How Neighborhood Structural and Individual Characteristics Affect Frailty Progression: Evidence from the China Health and Retirement Longitudinal Study

Abstract

Objectives

(1) To characterize the average trajectories of frailty over time in Chinese community-dwelling older adults; (2) To assess the effects of neighborhood structural and individual characteristics on frailty progression.

Design

A nationally representative prospective cohort study.

Setting

Communities in 28 provinces, China.

Participants

6238 respondents aged 60 and above in 447 communities from four waves of the China Health and Retirement Longitudinal Study.

Measurements

Frailty was measured using the 61-item Frailty Index (FI).

Results

The trajectory of FI was nonlinear, with an average growth rate of 0.025 that significantly slows down at the rate of 0.002 per year. Older, male, and uninsured respondents showed faster rates of growth in FI over time than younger, female, and insured counterparts. Respondents living in neighborhoods with a higher percentage of the older population and rural villages showed slower rates of growth in FI over time.

Conclusion

Expanding health insurance coverage and keeping a high clustering of the elderly in neighborhoods may be the potential strategies for population-level frailty prevention and interventions.

Introduction

Frailty, a dynamic process marked by decreased physiological reserves and increased vulnerability to stressors (1), is a severe challenge in an aging society. Evidence suggests that rapid frailty progression is a risk sign of adverse health outcomes (2, 3). Assessing the pattern of change in frailty over time can enlighten future intervention strategies to help older adults move toward a less harmful frailty pathway. A few studies have estimated frailty progression in developed countries (4, 5, 6). Heterogeneity in the overall shape of the trajectories and the rate of change in frailty has been found. Few know how overall frailty progresses over time in Chinese older adults.

There is strong interest in identifying factors that influence progression to higher or lower degrees of frailty because some populations may experience faster progression than others. This knowledge can help identify older adults who may be more likely to improve or be at risk for a rapid decline. Most studies have examined individual-level factors, with mixed findings. Few studies have focused on neighborhood effects, despite the significant role of the neighborhood in health. Social disorganization theory suggests that the neighborhood structural disadvantage, such as low SES and more residential instability, could attenuate the residents' ability to establish social order and maintain informal social control and collective efficacy, leading to poor health (7, 8, 9). Furthermore, neighborhood structural characteristics typically determine the location and allocation of resources (10), which could affect the health needs of older adults.

A limited number of studies have explored the effect of neighborhood structural characteristics on frailty. However, there are a few limitations. Most did not consider clustering by neighborhood, which may lead to incorrect statistical inferences (11, 12, 13). Besides, many of the studies were cross-sectional studies that failed to capture the dynamic nature of the frailty and the accumulation of neighborhood effects over time, given that older adults typically spend more time in their neighborhoods (12, 14). Also, few studies considered the neighborhood age structure, which could confound the relationship between SES disadvantage, residential instability, and frailty (15). Evidence highlights the need to consider the neighborhood age structure- typically measured as the percentage of older adults- in neighborhood and health research (15, 16). No studies have reported the influence of neighborhood structural context on frailty in a Chinese setting.

Therefore, this study sought to 1) characterize the average trajectories of frailty over time in Chinese community-dwelling older adults and 2) assess the effects of neighborhood structural and individual characteristics on frailty progression.

Methods

Study Design and Sample

This study used data from four waves of the China Health and Retirement Longitudinal Study (CHARLS 2011–2018), which collects health and retirement information from a nationally representative sample of the Chinese middle-aged and elderly population. CHARLES conducted the baseline survey in 2011 and three follow-up surveys in 2013, 2015, and 2018, covering 28 provinces, 150 counties/districts, 450 communities, 10,257 households, and 17,708 individuals aged 45 years and older. It also collected community information at baseline from the local community officials and the interviewers. More details about the CHARLS have been available elsewhere (17). The CHARLS study had been reviewed and approved by the Institutional Review Board at Peking University.

Among 17708 respondents at baseline, we excluded 9906 respondents younger than 60, 1070 respondents with invalid frailty measure (more than 20% of items were missing according to commonly used criterion (18)), and three respondents with missing data on age and gender. Given that the current study focuses on frailty progression, we further restricted the sample to respondents who completed at least one follow-up survey for frailty. The final study sample included 6238 respondents in 447 communities. Of those study samples, 93.9% and 71.1% participated in three and four waves, respectively.

