Analyzing Frailty Trajectories and Predictive Factors in Older Chinese Adults: A Group‐Based Trajectory Modeling

ABSTRACT

Background

Existing interventions for geriatric frailty have demonstrated limited overall efficacy, underscoring the need to identify more precise frailty trajectories and develop targeted therapeutic strategies. Although substantial research has examined frailty prevalence among Chinese older adults, few studies have explored longitudinal trajectories, and limited evidence suggests that patterns resemble those observed in other countries. This study aims to identify distinct frailty trajectories and their multidimensional predictors, thereby informing personalized interventions to slow frailty progression.

Methods

This study analyzed data from 11,988 participants across four waves (2008–2018) of the Chinese Longitudinal Healthy Longevity Survey (CLHLS). Using group‐based trajectory modeling (GBTM), we iteratively fitted models of varying orders to identify optimal frailty development patterns in the elderly population. Next, frailty index (FI) levels and trajectory group effects were assessed using generalized estimating equations (GEEs). Finally, multinomial logistic regression was employed to identify multidimensional predictors of trajectory group membership.

Results

The optimal classification of frailty trajectories among Chinese older adults identified three groups: the low and stable group (53.0%), the moderate increasing group (28.4%), and the high‐level frailty group (18.7%). The GEE analysis revealed significant differences in frailty trajectories among participants (p < 0.05). Multinomial logistic regression results indicated that gender, age, education, residence, residential status, economic status, dietary diversity (DD), alcohol consumption, physical activity, cognitive activity, social activity, anxiety, and loneliness were significant predictors of frailty trajectory membership (p < 0.05). The high‐risk group exhibited a “cumulative disadvantage” profile characterized by female sex, age ≥85 years, rural residence, low economic status, experience of childhood hunger, moderate alcohol consumption, poor DD, physical inactivity, low cognitive activity, severe anxiety, pronounced loneliness, and low social participation.

Conclusion

This study identified three distinct frailty trajectories among older adults, revealing that high‐risk groups face a “cumulative disadvantage” arising from the convergence of socioeconomic deprivation, behavioral risks, and psychological vulnerability. These findings underscore the need for targeted early interventions and personalized management strategies to effectively address frailty.

Clinical Study Registration

Not applicable.

Abbreviations

CLHLS

Chinese Longitudinal Healthy Longevity Survey

FI

Frailty index

GBTM

Group‐based trajectory modeling

GEE

Generalized estimating equation

1. Introduction

Frailty, a transitional state between health and disease in older adults (Kim and Rockwood 2024), is characterized by a multidimensional and dynamic decline in physiological reserve, resulting in reduced resilience and adaptability (Fried et al. 2021; Clegg et al. 2013). This condition leads to results in increased dependency, vulnerability, and a heightened risk of mortality (Morley et al. 2013), affecting approximately 10% of individuals aged 65 and older and 25%–50% of those aged 85 and above (Cesari et al. 2016). Regionally, prevalence is highest in Oceania (31%), followed by Asia (25%), the Americas (23%), and Europe (22%) (O'Caoimh et al. 2021). The accelerated global trend of population aging underscores the urgency of addressing frailty. China, one of the world's most populous aging nations (Chen et al. 2022), is projected to have 395 million older adults by 2050, nearly 1.2 times the current population of the United States (United Nations 2022). Frailty contributes to increasing late‐life dependency (Morley et al. 2013). As frailty progresses, its progression elevates the risk of falls, disability, cognitive decline, and dependence, often leading to long‐term disability, hospitalization, and reliance on home care. These outcomes impose substantial burdens on individuals, families, and society (Morley et al. 2013; Álvarez‐Bustos et al. 2022). Consequently, frailty is recognized as growing public health priority amid China's rapidly aging population (Kim and Rockwood 2024; Fried et al. 2021; Cesari et al. 2017).

Frailty is a critical criterion for evaluating the overall health status of older adults, providing a precise assessment of multifaceted functional decline (Dent et al. 2017). It encompasses the deterioration and imbalance across physical, psychological, and social functions (Walston et al. 2018), thereby offering a comprehensive representation of the measure of overall health and functional capacity. Globally, various methodologies have been developed to assess frailty status, including the frailty phenotype (FP) (Hoogendijk et al. 2019), frailty index (FI) (Mitnitski et al. 2001), and clinical frailty scale (Lopez et al. 2012). Among these, the FP and FI are currently the two most widely recognized and internationally utilized approaches for frailty assessment. The FP, operationalized through five physical criteria–unintentional weight loss, reduced grip strength, self‐reported exhaustion, slowed walking speed, and diminished physical activity–defines frailty as the presence of three or more impaired indicators (Op het Veld et al. 2015). This method is well suited for rapid clinical screening; however, its limited consideration of multiple dimensions of frailty may introduce assessment bias (Veronese et al. 2021). First proposed by Rockwood and Mitnitski in 2001, the FI quantifies frailty by calculating the ratio of accumulated health deficits to the total number of health indicators assessed (Kojima et al. 2018). Compared with the FP, the FI demonstrates greater sensitivity to changes in exposure factors, making it more appropriate for longitudinal cohort studies (Malmstrom et al. 2014).

