curve modeling (GCM) approach to understand the longitudinal associations between climate change, PM_2.5 exposure, and multimorbidity, and explore the rural–urban disparities in these associations in four waves of the CHARLS 2011–2018. GCM is able to examine the trajectories of individuals over time and distinguish within-individual from between-individual heterogeneities in multimorbidity changes attributed to other variables (Luo et al., 2019; Rabe-Hesketh & Skrondal, 2012). GCM is a kind of multilevel modeling approach with random slope, and in this study, we used three-level linear models with random slope using longitudinal data with four waves across 8 years of data collection (Level 1) embedded in respondents (Level 2) embedded in cities (Level 3; Beroho et al., 2020).

We estimated three separate sets of models, each within different model designs. First, GCM is used to explore the association between temperature, PM_2.5 exposure, and multimorbidity. Second, we used extreme temperature events to replace average temperature in GCM models. Third, we separated rural and urban samples to re-run GCM for identifying the rural–urban disparities in multimorbidity. Fourth, we interacted categorical PM_2.5 exposure with temperature, establishing a joint variable with 16 categories as the interaction between air pollution and climate change. This takes into consideration that air pollution is highly related to temperature and weather context (Qian et al., 2021), which would have caused a problem of multicollinearity if assessed simultaneously in the same model (Hu et al., 2023). Finally, we added the interaction terms between age, age-squared, PM_2.5 exposure, or temperature to explore the multimorbidity trajectories.

Additionally, there were several strategies for robustness checks. First, as multimorbidity is a count variable, we assumed a Poisson distribution to estimate the association between temperature, PM_2.5, and multimorbidity. Second, considering the nonlinear relationship between temperature, PM_2.5, and multimorbidity, we used the categorical PM_2.5 exposure and temperature to replace continuous measures. Third, due to the missingness of predictors (details in Supplementary Appendix), we employed the multiple imputation (MI) approach based on chained equations to fill our data set and run the same models using the MI data to check the consistency of our main results. Fourth, we controlled more weather information (including humidity and wind speed), access to health care, life behaviors, and indoor air pollution to exclude the bias from related confounders. Fifth, due to consecutively recruiting samples aged over 45 in each wave, it is necessary to consider the strata for the wave to reduce the bias of birth cohort. Sixth, considering the nonlinear relationship between temperature and health risks (Abbafati et al., 2020; Kan et al., 2007; Tan et al., 2011), we checked the association between temperature, temperature square, and multimorbidity.

Results

We first observed the statistical characteristics of study samples from the CHARLS 2011–2018 in Table 1. Analytical samples at the baseline (CHARLS 2011) were aged at 59 years. Fifty percent were men, and 40% of the samples with primary education, 35% having middle or higher education. Twenty-two percent of respondents had urban HuKou; 35% samples were retired, 38% worked for agricultural jobs and 28% with nonagricultural jobs. Eighty-eight percent of respondents were partnered with their spouses or partners. Average multimorbidity score was 1.43 at the CHARLS 2011, increasing to 2.20 by 2018. From 2011 to 2018, average PM_2.5 concentrations in 5 years declined slightly from 49 to 45 µg/m³, and 5-year average temperature remained stable at 14°C. However, extreme events of temperature in 5 years had a sharp decline from 5.80 in 2011 to 2.86 in 2018, representing that the variation of temperatures was lowering (Table 1).

Table 1.