Measurements

Frailty

We used Frailty Index (FI) to measure frailty, considering that FI could adequately capture changes in frailty over time, even subtle ones, due to its continuous scale (19, 20). FI is a count of impairments and diseases, collectively known as deficits (21). Following previous studies using CHARLS (22, 23), we constructed a 61-item FI, covering multiple domains of health deficits (see Supplement Table 1). FI score of each participant was defined as the ratio of the existing deficits in the total number of evaluated items. If the health deficits are missing, we excluded the corresponding ones from the denominator and the numerator. Thus, the FI varied from 0 to 1, with a higher FI score indicating greater frailty.

Individual characteristics

We included several socio-demographic characteristics at baseline, such as age, gender (male, female), marital status (never married/widowed/divorced, and married), education (lower than elementary, and higher than/equal to elementary), and financial status (average/over, and poor). Due to many missing values for the income- or expenditure-related variables, we used the self-rated standard of living to assess individuals' financial status, as in previous CHARLS studies (24). We included three measures of health behaviors, i.e., smoking (at least one cigarette per day and currently smoking, yes/no), drinking (alcoholic use more than once a month in the past years, yes/no), and social engagement (participating in social activities in the past month, yes/no), because of their association with frailty (25, 26). We also included an indicator for health insurance (yes/no).

Neighborhood structural characteristics

We included neighborhood age structure (% of the elderly aged 65 and older), residential stability (% of migrant people living less than half a year in a village/community), and neighborhood SES (a 7-point scale, ranging from 1 for poor and 7 for rich). We also included neighborhood education (% of illiterate residents), considering its independent influence on the health of old adults (27). The CHARLS measures this variable as a percentage value or a 20% interval range (e.g., 0–20%, 20–40%). Following prior work (27), we transform the interval into an exact percentage using the midpoint. Considering the significant urban-rural differences in the neighborhood structural characteristics, we also included an indicator of neighborhood type, rural villages versus urban communities.

Statistical Analyses

Descriptive statistics at neighborhood and individual levels were reported as means ± standard deviations (SD) or percentages. The trajectories of FI over seven years were assessed by the three-level growth curve model, accounting for the three-level data hierarchy: up to four measurements of frailty (level 1) were nested within individuals (level 2) who were nested within neighborhoods (level 3). We handled missing values using an extension of the Fully Conditional Specification (FCS) multiple imputation approaches, which have been developed specifically for imputing three-level data (28). This approach preserves the correlation among individuals within the same neighborhood and the clustering of repeated measures within individuals during the imputation stage (29). The ml. lmer function in the R package “miceadds” was used for multiple imputations. We produced twenty imputed data sets for model analysis.

We first conducted an unconditional model with no predictors to examine how much of the total variation in frailty trajectories can be accounted for by the individuals and neighborhoods, respectively. By computing the Intraclass Correlation Coefficient (ICC), 40.7% and 7.4% of the variability in FI trajectories can be explained by individuals and neighborhoods, respectively, which provides support for utilizing a three-level analytical model.

Then, unconditional growth models that contained the term(s) for time (years since baseline) were used to determine the trajectory pattern of FI over time and whether the trajectories of FI vary by individuals and neighborhoods. We compared the models with different functional forms (i.e., linear, quadratic) and random effect structures (i.e., random intercepts, random linear slopes, random quadratic slopes). A quadratic model with random intercepts and random linear slopes of the individual and neighborhood levels was fit best according to the likelihood ratio test. In the quadratic model, the intercept represented the baseline FI; the linear slope corresponded to the rate of change over time; the quadratic term indicated the change in the rate of change, that is, acceleration or deceleration over time.

Next, we added individual-level measures to the model at level 2 to examine the effect of the individual characteristics on the trajectories of FI. Specifically, we first examined the crude effect of every feature (model 1) and then its independent effect net of the others (model 2). Similarly, we added neighborhood-level measures to the model at level 3 to examine the effect of the neighborhood structural characteristics on the FI trajectories. Self- or social selection is a common issue in community-level analyses. Frail persons may select into neighborhoods actively or forcibly, based on individual attributes, which requests caution in interpretation because the associations between neighborhood characteristics and frailty may reflect the compositional effects rather than the contextual effects of neighborhood characteristics on health. Therefore, we examined the crude effect of every structural feature (model 3) and whether this effect remained significant after controlling for individual attributes that aim to minimize selection bias in the results (model 4). These analyses also can test whether neighborhood structural characteristics affect FI trajectories above and beyond individual features. Finally, we performed a full model (model 5), including all measures.