Research indicates that frailty can improve or worsen over time in response to external interventions, with considerable heterogeneity among individuals (Jang et al. 2022). Longitudinal studies can help establish more precise care pathways by distinguishing between different trajectories and their associated risk factors; however, this field remains in its early stages (Hoogendijk et al. 2019). Longitudinal studies examining transitions between frailty states based on FPs (non‐frail, pre‐frail, and frail) suggest that frailty is a heterogeneous and dynamic process (Pollack et al. 2017; Trevisan et al. 2017). Although most individuals maintain their baseline frailty status during follow‐up periods of 1–5 years, up to 37% experience at least one transition, involving either deterioration or improvement. Several longitudinal studies on frailty trajectories indicate that frailty generally increases with age (Stolz et al. 2017; Hoogendijk et al. 2018). However, there is considerable variation exists among older adults. Furthermore, research on the impact of birth cohorts on frailty is limited, and most studies rely on only two time points, leading to inconsistent findings (Chen et al. 2022; Walston et al. 2018). Overall, empirical observations reveal three broad patterns: (1) some older adults maintain robust physical functioning throughout their lifespan; (2) some experience acute declines in function following specific health events or environmental triggers; and (3) others exhibit progressive, age‐related deterioration in physical capabilities (Hwang et al. 2021). These divergent pathways in frailty progression contribute to marked disparities in physical health among older adults of the same chronological age (Mutambudzi et al. 2019). Consequently, stratifying elderly populations according to frailty status, using physiological and functional characteristics, is essential to elucidate frailty trajectories and guide targeted interventions.

Developmental trajectories describe the dynamic evolution of variables over time. Traditional longitudinal methods, such as mixed‐effects models and latent growth curve models, are limited by the assumption of uniform developmental trends across populations (Junjie et al. 2024). To overcome this limitation, the group‐based trajectory modeling (GBTM) was developed to identify heterogeneous subgroups with distinct temporal patterns (Elmer et al. 2018). Internationally, GBTM has been widely applied to classify frailty trajectories in older populations. For example, studies in Europe have identified trajectories such as stable progression, gradual decline, and rapid deterioration (Jenkins et al. 2023), while research in Korea has delineated subgroups including non‐frail, gradual frailty progression, and high‐severity frailty (Jang et al. 2022). The Toledo Healthy Aging Study further differentiated five trajectory types, ranging from improvement to severe frailty escalation (Álvarez‐Bustos et al. 2022). In contrast, evidence on frailty trajectories among Chinese older adults remains limited. Although a small number of recent studies have applied GBTM to frailty‐related outcomes in Chinese populations (Huan et al. 2023), these investigations have primarily focused on multi‐trajectory patterns involving frailty in conjunction with cognitive or psychological domains, or have been restricted to specific subpopulations. Importantly, a substantial proportion of existing studies in China continue to rely on mixed‐effects models, which estimate average population‐level trends but are limited in their ability to capture heterogeneity in individual developmental pathways (Yuan et al. 2023). Therefore, employing GBTM to systematically map long‐term frailty trajectories in the Chinese aging population remains essential. This approach not only facilitates a more nuanced understanding of context‐specific frailty dynamics but also provides an empirical foundation for developing precision‐oriented geriatric care strategies. In contrast, evidence on frailty trajectories among Chinese older adults remains limited. Importantly, existing studies in China predominantly rely on mixed‐effects models, which estimate average population‐level trends but fail to capture individual heterogeneity in developmental pathways (Yuan et al. 2023). Therefore, employing GBTM to systematically map frailty trajectories in the Chinese aging population is essential. This approach would not only clarify context‐specific dynamics but also establish a foundation for precision geriatric care strategies.

A longitudinal cohort study found that approximately 6%–7% of Chinese adults aged 60 years and older in community settings exhibit frailty. Timely interventions targeting modifiable risk factors may reduce mortality risk by 3%–5% (Chun‐xia et al. 2017). Another review reported that lifestyle and clinical risk factors associated with frailty can be modified through targeted interventions and preventive measures (Hoogendijk et al. 2019). These findings underscore the potential to mitigate frailty by systematically addressing controllable determinants. According to the Gobbens frailty integration model, frailty is influenced by various factors, including personal characteristics and diseases, which contribute to the gradual development of adverse outcomes. Life course determinants—such as socio‐demographic factors, socio‐economic status, and lifestyle—are related to the degree of frailty (Huan et al. 2023). However, current analyses of predictors of frailty trajectories remain limited. Exploring the interplay between these multifactorial influences and frailty trajectories is critical for designing stratified early interventions and evidence‐based policies tailored to diverse aging populations.