Characteristics of the Study Population in the CHARLS 2011–2018

Variable	CHARLS 2011	CHARLS 2013	CHARLS 2015	CHARLS 2018
Multimorbidity score, mean (SD)	1.43 (1.42)	1.57 (1.51)	2.04 (1.74)	2.20 (1.93)
5-Year PM2.5 exposure, mean (SD)	48.64 (14.53)	48.63 (14.80)	48.84 (14.93)	44.27 (15.29)
5-Year temperature, mean (SD)	14.46 (4.77)	14.03 (4.99)	14.19 (4.74)	14.54 (4.80)
5-Year extreme heat events, mean (SD)	5.80 (36.67)	1.23 (2.69)	1.48 (5.10)	2.86 (1.49)
Age, mean (SD)	59.26 (9.80)	59.75 (9.52)	61.11 (9.27)	61.94 (9.99)
Gender, n (%)
Men	6,401 (49.93)	5,914 (49.61)	5,381 (49.16)	8,206 (48.45)
Women	6,419 (50.07)	6,007 (50.39)	5,565 (50.84)	8,732 (51.55)
Education, n (%)
No schooling	3,307 (25.80)	2,894 (24.28)	2,544 (23.24)	3,727 (22.00)
Primary	5,075 (39.59)	4,843 (40.96)	4,482 (40.95)	7,289 (43.03)
Middle+	4,438 (34.62)	4,184 (35.10)	3,920 (35.81)	5,922 (37.96)
Expenditure (log), mean (SD)	8.16 (1.70)	8.27 (1.88)	8.17 (2.02)	8.64 (1.97)
HuKou, n (%)
Rural	9,977 (77.82)	9,334 (78.30)	8,544 (78.06)	13,121 (77.46)
Urban	2,843 (22.18)	2,587 (21.70)	2,402 (21.94)	3,817 (22.54)
Occupations, n (%)
Agricultural	4,790 (37.36)	4,830 (40.52)	4,126 (37.69)	6,066 (35.81)
Nonagricultural	3,511 (27.39)	3,276 (27.48)	3,187 (29.12)	5,054 (29.84)
Retired	4,519 (35.25)	3,815 (32.00)	3,633 (33.19)	5,818 (34.35)
Marital, n (%)
Partnered	11,341 (88.46)	10,674 (89.54)	9,735 (88.94)	14,610 (86.26)
Single	1,479 (11.54)	1,247 (10.46)	1,211 (11.06)	2,328 (13.74)
Population density (log), mean (SD)	5.91 (0.88)	5.85 (0.94)	5.91 (0.90)	5.80 (1.03)
Log GDP, mean (SD)	10.32 (0.55)	10.46 (0.64)	10.63 (0.52)	10.80 (0.52)
Number of respondents	12,820	11,921	10,946	16,938

Notes: CHARLS = China Health and Retirement Longitudinal Study; GDP = gross domestic product; PM_2.5= particulate matter 2.5; SD = standard deviation. All statistics are calculated after multiple imputation (see details in the “Method” section).

Table 2 shows the results from the growth curve models regarding the associations between PM_2.5 exposure, temperature, and multimorbidity. First, Model 1 shows the negative association between 1-year PM_2.5 exposure and multimorbidity (β = −0.076; 95% confidence interval (CI): −0.097, −0.055). With the longer exposure of PM_2.5, coefficients of the association between PM_2.5 exposure and multimorbidity are rising. For example, Model 2 presents that 10 µg/m³ increase in 2-year PM_2.5 exposure decreased 0.038 (95% CI: −0.061, −0.015) scores in multimorbidity, whereas the same increase in 3-year PM_2.5 exposure is associated with 0.083 increase in multimorbidity score (β = 0.083; 95% CI: 0.057, 0.108). In Model 4 using 4-year exposure period, 10 µg/m³ increase in PM_2.5 exposure rises 0.116 (95% CI: 0.088, 0.144) scores in multimorbidity, whereas we found that the figure for 5-year exposure period declines to 0.083 (95% CI: 0.046, 0.120) in Model 5. Second, the associations between temperature and multimorbidity score are relatively stable, revealing that rising temperature is associated with a higher multimorbidity score (worse chronic health). In terms of temperature, the coefficients in all models are positive. Among them, the coefficient of 3-year temperature is 0.432 (95% CI: 0.403, 0.460), larger than other models, representing that a 1°C increase in temperature rises 0.432 scores in multimorbidity. We found that with the increasing period, the pattern of temperature coefficients shows an inverted u-shaped curve.

Table 2.

