We used data from the Chinese Longitudinal Health and Longevity Survey (CLHLS), an on-going national representative prospective cohort study of Chinese individuals aged ≥60 years, covering approximately 85% of the population in China (13, 14). Since 1988, CLHLS has been gathering demographic, lifestyle, and health information every 2–3 years through face-to-face interviews conducted by trained health workers from local Centers for Disease Control and Prevention and community hospitals (14). Details about CLHLS have been described elsewhere (13). CLHLS is considered to be of high quality because of its few missing data, high response rate, low dropout rate (average per wave was 4.85%), and compatibility with other countries' data (15). The CLHLS was approved by the Research Ethics Committee of Peking University (IRB00001052-13074), and the written informed consent of all participants were obtained.

BMI variables were collected from 2008, so a total of 5,245 participants (≥60 years) surveyed in 2008, 2011, and 2014 were eligible for this study. We excluded 1,158 participants with five major chronic diseases at baseline (8, 16, 17), 353 participants with missing weight data in any one of these three surveys, 809 participants without follow-up data after 2014, and 227 participants with missing healthy aging data (2018), leaving 2,698 eligible participants for this study (eFigure 1). To assess the possibility of potential selection bias, we compared the included participants (n = 2,698) with those excluded due to the lack of follow-up or healthy aging data (n = 1,036) and found that the basic characteristics were largely comparable between these included and excluded participants, except that excluded participants were more likely to be from urban areas (eTable 1). This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Weight and height of all participants were measured three times in 2008, 2011, and 2014. It was measured by trained medical staff using standardized programs. BMI is calculated by dividing the weight (kg) by the square of the height (m). According to the recommended standards of adults in China (18), body type was divided into underweight (<18.5 kg/m²), normal (18.5–23.9 kg/m²), overweight (24.0–27.9 kg/m²), and obesity (≥28.0 kg/m²). Based on the BMI in 2008, 2011, and 2014, we identified BMI trajectories using Latent Class Growth Modeling (LCGM) (19). In the secondary analysis, we used the standard deviation of BMI from three wave surveys to evaluate the variability of BMI.

Based on previous research and the framework of Rowe and Kahn's concepts (5, 16), healthy aging was evaluated in 2018 and was defined as having no five major chronic diseases (cancer, chronic lung disease, diabetes, heart disease and stroke), no cognitive impairment (assessed by the Chinese version of Mini-Mental State Assessment (MMSE) and education-stratified cutoffs), no physical limitations [evaluated by the Katz Index of Independence in Activities of daily living (ADL)] and good mental health [no depression (Center for Epidemiologic Studies Depression Scale, CESD-10) or anxiety (Generalized Anxiety Disorder Scale, GAD-7)]. If any one of the criteria was not met, it was defined as usual aging. Detailed descriptions of each of the 4 domains of healthy aging are listed in Table 1 (7, 20–29).

The covariate information was collected in 2008 based on previous studies (15). Demographic characteristics included age, gender (male, female), ethnicity (Han Chinese, others) (30), residence district (city, town, rural), education (illiteracy, primary school and below, junior high school and above), marital status (15) (married and living with spouse, others), living expenses (adequate, inadequate), and occupation (farmer, others). Lifestyle information included sleep quality (good, fair, poor), sleep duration (31) (<7, 7–8, >8 h/day), smoking status (never, former, current), drinking status (never, former, current), exercise status (never, former, current) and dietary diversity (Dietary Diversity Score, DDS), and BMI at baseline (underweight, normal, overweight, and obesity). If the participant was unable to answer, then the questions (except those that had to be filled out by the participant) could be answered by an adult family member who was most familiar with the participant's living situation.

When the covariates in 2008 were missing (variables are missing in <5% of the participants), we used the data collected in 2011 or 2014 to impute them. Finally, the proportion of missing values in covariates was 0.07% (missing values for occupation and education were 2 and 2, respectively).

We used Latent Class Growth Modeling (LCGM) (19) (R [lcmm package]) to identify subgroups with similar latent BMI trajectories in three waves. Generally, we fitted the longitudinal BMI data into the model with a mixture of multiple latent trajectories by the maximum likelihood method (17). We implemented linear, quadratic and cubic models and used the Bayesian Information Criterion (BIC) and entropy to identify the optimal number and shapes of trajectory groups (eTable 2), and the proportion of participants in each trajectory should meet the condition of at least 5% of the population (32). Then, the posterior probabilities computed via Bayes' theorem were used to test the identification ability, with detailed methods found in the LCGM tutorial (19, 33).