Using longitudinal data from the Chinese Longitudinal Healthy Longevity Survey (CLHLS), this study identifies distinct frailty trajectory patterns and examines their heterogeneity among middle‐aged and older adults in China through the application of GBTM. Based on these classifications, factors influencing frailty are selected according to the Gobbens frailty integration model. This analysis explores the associations between demographic characteristics, socioeconomic status, health behaviors, social participation, and mental health indicators with frailty trajectories. The findings aim to inform tailored interventions to slow frailty progression and contribute to evidence‐based strategies for promoting healthy aging.

2. Methods

2.1. Data Source

This study utilizes data from the CLHLS, a nationally representative, large‐scale prospective cohort study initiated in 1998 to investigate the social, environmental, and genetic determinants of aging among older adults in China. The CLHLS aims to improve health outcomes and promote longevity in aging populations by employing a sampling framework that covers approximately 85% of the national population ensuring broad geographic and demographic representation. Ethical approval for the survey was obtained from the Institutional Review Board of Peking University (Approval No. IRB00001052‐13074), and all participants provided written informed consent after receiving comprehensive information about the study's objectives and methods. The analytical cohort was derived from four waves of CLHLS data (2008, 2011, 2014, and 2018), initially comprising 16,954 individuals. Exclusion criteria were strictly applied as follows: (1) age under 65 years at baseline (n = 110); (2) missing critical baseline variables (n = 142); and (3) non‐mortality attrition during follow‐up (n = 4714). After these exclusions, the final cohort included 11,988 participants, with longitudinal follow‐up numbers of 6467 in 2011, 3887 in 2014, and 2351 in 2018.

2.2. Variable Selection

2.2.1. Frailty Index

This study employs the FI, originally introduced by Mitnitski and Rockwood in 2001 (Mitnitski et al. 2001), which is widely used to assess frailty in older adults. The FI encompasses various dimensions, including self‐care ability, chronic illness status, self‐reported health, and cognitive function. The specific indicators are as follows:

1. Self‐Care Ability

   The Activities of Daily Living (ADL), Instrumental Activities of Daily Living (IADL), and Instrumental Obstacles scales consisted of 6, 8, and 6 items, respectively. The scoring criteria were as follows: ADL items were scored 0 (no assistance required), 0.5 (partial assistance needed), or 1 (complete assistance needed); IADL items were scored 0 (able to complete the task), 0.5 (some difficulty completing the task), or 1 (unable to complete the task); and Instrumental Obstacles items were scored 0 (task can be completed with either hand), 0.5 (task can be completed with one hand only), or 1 (task cannot be completed with either hand).

2. General Health and Disease

   The self‐reported health status sections each consisted of a single item scored as follows: 0 = excellent, 0.25 = good, 0.5 = fair, 0.75 = poor, and 1 = very poor. Visual impairment was recorded as follows: 0 = no impairment, 0.5 = unclear vision, and 1 = blindness. Auditory impairment was recorded as follows: 0 = no difficulty, 0.5 = impairment in the left or right ear, and 1 = impairment in both ears. Chronic diseases were recorded using a cumulative score of up to 14, including conditions such as hypertension, diabetes, heart disease, stroke, and cerebrovascular diseases. The presence of each disease was scored as 1. Severe illnesses were rated as: 0 = no severe illness, 1 = one severe illness, and 2 = two or more severe illnesses.

3. Cognitive function was assessed using the cognitive assessment module available in the CLHLS dataset. To avoid reproducing item‐level or proprietary content, this study utilized only the total cognitive score, which was recoded as a frailty deficit component. The cognitive deficit component of the FI was coded as follows: 0–10 = 1, 11–17 = 0.75, 18–24 = 0.5, and 25–30 = 0. Higher values indicated a greater cognitive deficit burden.

The FI model can be represented as follows:

$$I_F=U/T$$

where $I_F$ represents the FI, U signifies the actual score for the above indicators, and T denotes the theoretical total score (fixed at T = 41 in this study). An FI closer to 0 indicated better health status, whereas an index closer to 1 indicated worse frailty status.

2.3. Influencing Factors

Based on the Gobbens Frailty Model (Yuan et al. 2023; Chun‐xia et al. 2017), the influencing factors were systematically categorized into five core domains: demographic characteristics, socioeconomic indicators, lifestyle and health behaviors, social engagement, and mental health. Demographic variables included sex, age, and ethnicity. Socioeconomic indicators encompassed educational attainment, residence, residential status, marital status, economic status, insurance coverage, and childhood hunger experience.