Associations Between PM_2.5 (10 µg/m³), Temperature, and Multimorbidity

Variable	Model 1	Model 2	Model 3	Model 4	Model 5
	1 year	2 years	3 years	4 years	5 years
PM2.5	−0.076*** (−0.097 to −0.055)	−0.038** (−0.061 to −0.015)	0.083*** (0.057–0.108)	0.116*** (0.088–0.144)	0.083*** (0.046–0.120)
Temperature	0.154*** (0.137–0.171)	0.300*** (0.280–0.319)	0.432*** (0.403–0.460)	0.343*** (0.316–0.370)	0.185*** (0.159–0.212)
Age	0.067*** (0.051–0.082)	0.058*** (0.043–0.074)	0.057*** (0.042–0.072)	0.060*** (0.045–0.075)	0.065*** (0.050–0.080)
Age × Age	−0.0001* (−0.0003 to −1.75e−06)	−9.11e−05 (−0.0002–3.31e−05)	−7.55e−05 (−0.0002–4.88e−05)	−8.77e−05 (−0.0002–3.69e−05)	−0.0001 (−0.0002–2.32e−05)
Gender (ref: man)
Woman	0.179*** (0.138–0.220)	0.169*** (0.128–0.209)	0.170*** (0.129–0.211)	0.174*** (0.133–0.215)	0.182*** (0.141–0.223)
Education (ref: no schooling)
Primary	0.103*** (0.0612–0.144)	0.0849*** (0.0435–0.126)	0.0825*** (0.0411–0.124)	0.0901*** (0.0486–0.132)	0.106*** (0.0640–0.147)
Middle+	0.044# (−0.0078–0.095)	0.017 (−0.034–0.068)	0.020 (−0.031–0.072)	0.031 (−0.020–0.083)	0.053* (0.002–0.104)
HuKou (ref: rural)
Urban	0.070** (0.018–0.122)	0.082** (0.031–0.134)	0.081** (0.0289–0.132)	0.076** (0.025–0.128)	0.069** (0.018–0.121)
Occupations (ref: agricultural)
Nonagricultural	−0.0252* (−0.050 to −0.0002)	−0.0318* (−0.057 to −0.007)	−0.0319* (−0.057 to −0.007)	−0.0283* (−0.053 to −0.003)	−0.0230# (−0.048–0.002)
Retired	0.125*** (0.101–0.150)	0.121*** (0.0971–0.146)	0.118*** (0.0942–0.143)	0.121*** (0.0966–0.145)	0.125*** (0.100–0.150)
Expenditure	0.032*** (0.027–0.037)	0.031*** (0.026–0.035)	0.030*** (0.025–0.035)	0.031*** (0.026–0.035)	0.032*** (0.027–0.037)
Marital (ref: partnered)
Single	0.112*** (0.066–0.158)	0.103*** (0.057–0.149)	0.1000*** (0.054–0.146)	0.103*** (0.057–0.149)	0.108*** (0.061–0.154)
Log population density	−0.0339 (−0.098–0.030)	0.00302 (−0.062–0.068)	0.0160 (−0.049–0.081)	−0.00146 (−0.066–0.064)	0.0272 (−0.037–0.092)
Log GDP	0.737*** (0.693–0.780)	0.561*** (0.513–0.609)	0.708*** (0.662–0.754)	0.830*** (0.784–0.876)	0.928*** (0.883–0.973)
Constant	−11.71*** (−12.49 to −10.92)	−11.91*** (−12.73 to −11.10)	−15.93*** (−16.81 to −15.06)	−16.15*** (−17.01 to −15.30)	−15.24*** (-16.08 to −14.40)
Random effects
Within individual
Change rate (age)	0.005***	0.005***	0.005***	0.005***	0.005***
Intercept	14.385***	14.083***	14.341***	14.180***	13.825***
Covariance	−0.257***	−0.252***	−0.256***	−0.253***	−0.247***
Between cities
Residuals	1.302***	3.192***	6.125***	4.275***	1.933***
Between individuals
Residuals	0.488***	0.481***	0.479***	0.484***	0.493***
Observations	52,625	52,625	52,625	52,625	52,625
Number of cities	125	125	125	125	125

Notes: GDP = gross domestic product; PM_2.5= particulate matter 2.5.

***p < .001. **p < .01. *p < .05. ^#p < .1.

Third, the associations between multimorbidity and other covariates are expected. In Model 4, for example, women have higher risks in multimorbidity than men. With the increase of age, the score of multimorbidity is also higher but the expected curvilinear association was not evidenced in Table 2. Higher SES (higher educational attainment, urban HuKou, and higher household expenditure) and retired status are associated with higher multimorbidity scores, whereas the similar association is not found in samples with nonagricultural occupations. Compared with respondents with spouses or partners, single respondents have higher risks in multimorbidity status (β = 0.113; 95% CI: 0.067, 0.160). Model 4 also shows that higher population density or higher GDP is associated with higher multimorbidity scores.