We tested for multicollinearity among the variables using the variance inflation factor (VIF); the mean VIF was 175.91, showing high multicollinearity. To reduce multicollinearity and facilitate parameter interpretation, we followed the recommendation by Enders and Tofighi (30) to center all continuous variables. Specifically, we centered the time variables at the baseline survey year (2011), the individual-level variables at the group mean, and the community-level variables at the sample mean. Binary variables were not centered because they already had a “meaningful zero” value for interpretation (30).

Results

This study included 6238 respondents (mean age 68.00 ± 6.62, 50.4% male) in 447 neighborhoods, with 1576 respondents living in 149 urban communities and 4662 living in 298 rural villages (Table 1). Of all respondents, the average FI at the baseline survey was 0.18. It increased steadily at follow-up two waves and then leveled off.

Table 1.

Descriptive statistics of neighborhood structural and individual characteristics

Variables	N (%) / Mean ± SD
Individual characteristics	N = 6238
Frailty Index 2011	0.18 ± 0.11
Frailty Index 2013	0.20 ± 0.13
Frailty Index 2015	0.26 ± 0.22
Frailty Index 2018	0.25 ± 0.15
Age, years	68.00 ± 6.62
Male	3146 (50.4)
Never Married/Divorced/Widowed	1276 (20.5)
Lower than elementary	3463 (55.5)
Poor standard living	2643 (42.4)
Current smoker	1909 (30.6)
Drinker	2007 (32.2)
Health insurance	5873 (94.2)
Social engagement	2947 (47.2)
Neighborhood structural characteristics	N = 447
Older population ratio (%)	15.27 ± 11.05
Illiterate resident ratio (%)	11.49 ± 12.55
Migrant population ratio (%)	6.95 ± 13.53
Rural village	298 (66.7)
Neighborhood SES	3.84 ± 1.34

SES= social economic status; SD=standard deviation

Table 2 shows the results of the unconditional growth model of FI. The likelihood ratio test showed that the quadratic model was better than the linear model (p<0.001). Results of the unconditional quadratic model show that the trajectory of FI was nonlinear, with an average growth rate of 0.025 (p<0.001) that significantly slows down at the rate of 0.002 (p<0.001) per year. The intercepts and linear growth rates demonstrated significant random variance across individuals and neighborhoods, indicating great individual and neighborhood variability in baseline level and growth trajectories of FI. There was significant covariance between the intercept and the linear growth rate at the individual level, but this covariance was not significant at the neighborhood level.

Table 2.

Unconditional growth model of Frailty Index

Effect	Linear model	Quadratic model
	Coefficient (SE)	Coefficient (SE)
Fixed effect
Intercept	0.187 (0.002) ***	0.175 (0.002) ***
Linear change rate	0.011 (0.000) ***	0.025 (0.001) ***
Quadratic change rate		−0.002 (0.000) ***
Random effect
Individual-level
Intercept	0.006 (0.000) ***	0.006 (0.000) ***
Linear change rate	0.000 (0.000) ***	0.000 (0.000) ***
Covariance	0.001 (0.000) ***	0.001 (0.000) ***
Neighborhood-level
Intercept	0.001 (0.000) ***	0.001 (0.000) ***
Linear change rate	0.000 (0.000) ***	0.000 (0.000) ***
Covariance	0.000 (0.000)	0.000 (0.000)
Residual variance	0.013 (0.000) ***	0.012 (0.000) ***
Likelihood-ratio tests	228.221***

SE= standard error; * p < 0.05; ** p < 0.01; ***p < 0.001;

Table 3 presents the crude and independent associations between individual characteristics and FI trajectories using three-level growth curve models. The results show that the trajectories of FI differ significantly by individual characteristics. First, in the adjusted model, age at baseline, gender, education, financial status, and social engagement was significantly associated with baseline FI. Specifically, one year increase in the baseline age was associated with a 0.003 higher baseline FI (p<0.001). The effect of gender was stronger than that of baseline age: Being female was associated with a 0.038 higher baseline FI (p<0.001). And the impact of education was the strongest of all: Having less than an elementary school education increased the baseline FI by 0.045 (p<0.001). Similarly, a poor standard of living increased one's baseline FI by 0.035 (p<0.001). Social engagement decreased the baseline FI by 0.026 (p<0.001).

Table 3.