The Chi-square test was used to compare differences in categorical variables and Kruskal–Wallis test in ordinal data and non-normal continuous data. We used logistic regression models to estimate the odds ratios (ORs) and 95% confidence intervals (CIs). Model I was adjusted for baseline age, gender. Model II was additionally adjusted for residence district, ethnic group, educational level, marital status, living expenses and occupation. Model III was additionally adjusted for lifestyle factors, including smoking history, drinking status, exercise, dietary diversity, sleep quality, and sleep duration. Model IV was additionally adjusted for body mass index at baseline. To explore the stability and heterogeneity of the associations between different BMI trajectories with healthy aging in different lifestyle situations, we further performed stratified analyses by sex, age and lifestyle (smoking, drinking, exercise, diet, sleep quality and sleep duration). We used the likelihood ratio test to evaluate the statistical significance of the multiplicative interaction between lifestyles and BMI trajectories.

To compare with previous studies, we conducted a secondary analysis utilizing 2014 BMI data to investigate its correlation with healthy aging and its various dimensions through the application of restricted cubic spline curves and threshold effect analysis. For verifying the robustness of the results, we conducted a series of sensitivity analyses, the specific methods are presented in Supplementary data (Appendix 1).

All statistical analyses were performed using Empower (R) (https://www.empowerstats.net/cn/, X & Y solutions, Inc. Boston MA) and R (https://www.r-project.org/). All P-values were 2-sided and P < 0.05 was considered statistically significant.

Three different BMI trajectories in older adults from 2008 to 2014 were identified (eTable 2): the average BMI of 1,330 (49.30%) participants maintained around 19 kg/m² (low-normal stable trajectory), 1,162 (43.07%) participants had a moderate increase in BMI from 21.77 kg/m² to 23.17 kg/m² (normal slight increase trajectory), and 206 (7.64%) participants had an increase in BMI from overweight to obesity (overweight to obesity trajectory) (Figure 1). We calculated the posterior probability of each individual being assigned to each trajectory group and assigned it to the trajectory group with the highest posterior probability. The average posterior probabilities for each trajectory groups were >0.70 (0.853, 0.888, and 0.903, respectively), suggesting that each trajectory we fitted had high internal reliability and was sufficiently distinguishable among participants with different BMIs.

Table 2 showed the age-adjusted baseline characteristics of participants with different BMI trajectories. Compared with the normal slight increase trajectory, participants with the overweight to obesity trajectory were slightly younger, tended to be female, non-farmers, had exercise, and had better dietary diversity. Conversely, participants with the low-normal stable trajectory were slightly older, more likely to be female, illiterate, farmer, living alone, non-smokers, non-drinkers, non-exercisers, had worse sleep quality, and had worse dietary diversity. All P < 0.05.

Table 3 summarized the ORs of healthy aging and its four dimensions associated with BMI trajectories. Compared with the normal slight increase trajectory, the overweight to obesity trajectory was significantly associated with lower odds of healthy aging after adjustment for potential confounders. The multivariable-adjusted ORs for Model 4 were 0.90 (95%CI: 0.70, 1.15) for the low-normal stable trajectory and 0.57 (95%CI: 0.36, 0.93) for the overweight to obesity trajectory. The association between BMI trajectories and the four dimensions of healthy aging differed. Compared with the normal slight increase trajectory, older adults in the overweight to obesity trajectory had a multivariable-adjusted OR of 0.46 (95%CI: 0.31, 0.69) for being free of chronic disease. In contrast, those in the low-normal stable trajectory had a multivariable-adjusted OR of 0.70 (95%CI: 0.56, 0.87) for no cognitive impairment.

In the stratified analyses, the associations between BMI trajectories and healthy aging were consistent among subgroups of different gender, age, BMI, lifestyles including smoking, drinking, exercise, diet, sleep quality and sleep duration, all P for interaction >0.05 (Figure 2). Specifically, in almost every subgroup, the overweight to obesity trajectory was consistently associated with lower odds of healthy aging.

Based on the results of Appendix 2, Table 4, and eTable 3, we discovered our findings remained robust in several sensitivity analyses.

Appendix 2 and Figure 3 indicated that compared to individuals with a normal slight increase trajectory and the least BMI variability, the odds of healthy aging were lowest among those with an overweight to obesity trajectory and a BMI standard deviation in the fourth quartile (OR = 0.39, 95%CI: 0.18, 0.86). Higher BMI variability is an important marker of healthy aging independent of baseline BMI.

In comparison to older adults with a normal weight range (BMI = 18.5–23.9 kg/m²), being underweight (BMI <18.5 kg/m²) poses a risk to mental health and cognitive function, while obesity (BMI ≥ 28 kg/m²) is associated with risk for chronic disease but is considered a protective factor for cognitive function. Restrictive cubic spline plots demonstrated an inverted U-shaped correlation between BMI and all three dimensions of healthy aging except for cognitive function (eTable 5; eFigure 2).