In terms of random effects in Model 4 of Table 2, the variance of the random intercept is 13.8 (p < .001), representing the significant variation in multimorbidity scores between individuals. The variance of the random slope of age (0.005, p < .001), reflecting a tiny, predictable, and downward trajectory of multimorbidity with age, as previously shown in health studies (Hale, 2017; Hu et al., 2023).

Considering the difference between average temperature and extreme heat events, we estimate the relationship between PM_2.5, extreme heat events (temperature variations), and multimorbidity in Table 3. Similar to Table 2, the associations between PM_2.5 exposure and multimorbidity are positive when the exposure period is more than 3 years. Unlike Table 2, however, 1-year extreme heat events are not significantly associated with multimorbidity score. Model 2 shows the largest coefficient of extreme heat events (β = 0.187; 95% CI: 0.163, 0.212) among these five models in Table 3, and the figure for extreme heat events gradually declines with the increase of exposure period. In Model 5, one-unit increase in extreme heat events is only associated with 0.015 (95% CI: 0.007, 0.024) increased scores in multimorbidity.

Table 3.

Associations Between PM_2.5 (10 µg/m³), Extreme Heat Events, and Multimorbidity

Variable	Model 1	Model 2	Model 3	Model 4	Model 5
	1 year	2 years	3 years	4 years	5 years
PM2.5	−0.103*** (−0.124 to −0.083)	−0.013 (−0.035–0.009)	0.045*** (0.020–0.070)	0.064*** (0.037–0.091)	0.051** (0.016–0.086)
Heat events	−0.004 (−0.034–0.025)	0.187*** (0.163–0.212)	0.078*** (0.069–0.088)	0.059*** (0.050–0.068)	0.015*** (0.007–0.024)
Age	0.070*** (0.055–0.085)	0.065*** (0.050–0.080)	0.062*** (0.047–0.077)	0.063*** (0.048–0.078)	0.066*** (0.050–0.081)
Age × Age	−0.0001* (−0.0003 to −5.44e−06)	−9.66e−05 (−0.0002 to 2.85e−05)	−7.08e−05 (−0.0002 to 5.42e−05)	−7.69e−05 (−0.0002 to 4.82e−05)	−8.88e−05 (−0.0002 to 3.66e−05)
Gender (ref: man)
Woman	0.186*** (0.145–0.227)	0.183*** (0.143–0.224)	0.183*** (0.142–0.224)	0.184*** (0.143–0.225)	0.187*** (0.146–0.228)
Education (ref: no schooling)
Primary	0.114*** (0.0723–0.156)	0.111*** (0.069–0.152)	0.111*** (0.069–0.152)	0.113*** (0.071–0.155)	0.118*** (0.0764–0.160)
Middle+	0.061* (0.009–0.112)	0.059* (0.008–0.111)	0.0579* (0.007–0.109)	0.060* (0.008–0.111)	0.065* (0.013–0.116)
HuKou (ref: rural)
Urban	0.056* (0.004–0.107)	0.050# (−0.002–0.101)	0.051# (−0.0004–0.103)	0.055* (0.003–0.107)	0.059* (0.007–0.110)
Occupations (ref: agricultural)
Nonagricultural	−0.022# (−0.047–0.003)	−0.024# (−0.049–0.001)	−0.0244# (−0.050–0.001)	−0.0232# (−0.048–0.002)	−0.0210 (−0.046–0.004)
Retired	0.126*** (0.101–0.151)	0.127*** (0.103–0.152)	0.127*** (0.103–0.152)	0.127*** (0.102–0.151)	0.127*** (0.102–0.152)
Expenditure	0.033*** (0.028–0.038)	0.033*** (0.029–0.038)	0.034*** (0.029–0.039)	0.034*** (0.029–0.039)	0.034*** (0.029–0.039)
Marital (ref: paternal)
Single	0.113*** (0.067–0.160)	0.114*** (0.068–0.161)	0.114*** (0.068–0.161)	0.114*** (0.068–0.160)	0.113*** (0.067–0.160)
Log population density	0.006 (−0.053–0.065)	0.073* (0.014–0.133)	0.082** (0.022–0.142)	0.068* (0.008–0.129)	0.051# (−0.010–0.111)
Log GDP	0.830*** (0.788–0.871)	0.936*** (0.895–0.978)	0.983*** (0.941–1.025)	1.007*** (0.963–1.050)	0.970*** (0.926–1.013)
Constant	−10.80*** (−11.54 to −10.05)	−12.57*** (−13.31 to −11.83)	−13.27*** (−14.01 to −12.53)	−13.58*** (−14.33 to −12.83)	−13.13*** (−13.90 to −12.37)
Random effects
Within individual
Change rate (age)	0.005***	0.005***	0.005***	0.005***	0.005***
Intercept	14.135***	13.721***	13.690***	13.710***	13.964***
Covariance	−0.253***	−0.245***	−0.245***	−0.245***	−0.250***
Between cities
Residuals	0.500***	0.605***	0.670***	0.688***	0.644***
Between individuals
Residuals	0.495***	0.495***	0.494***	0.495***	0.497***
Observations	52,625	52,625	52,625	52,625	52,625
Number of cities	125	125	125	125	125