The crade and independent effects of individual characteristics on frailty progression using three-level growth curve models

	Model 1									Model 2
	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)
Intercept	0.175 (0.002) ***	0.202 (0.003) ***	0.167 (0.003) ***	0.138 (0.003) ***	0.160 (0.003) ***	0.186 (0.003) ***	0.183 (0.003)***	0.185(0.007) ***	0.190 (0.003) ***	0.162(0.008) ***
Linear change rate	0.025 (0.001)***	0.021 (0.001) ***	0.024 (0.001) ***	0.023(0.001) ***	0.026 (0.001) ***	0.025(0.001) ***	0.025 (0.001) ***	0.038 (0.004) ***	0.027(0.001) ***	0.031 (0.005) ***
Quadratic change rate	−0.002 (0.000) ***	−0.002(0.000) ***	−0.002 (0.000) ***	−0.002 (0.000) ***	−0.002 (0.000) ***	−0.002 (0.000) ***	−0.002 (0.000) ***	−0.003 (0.001) ***	−0.002 (0.000) ***	−0.003 (0.001) ***
Intercept on
Age	0.003 (0.000) ***									0.003(0.000) ***
Male		−0.054(0.003) ***								−0.038(0.004) ***
Never Married/Divorced/Widowed			0.036 (0.004) ***							0.007(0.004)
Lower than elementary				0.068(0.003) ***						0.045(0.004) ***
Poor standard living					0.036 (0.003) ***					0.035(0.003) ***
Current smoker						−0.035(0.004) ***				−0.009(0.004)
Drinker							−0.027 (0.004) ***			0.006(0.004)
Health insurance								−0.011 (0.007)		0.005 (0.007)
Social engagement									−0.031 (0.004) ***	−0.026 (0.003) ***
Linear change rate on
Age	0.002 (0.000) ***									0.002(0.000) ***
Male		0.008 (0.002) ***								0.011 (0.003) ***
Never Married/Divorced/Widowed			0.007 (0.002) **							−0.001 (0.003)
Lower than elementary				0.004 (0.002) *						0.002 (0.002)
Poor standard living					−0.001 (0.002)					−0.001 (0.002)
Current smoker						−0.000(0.002)				−0.003 (0.002)
Drinker							0.001 (0.002)			−0.002 (0.002)
Health insurance								−0.013(0.004) **		−0.009 (0.004) *
Social engagement									−0.003 (0.002)	−0.002 (0.002)
Quadratic change rate on
Age	−0.000 (0.000) ***									−0.000 (0.000) ***
Male		−0.001 (0.000) ***								−0.001 (0.000) ***
Never Married/Divorced/Widowed			−0.001 (0.000) *							0.000 (0.000)
Lower than elementary				−0.001 (0.000) *						−0.000 (0.000)
Poor standard living					0.000 (0.000)					0.000 (0.000)
Current smoker						0.000 (0.000)				0.000 (0.000)
Drinker							−0.000 (0.000)			0.000 (0.000)
Health insurance								0.002(0.001)**		0.001 (0.001) *
Social engagement									0.000(0.000)	0.000 (0.000)

Notes: Estimates presented above are fixed effect coefficients; random effects are the same as table 2 but omitted. SE= standard error. Model 1: crude; Models 2: adjusted for all individual characteristics. * p < 0.05; ** p < 0.01; ***p < 0.001.

Second, age at baseline, gender, and health insurance was significantly associated with both the linear and quadratic growth rates in FI over time. Specifically, each year increase in the baseline age was associated with a 0.002 faster rate of growth in FI over time, while this growth slowed slightly over time (-0.0001, p<0.001). Males showed a faster rate of growth in FI over time than females (0.011, p<0.001), but this growth slowed significantly over time (−0.001, p<0.001). Figure 1 graphically depicts the trajectories of estimated FI for males and females that converge over time. Similarly, uninsured respondents showed a faster rate of growth in FI over time than insured counterparts (0.009, p<0.001), but this growth also slowed significantly over time (−0.001, p<0.001). Because the effect of health insurance on the baseline FI was not statistically significant, the trajectories of estimated FI for those with and without health insurance diverged over time from the same starting point, as shown in Figure 2.

Figure 1.