We identified three BMI trajectories from 2008 to 2014 in this cohort study and found that the BMI trajectory of most Chinese older people was stable at a normal or low-normal level. Compared with participants with a normal slight increase trajectory, those with overweight to obesity trajectory had significantly lower odds of healthy aging, and this association was more pronounced on the chronic disease dimension. We also observed a linear negative association between the variation (standard deviation) in BMI and healthy aging. Moreover, compared to participants with normal slight increase trajectory and the least BMI variability, the subsequent healthy aging rate is significantly reduced for those with the overweight to obesity trajectory and with a large variation in BMI.

This study used three repeated measurements to model BMI trajectories during older adulthood among Chinese older adults, an approach validated in previous studies (19, 34, 35). Existing studies in older adults (36–38), further confirms that this measurement frequency are sufficient to model BMI trajectories in this population. The trajectory analysis revealed that the BMI trajectories of most Chinese older adults tend to stay within the normal weight range (BMI: 18.5–23.9 kg/m²), which was consistent with the survey results in Taiwan (11). However, the distribution of BMI trajectories in older adults differed by region. An early study of the BMI trajectory in the United States found that 70%-80% of middle-aged and older people were overweight and obesity (39). The first survey conducted in Asia on the BMI trajectory (38) found that BMI for older Japanese was mainly within the normal weight range (BMI: 18.5–25.0 kg/m²), which is also similar to our results. Such differences may be attributed to racial differences, lifestyle and dietary habits. A study involving the analysis of BMI trajectories in older adults showed significant racial differences among Black, White, and Hispanic individuals (39). Therefore, it is recommended to focus on BMI trajectories between different populations and further study the BMI trajectories and their changing mechanisms in Asian and non-Asian older adults.

To the best of our knowledge, this is the first prospective study to investigate the association between BMI trajectories and healthy aging in older adults. Many current studies had focused on BMI and individual health outcomes among adults, and found that having a high BMI was related to chronic diseases (8), psychological problems (40), cognitive impairment (41) and disability (9, 26). A Canadian study (10) showed that individuals in the Overweight-Stable, Obesity I-Stable, and Obesity II -Stable groups experienced more chronic diseases, cognitive impairment, and declines in self-rated overall health than those in the Normal-Stable group. Therefore, our study combines the four aspects of chronic diseases, cognitive health, mental and physical function to evaluate healthy aging, which provides more concrete evidence for the association between BMI trajectories and the overall health of older adults. Previous studies commonly used frailty as the primary outcome, for example, studies conducted in the USA (42), Finland (43), Taiwan (China) (11) have shown that low and very high BMI are associated with a higher risk of frailty development in middle-aged and older people. Although frailty and healthy aging were defined differently, they are both an overall indicator that can measure the health and quality of life of older people. Thus, the findings of our study are similar to these findings, in which we observed that individuals in the overweight to obesity trajectory were significantly associated with reduced rates of healthy aging compared with those in the normal slight increase trajectory.

According to a literature review, some cross-sectional studies have shown a significant association between lower body weight or BMI and successful aging (44, 45). In our study, this result is not significant but a similar trend exists. Similar to our findings, several studies showed that overweight was significantly associated with a shorter healthy life expectancy (20). For example, a higher BMI in midlife was associated with significantly lower odds of healthy aging in French (8), and relevant research in China (46) showed that preventing central obesity and avoiding BMI increases were beneficial to the successful aging of older adults, which is consistent with our research results. Therefore, our study supplements evidence on BMI trajectories and healthy aging in older adults.

The impact mechanisms of BMI on healthy aging are complex. Firstly, the increase in body fat (which comes with a high BMI) is what brings about metabolic consequences due to its inflammatory effects (such as cytokine release, adipocyte death, and macrophage infiltration) in adipose tissue (47), along with high leptin (due to the increase in fat mass itself) (48). This inflammatory process generates insulin resistance, elevated blood glucose, and, in turn, increased sorbitol levels, which reduce anti-inflammatory molecules (like glutathione), producing fewer vasodilators (49). This exposure leads to greater susceptibility to atherosclerotic processes (17). Secondly, an increase in pro-inflammatory adipocytokines secreted by adipose tissue may lead to low-grade systemic inflammation, inducing various chronic diseases (8, 11). Higher body weight, especially obesity may cause multiple comorbidities and pose greater risks to human health. In addition, in present society, being overweight may lead to a person's low self-esteem and subsequent depressive symptoms, which in turn can have a negative impact on healthy aging (44). These mechanisms suggest that the overweight to obesity trajectory in older adults is an important marker of unhealthy aging and should be alerted.