Notes: GDP = gross domestic product; PM_2.5= particulate matter 2.5.

***p < .001. **p < .01. *p < .05. ^#p < .1.

To explore the rural–urban disparities in multimorbidity attributed to PM_2.5, we designed two sets of models in rural and urban samples. Figure 1A shows the associations between PM_2.5 exposure and multimorbidity in rural samples, similar to the results of Table 2. In Figure 1B, however, we found that there are stronger associations between PM_2.5 exposure and multimorbidity in urban samples than those in rural samples. For example, 10 µg/m³ increase in 4-year PM_2.5 decreased 0.085 (95% CI: 0.054, 0.116) scores in multimorbidity for rural samples, whereas the figure for urban samples is 0.097 (95% CI: 0.042, 0.153). Also, in 1- and 2-year PM_2.5 exposure, there are negative associations between PM_2.5 exposure and multimorbidity for rural samples, but for urban samples, it is not as strong as rural samples.

Figure 1.

Coefficients of PM_2.5 exposure with 95% confidence intervals in different periods. (A) Models only with rural samples, adjusted covariates including temperature, age, age squared, gender, marital status, education, HuKou, occupations, household expenditure (log-transformed), annual GDP, and population density at the city level (log-transformed). (B) Models only with urban samples, adjusted covariates including temperature, age, age squared, gender, marital status, education, HuKou, occupations, household expenditure (log-transformed), annual GDP, and population density at the city level (log-transformed). GDP = gross domestic product; PM_2.5= particulate matter 2.5.

Following the strategy of Figure 1, we explored the rural–urban difference in multimorbidity related to rising temperature in Figure 2. Unlike Figure 1, however, Figure 2 shows an opposite pattern of rural–urban disparities, revealing a more harmful impact of temperature on multiple chronic diseases in rural samples than that of rural samples. Specifically, a 1°C increase in the 3-year average temperature increased 0.394 (95% CI: 0.362, 0.426) scores in multimorbidity for rural samples, whereas the figure for urban samples is 0.106 (95% CI: 0.075, 0.137). Supplementary Figure 2 shows a similar trend of extreme heat events on multimorbidity scores, suggesting that rural residents have a higher prevalence of multimorbidity related to temperature.

Figure 2.

Coefficients of temperature with 95% confidence intervals in different periods. (A) Models only with rural samples, adjusted covariates including PM_2.5 exposure, age, age squared, gender, marital status, education, HuKou, occupations, household expenditure (log-transformed), annual GDP, and population density at the city level (log-transformed). (B) Models only with urban samples, adjusted covariates including PM_2.5 exposure, age, age squared, gender, marital status, education, HuKou, occupations, household expenditure (log-transformed), annual GDP, and population density at the city level (log-transformed). GDP = gross domestic product; PM_2.5= particulate matter 2.5.