Estimated trajectories of the frailty index (FI) by gender

Note: We calculated the estimated value of FI based on model 2. When we calculated the FI for males and females, the other variables were held constant or for the reference group. For example, for males (gender =1) in the year 2011 (time =0), the estimated value of FI: FÎ=0.162−0.038x1+0.031×0+0.011×1×0−0.003×0²−0.001×1×0²=0.124. Subsequent figures were constructed similarly.

Figure 2.

Estimated trajectories of the frailty index (FI) by health insurance

Table 4 shows the crude and independent associations between neighborhood structural characteristics and FI trajectories using three-level growth curve models. The results show that FI trajectories also differ significantly by neighborhood structural characteristics. First, in the fully-adjusted model, neighborhood education and neighborhood type were significantly associated with the baseline FI. Specifically, each unit increase in the percentage of illiterate residents was associated with a 0.0001 higher baseline FI (p<0.05). The effect of neighborhood type is stronger than that of neighborhood education: living in rural villages increased the baseline FI by 0.022 (p<0.001). A higher percentage of the older population was also significantly associated with higher baseline FI. This association remained significant after controlling for individual characteristics, but no longer significant after further controlling for other neighborhood structural characteristics, suggesting that other structural characteristics may be the main contributors to this relationship.

Table 4.

The crude and independent effects of neighborhood structural characteristics on frailty progression using three-level growth curve models

	Model 3					Model 4					Model 5
	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)	Coefficient (SE)
Intercept	0.168 (0.004) ***	0.165(0.003) ***	0.176(0.003) ***	0.146(0.004) ***	0.191 (0.011) ***	0.156 (0.008) ***	0.156(0.008) ***	0.163 (0.008) ***	0.149(0.008) ***	0.175(0.012) ***	0.150(0.012) ***
Linear change rate	0.028 (0.002) ***	0.026(0.001) ***	0.025(0.001)***	0.029 (0.002) ***	0.026(0.002) ***	0.035 (0.005) ***	0.032(0.005) ***	0.031 (0.005) ***	0.034(0.005) ***	0.032 (0.005) ***	0.038(0.006) ***
Quadratic change rate	−0.002 (0.000) ***	−0.002(0.000) ***	−0.002(0.000) ***	−0.002 (0.000) ***	−0.002 (0.000) ***	−0.003 (0.001) ***	−0.003 (0.001) ***	−0.003 (0.001) ***	−0.003 (0.001) ***	−0.003 (0.001) ***	−0.003 (0.001) ***
Intercept on
Older population ratio (%)	0.000 (0.000) *					0.000 (0.000) *					0.000(0.000)
Illiterate resident ratio (%)		0.001 (0.000) ***					0.001 (0.000) ***				0.000 (0.000) *
Migrant population ratio (%)			−0.000 (0.000)					−0.000 (0.000)			−0.000(0.000)
Rural village				0.041 (0.005) ***					0.026 (0.005) ***		0.022(0.005) ***
Neighborhood SES					-0.004(0.003)					−0.003 (0.002)	−0.002(0.002)
Linear change rate on
Older population ratio (%)	−0.000 (0.000) *					−0.000(0.000) *					−0.000 (0.000) *
Illiterate resident ratio (%)		−0.000(0.000)					−0.000(0.000)				−0.000(0.000)
Migrant population ratio (%)			0.000 (0.000)					0.000 (0.000)			0.000(0.000)
Rural village				−0.005 (0.002) *					−0.006(0.002) **		−0.006 (0.002) *
Neighborhood SES					-0.000(0.001)					-0.000 (0.001)	−0.000(0.001)
Quadratic change rate on
Older population ratio (%)	0.000(0.000)					0.000(0.000)					0.000(0.000)
Illiterate resident ratio (%)		0.000 (0.000)					0.000 (0.000)				0.000(0.000)
Migrant population ratio ( %)			−0.000 (0.000)					−0.000 (0.000)			−0.000(0.000)
Rural village				0.001 (0.000)					0.001 (0.000) *		0.001 (0.000)
Neighborhood SES					0.000 (0.000)					0.000 (0.000)	0.000 (0.000)

Notes: Estimates presented above are fixed effect coefficients; random effects are the same as table 2 but omitted SE= standard error; SES= social economic status. Model 3: crude; model 4: adjusted for all individual characteristics; model 5: adjusted for all individual and neighborhood structural characteristics. *p<0.05; **p<0.01; ***p< 0.001.