Previous studies evidence a high correlation between PM_2.5 exposure and temperature (Kinney, 2018; Orru et al., 2017), so we further examined the impacts of the interaction between PM_2.5 exposure and temperature on multiple chronic diseases. Overall, we found a strong correlation between PM_2.5 exposure and temperature in Supplementary Table 9 (more people exposed to higher PM_2.5 are living with higher temperature, such as higher proportion in “3-2,” “3-3,” and “3-4” groups). In Figure 3, compared with the “1-1” group, both “1-2” and “2-1” groups show a significantly positive association with multimorbidity, suggesting a greater effect size in temperature and PM_2.5, whereas the coefficient of the “2-1” group is not significant at 95% level. Specifically, respondents in the “1-2” group have 0.28 (95 CI: 0.203, 0.361) higher scores in multimorbidity than those in the “1-1” group. Similar findings are also found in other groups (e.g., “2-1,” “2-2,” and “3-1” groups). These findings suggest that the harmful impacts of temperature are more substantive than that of PM_2.5 exposure.

Figure 3.

Coefficients of the interaction between PM_2.5 exposure and temperature with 95% confidence intervals. Results were from multilevel models with interactions 5-year PM_2.5 exposure and temperature, adjusted covariates including age, age squared, gender, marital status, education, HuKou, occupations, household expenditure (log-transformed), annual GDP, and population density at the city level (log-transformed). The first number in “PM_2.5 # temperature” is PM_2.5 (1: 0–35 µg/m³; 2: 36–50 µg/m³; 3: 51–75 µg/m³; 4: 76+ µg/m³), and the second represents temperature (1: first quartile of temperature in 5 years; 2: second quartile; 3: third quartile; 4: fourth quartile). GDP = gross domestic product; PM_2.5= particulate matter 2.5.

To examine the role of age in the associations between PM_2.5 exposure, rising temperature, and multimorbidity, we interacted PM_2.5 exposure/temperature with age and age squared, and plotted the details in Figure 4. We can see higher PM_2.5 exposure or temperature was associated with worse chronic health in an expected trend, but the overlapping CIs in Figure 4 suggest that age is not always a significant factor to boost the effects of air pollution or temperature on multimorbidity at the life-course span. We can see, in Figure 4, it is only at certain older ages (80 years in Figure 4A or 75 in Figure 4B, for example) that there is a statistically significant difference in the associations between PM_2.5 exposure, rising temperature, and multimorbidity across age.

Figure 4.

Trajectories of multimorbidity by PM_2.5 exposure and temperature with 95% confidence intervals in different periods. (A) Results were from GCM with interactions between PM_2.5 exposure, age and age squared, adjusted covariates including temperature, gender, marital status, education, HuKou, occupations, household expenditure (log-transformed), annual GDP, and population density at the city level (log-transformed). (B) Results were from GCM with interactions between temperature, age and age squared, adjusted covariates including PM_2.5 exposure, gender, marital status, education, HuKou, occupations, household expenditure (log-transformed), annual GDP, and population density at the city level (log-transformed). GCM = growth curve modeling; GDP = gross domestic product; PM_2.5= particulate matter 2.5.

Furthermore, we also ran a set of robustness checks to validate our findings above. First, we ran the Poisson regression models with covariates the same as in Table 2. Supplementary Table 1 (average temperature) and Supplementary Table 2 (extreme heat events) show no significant differences with the results in Tables 2 and 3, revealing that the distribution of multimorbidity might not bias our estimates. Second, we examined the nonlinear associations between temperature, PM_2.5, and multimorbidity using the categorical PM_2.5 exposure and temperature to replace continuous measures in Supplementary Tables 3 and 4. We found that results in models with categorical measures were consistent with the main findings, still presenting the associations of higher PM_2.5 exposure and temperature with higher multimorbidity scores. Third, we ran the same model specification (Table 2) using the MI data in Supplementary Tables 4 and 5. Comparing all results from the original data, we found that the MI results were consistent with the main findings, suggesting that the missingness might not influence the core findings. Fourth, considering the bias from unobserved information, we controlled the extra variables (access to health care, life behaviors, indoor air pollution, humidity, and wind speed) based on Tables 2 and 3. In Supplementary Tables 6, 7, and S14, we found that there are no significant differences with Tables 2 and 3 except for some slight changes in effect size. Fifth, Supplementary Table 10, controlling the strata for the wave, shows similar findings to Table 2, which suggests that the main results were not biased due to the effect of birth cohort. Sixth, we checked the nonlinear relationship between temperature and health risks in Supplementary Table 13, which shows an inverted u-shaped relationship between temperature and multimorbidity (also plotted the pattern in Supplementary Figure 4). These results, together with results in models with categorical measures, reflect the nonlinear relationship between temperature and multiple diseases as previously found (Abbafati et al., 2020; Kan et al., 2007; Tan et al., 2011).

Discussion

This study investigates the relationship between climate change, air pollution, and multimorbidity through linking the CHARLS (2011–2018), a nationally representative data set, with temperature records from the China Meteorological Administration and historical PM_2.5 records derived from remotely sensed satellite data. With a sequence of longitudinal analyses, the main contribution of this study consists of a detailed analysis of the interaction between temperature and PM_2.5 exposure to untangle the associations between climate change, air pollution, and multimorbidity.

First of all, this study shows that rising temperature is associated with higher multimorbidity among Chinese middle-aged and older adults, as reported in previous studies, in which global warming is a risk factor for chronic health (Bartholy & Pongrácz, 2018; Haines, 1991; Maibach et al., 2019; Staropoli, 2002). First, we used the intensity of temperature (average temperature) as the measure for climate change over periods ranging from 1 to 5 years. The findings show that coefficients of temperature follow a u-shaped pattern when the exposure period is increased from 1 to 5 years. One potential reason may be the adaptability of respondents living with rising temperature in such a long term. Second, we found that more extreme heat events (the ratio of the standard deviation to the mean of temperature) were associated with more multiple chronic diseases, consistent with previous research regarding extreme weather (Weilnhammer et al., 2021). However, our findings show a downward pattern of coefficients of extreme heat events when the exposure period gradually increased. Similar to the findings of temperature earlier, respondents may adapt to the change of climate and understand how to prevent the risks from extreme weather events in a long-term period.

In terms of air pollution, our study confirms the relationship between long-term PM_2.5 exposure and multimorbidity in previous studies (Arku et al., 2020; Atkinson et al., 2013; Hu et al., 2022). However, our findings show that 3-year exposure or more to PM_2.5 is significantly associated with multimorbidity while there are no significant associations between 1- and 2-year PM_2.5 exposure. Although previous studies evidence the nonsignificant associations between air pollution and health of older adults (Adar et al., 2018; Hu et al., 2020), these associations focus on short-term exposure to air pollution (e.g., less than 1 year). Our study further suggests that only longer-term exposure to air pollution is associated with multiple chronic diseases, and the harmful impacts of air pollution might not be detected in a short-term period (e.g., less than 2 years). In particular, the association between air pollution and memory-related diseases or nervous problems should be examined in a long-term period (3 years or longer; Babadjouni et al., 2017; Younan et al., 2019). In addition, air pollution usually coincides with economic development in the Chinese context (W. Liang & Yang, 2019; Zhu et al., 2019). The short-term increase in air pollution is mainly due to rapid urbanization and industrialization, which lead to improved infrastructure and are also the main drivers of beneficial health status (Hou et al., 2019). With prolonging the period of PM_2.5 exposure, health advantages of economic progress along with air pollution are offset by the harmful effects of air pollution on health outcomes. This could be clarified in further research using detailed data sets to exclude more confounding factors.

In the Chinese context, rural–urban settings do not only represent spatial and regional distribution but reflect various inequalities in social welfare and living conditions. Previous studies confirmed that rural residents have higher risks in health diseases than urban ones (C. C. Liu et al., 2022; Wang et al., 2015). With the rapid economic development, rural–urban disparity in PM_2.5 exposure significantly increased as well (Xiao et al., 2020). In our study, the relationship between climate change, PM_2.5 exposure, and multimorbidity also shows rural–urban disparities. Yet rural respondents have more chronic diseases due to rising temperature or extreme heat events but lower prevalence of multimorbidity attributed to PM_2.5 exposure. The unexpected findings may be due to the unequal distribution of air pollution and extreme weather events. First, severe air pollution often happened in major cities (Chai et al., 2014), which might lead to higher prevalence of chronic diseases in urban areas. Second, global warming and rising temperatures are a common health risk for all nations and populations instead of regional issues. Theoretically, the harmful impacts of rising temperature should not be different in rural and urban areas. However, urban residents have more advantages in living conditions (e.g., air conditioner, fridge) and medical services than rural ones. These disparities cannot be removed even though we controlled household expenditure as a proxy for wealth. Thus, the rural–urban disparities in multimorbidity related to temperature might be due to the inequalities in individual and community SES or living conditions.