Second, neighborhood age structure and neighborhood type were significantly associated with the linear growth rate of FI while not with the quadratic growth rate. Specifically, each unit increase in the percentage of the older populations was associated with a 0.0001 slower rate of growth in FI over time(p<0.05). Living in rural villages showed a much slower rate of growth in FI over time (−0.006, p<0.05). Figure 3 shows the trajectories of estimated FI for different neighborhood types that converge over time and cross over around 2015.

Figure 3.

Estimated trajectories of the frailty index (FI) by neighborhood type

Note: We calculated the estimated value of FI based on model 5.

Discussion

Assessing overall frailty trajectories in older adults and understanding how neighborhood structural and individual characteristics affect frailty progression is crucial for Chinese communities to prevent and intervene in frailty at the population level. Using the three-level growth curve models, we found a non-linear growth pattern in FI over time, with the rate of growth slowing over time. Individual characteristics such as age at baseline, gender, and health insurance were significant determinants of delayed frailty progression. Even after controlling for individual variables, living in neighborhoods with a higher percentage of the older population and rural villages contributed to slowing down frailty progression.

Changes in FI have been portrayed essentially as trajectories corresponding to age (1, 19). Although changes in FI with age could help understand the relationship between frailty and aging, changes over time have potential clinical relevance due to their association with the risk of death (2, 31). We found that the FI of older adults in Chinese communities showed an overall nonlinear growth, and the growth rate slowed down over time, which may be due to ceiling effects on the magnitude of FI growth (32). Besides the population pattern and individual variability, there was also small but significant variability in baseline levels and growth of FI across neighborhoods, suggesting that neighborhood characteristics also play a role in frailty development. Notably, the neighborhood-specific means of baseline level and growth rate were not systematically related, suggesting that the neighborhood-specific baseline level may not predict subsequent frailty progression.

Consistent with extant findings, this study shows that frailty trajectories are age-specific, with older persons exhibiting a higher degree of loss in physiological reserve and faster deficit accumulation than younger counterparts. There are also significant gender differences in the trajectories of frailty. Females show higher degrees of vulnerability than males at baseline, which is consistent with a previous Dutch study (33). However, in contrast to previous studies (2, 33), we found gender differences in the rate of deficit accumulation. Females accumulate deficits at a lower rate, suggesting that their excess in frailty lessens over time. One potential explanation is that females have better tolerance to damage and illnesses (34). The convergence of frailty differences between females and males confirms, to some extent, the sex paradox in which females have poorer health yet live longer than men. As in other studies (6, 33), we also find higher levels of frailty at baseline for the less-educated and the poor, and lower levels of frailty for respondents engaging in social activities.

Our results show that health insurance does not significantly affect the baseline frailty level, perhaps because of the small effect of health insurance on near-term health (35). However, it influences the rate of deficit accumulation. The insured experience substantially slower frailty progression than the uninsured do, suggesting that health insurance could buffer the progression of frailty. This buffering effect may be because health insurance improves access to medical services, thereby delaying deterioration in health (36). Although previous studies documented a protective effect of health insurance on frailty (37), these findings provide new evidence that this effect occurs mainly through slowing its progression rather than reducing its severity.

We found slightly higher degrees of frailty for individuals living in neighborhoods with higher concentrations of older populations, even after adjusting for individual characteristics. This finding aligns with research showing that residents of areas with a higher proportion of older people have lower mental health status (38). However, these individuals exhibit slower rates of deficit accumulation, suggesting that an older age structure of the community has a protective effect on the progression of frailty. This effect operates perhaps in two ways. First, in general, the demographic composition of a neighborhood could affect the availability of services and resources to residents, with a higher proportion of older people usually implying better service provision targeted toward the elderly (10). Second, demographic concentrations are often along with dense social support networks (39). Older adults living in neighborhoods with high concentrations of the elderly have better access to social support, which could protect against the rapid progression of frailty (40). To our knowledge, no study is known about the association between neighborhood age structure and frailty progression. Our study fills this gap in the literature. Although this study demonstrates the importance of a high clustering of older people in neighborhoods from a health perspective, the benefits of intergenerational living communities cannot be ignored. An increasing body of empirical studies has shown that intergenerational living neighborhoods can help reduce ageism, loneliness, and social isolation, and promote physical activity and health among older adults (41, 42).