Many studies explore the interplay of climate change and air pollution on health (de Sario et al., 2013; Kinney, 2018; Orru et al., 2017). These studies suggest synergistic health effects of exposure to higher temperature and higher air pollution. In this study, we found that respondents exposed to higher PM_2.5 are more likely to live in surroundings with higher temperature. The reason is that climate change can influence the dispersion of air pollutants (e.g., PM) and accelerate the formation of secondary pollutants, such as volatile organic compounds (Orru et al., 2017). This is one of the important reasons why our findings show that temperature-related multimorbidity is higher than that related to PM_2.5 exposure. Another potential reason is that in China, air pollution has been a main environmental issue to overcome in public governance (Hu & He, 2023), whereas rising temperature has not been a public concern for all populations (X. Liu et al., 2019). The combination of the shortage of governing policies and insufficient prevention strategies among individuals leads to greater negative health effects of temperature in China.

In terms of the role of age, our findings show age was not a significant driver to worsen chronic diseases attributed to PM_2.5 exposure or rising temperature, which is similar to previous research (Hu et al., 2023). Although air pollution and climate change have nonnegligible effects on human health, they cannot change the trajectory of multimorbidity with age.

There are several limitations in this study. First, our study can only identify the association between temperature, PM_2.5, and multimorbidity at the city level, because we cannot obtain detailed respondents’ addresses to match PM_2.5 exposure and temperature data at the individual level. Second, the relationship between air pollution, climate change, and multiple chronic diseases should be a life-course issue. Although we have established a historical data set including PM_2.5 and temperature records in 10 more years, this cannot be used in a life-course analysis, especially for those aged 45 and over. Third, there are some unobserved confounders that we cannot include in this paper but may be associated with health outcomes as well as air pollution and temperature, such as individual or community SES factors. Fourth, as previously demonstrated (Yao et al., 2020; R. Zhang et al., 2019), we used self-reported doctor-diagnosed chronic diseases as the components for multimorbidity. However, this measure may underrate the prevalence of chronic diseases among respondents with low SES because they have less access to obtain medical diagnosis than those with high SES. Fifth, we used MI to solve the missingness bias; however, we also realized that the imputation strategy cannot remove all uncertain factors from missingness. Supplementary Tables 11–12 and Supplementary Figure 3 all show that the missing samples are not equally distributed in each wave and each province. Thus, the results from MI results can only be an auxiliary analysis, and in the future, we should collect more detailed data with less missingness to detect a more unbiased link.

Our study has two main contributions to methodological design. First, two measures (average and coefficient of variation) for temperature provide a more informative perspective to explore the relationship between climate change and multimorbidity. Second, we established a joint variable of rising temperature with PM_2.5 exposure to examine and compare their associations with multimorbidity, which can avoid the problem of multicollinearity if assessed these two variables simultaneously using continuous measurement in one model.

Conclusion

This study examines the relationship between temperature, PM_2.5 exposure, and multimorbidity among Chinese adults aged 45 and over, when controlling for demographic (gender, age, age squared, and marital status), SES (educational attainment, occupational stratus, HuKou, and household expenditure), and regional factors (GDP per capita and population density). Our findings confirm the rural–urban disparities in multimorbidity related to rising temperature and PM_2.5 exposure, which suggests the role of individual and community SES in environmental health inequality. In addition, our study further indicates that temperature-related multimorbidity is higher than PM_2.5-related multimorbidity, and suggests that efforts to both mitigate air pollution emissions and to adapt to the impacts of rising global temperatures are urgently required. Further research with more detailed data is needed to untangle the causal mechanisms underlying the relationship between climate change, long-term exposure to PM_2.5, and multiple chronic diseases.

Funding

This study was supported by the Fundamental Research Funds for the Central Universities (JKE02232201) and the Funds for the Construction of the First-class Disciplines, East China University of Science and Technology (SLE00234001).

Conflict of Interest

None reported.