In addition to neighborhood age structure, rural context appeared particularly protective for the frailty progression of older adults. We found that rural elderly had a higher degree of frailty at baseline but a slower increase over time than urban counterparts, resulting in decreasing frailty disparities. As shown in Figure 3, these disparities became insignificant around 2015. Since then, urban elderly has had a higher level of frailty than rural ones. The observed differences at baseline are consistent with previous studies (43). Although the finding for the rate of the increase seems to contradict the common impression that the health of people in poor areas worsens faster than those in affluent areas, it is not unexpected because we controlled neighborhood SES. Rural villages in China generally are less resourceful and economically disadvantaged, which could undermine health quickly (44). However, the health status of rural residents may weaken slowly after controlling for neighborhood SES. This slower deterioration of health may be due to tight connections between rural residents. Social connection is a well-established mechanism for protecting health, including frailty, either through increasing the ability of rural residents to access resources or seek support or contributing to the positive peer effects on health behaviors among rural residents (45, 46).

We did not find a significant association between neighborhood age structure, neighborhood type, and the quadratic growth rate of frailty. There are two possible explanations. First, neighborhood effects on frailty depend on an individual's exposure. The level of exposure determines the magnitude of the health change (47). It is possible that current exposure levels are not enough to drive changes in the rate of deficit accumulation. Second, from a stress-load perspective, the stress load associated with neighborhood structural characteristics is constant due to their relative constancy over time (48). Constant stress load corresponds to a linear change in the FI (49).

As expected, living in a neighborhood with a higher proportion of illiterate residents was associated with severe baseline frailty. This finding is consistent with other work examining neighborhood education and depression (27), highlighting the spillover effects of neighborhood education on individual health. Previous empirical studies have shown an association between neighborhood SES disadvantage and residential instability with frailty (11, 12, 13, 14). Our study did not confirm these findings, possibly because the data for neighborhood SES and residential stability come from the interviewers' or local officers' reports rather than more formal measurements. Future research could test these relationships using data from different sources, such as field surveys and census.

To our knowledge, this is the first study to quantify the frailty trajectories of older adults in Chinese communities using a nationally representative sample and to examine contributing factors at the individual and community levels. The main strength of this study was the use of a large and nationally representative sample which makes our research findings generalizable. Besides, we measured frailty using the FI that simultaneously encompasses multiple domains of health deficits (i.e., physiology, function, and psychology), which allows us to comprehensively conceptualize the contribution of different dimensions of neighborhood structural context to the physical and psychosocial frailty components. Moreover, this study used longitudinal models to assess the associations between individual and neighborhood characteristics and frailty trajectories, which reduces the risk of reverse causality and selection bias. In addition, we used three-level analysis strategies to address the nested, hierarchical nature of time-, individual- and neighborhood-level variables and to disentangle the effects of neighborhood-level variables from individual-level effects. More importantly, we used FCS multiple imputation approaches to deal with the missing data, which can reduce the potential sample attrition bias. Also, we took the clustered structure of the three-level data into account in the imputation process, which helps to produce unbiased parameter estimates.

Our study has several limitations. First, CHARLS collected neighborhood information only at baseline. We could not track the changes in the neighborhoods during the follow-up period. Therefore, this study only included individual and neighborhood characteristics at baseline. Future research could examine the effect of individual-level and neighborhood-level changes over time to provide valuable insights into individual and neighborhood conditions as causal determinants of frailty development. Second, neighborhood effects on frailty trajectories depend on an individual's location and exposure. Because we could not examine the respondents' location movement during the follow-up period, this study implicitly assumes that respondents keep the same position during the follow-up period as they did at baseline. Third, CHARLS has only collected four waves of data which is generally the minimum number of observations to determine growth trends (50). Future studies should use more waves of data to confirm our growth trajectories.

In conclusion, our study provides a comprehensive picture of how frailty develops over time and how neighborhood structural and individual characteristics affect frailty progression in China. Health insurance and living in neighborhoods with higher concentrations of older adults contribute to delaying frailty progression, which has important implications for policymakers and practitioners. Expanding health insurance coverage and keeping a high clustering of the elderly in neighborhoods may be the potential strategies for population-level frailty prevention and interventions. Rural settings also could protect against frailty progression.

Funding

None declared.

Conflict of Interest Statement

The authors declare no conflicts of interests.

Ethical standards

The CHARLS study complies with the current laws of the country in which it was performed. The study protocol was reviewed and approved by the Institutional Review Board at Peking University (IRB00001052-11015).

Electronic supplementary material

Supplementary material is available for this article at https://doi.org/10.1007/s12603-023-1916-1 and is accessible for authorized users.

Appendix 1