Lifestyle and health behaviors included dietary diversity (DD), smoking, alcohol consumption, physical activity, and cognitive activity. DD was measured by the variety of consumed food groups (e.g., vegetables, fruits, legumes, nuts, meat, eggs, fish, dairy, and tea), with participants classified into good or poor DD categories based on established scoring thresholds (Dent et al. 2017). Smoking status was self‐reported (Huan et al. 2023), while alcohol consumption was categorized as excessive, moderate, or abstinent (Huan et al. 2023). Physical and cognitive activity levels were stratified into low, medium, and high tertiles according to validated scoring systems. Social engagement was defined as participation in organized social activities and community services. Mental health was assessed using validated scales measuring loneliness (classified as severe, mild, or none) and anxiety severity (severe, mild, or none). Data analyzed in this study were obtained from the CLHLS and were collected by trained interviewers following standardized survey protocols to ensure data quality. Most variables, including sociodemographic characteristics, health behaviors, psychological factors, and health conditions, were self‐reported by participants. For respondents unable to complete the interview independently due to poor health or cognitive impairment, information was obtained from family members or other close proxies in accordance with CLHLS procedures. Objective measurements corresponding to several self‐reported indicators were not consistently available across survey waves, and detailed information on respondent type was not uniformly recorded. Therefore, all variables were analyzed as reported, without direct comparison with objective indicators or formal adjustment for respondent type. Detailed operational definitions and measurement protocols are provided in Table S1.

2.4. Data Analysis

2.4.1. Group‐Based Trajectory Models for FI

To examine patterns of frailty changes over time, we employed group‐based trajectory modeling (GBTM; Proc Traj in STATA), treating FI, a continuous variable ranging from 0 to 1, as a censored normal variable. The fundamental principle of the model posits that the longitudinal observation sequence of the ith individual at t time points is represented by [missing expression]. The notation signifies the probability associated with [missing expression]. Given that the overall population contains j distinct subgroups with different developmental trajectories, the model is formulated as follows:

$$P\left(y_i\right)=\sum _j^J{\pi }_j{P}^j\left(Y_i\right)$$

where ${\pi }_j$ represents the probability of group j; $P^j(Y_i)$ denotes the probability that individuals belong to group j. The model further assumes that the random variables $y_{it}$(t = 1, 2, …, T) are conditionally independent given that an individual belongs to group j. The formula can be represented as follows:

$$P^j\left(Y_i\right)=\prod _{t=1}^T{p}^{jt}(y_{it})$$

This represents the trajectory probability of individual i within group j in the GBTM, where T is the total number of observation time points. The term $p^{jt}(y_{it})$ denotes the probability of observing the outcome $y_{it}$ at time point t given that the individual belongs to group j. Longitudinal research often encounters nonrandom missing data, particularly in studies with high dropout rates, such as this one focusing on an elderly population. Having data that is missing not at random can compromise the validity of research findings, especially for subgroups with high dropout rates. To address this attrition bias, we utilized a dropout model in our trajectory analysis:

$$P^j\left(Y_i\right)=\prod _{t=1}^T\left[p^{jt}(y_{it})·\left(1-{\theta }_{jt}\right)\right]$$

where $p^{jt}\left(y_{it}\right)$ represents the probability of observing the outcome $y_{it}$ for an individual at time point t given that they have not exited the study; ${\theta }_{jt}$ represents the probability that individuals in group j exit the study at time point t, while $(1-{\theta }_{jt})$ denotes the probability that the individual has not exited the study before time point t. Consequently, the basic model extended to include dropout can be represented as follows:

$$P(y_i)=\sum _j^J{\pi }_j·\prod _{t=1}^T\left[{\rho }_j^t(y_{it})·(1-{\theta }_{jt})\right]$$

We fitted multiple development trajectories (ranging from 1 to 5 groups) with various functional forms (cubic, quadratic, linear, and intercept), and determined the optimal number of groups based on the Bayesian information criterion (BIC), which approached zero, the proportion of samples in each trajectory (minimum 5%), and average posterior probabilities (AvePP, near 1.0). Model selection prioritized the following criteria: (1) AvePP (AvePP ≥ 0.7 for acceptable classification; ≥ 0.8 for high reliability); (2) Bayesian and Akaike information criteria (BIC/AIC), with smaller values indicating better fit; and (3) statistical significance of trajectory parameters (p < 0.05) (Junjie et al. 2024; Elmer et al. 2018). As the BIC continued to improve with more groups, we primarily relied on the latter two criteria, and epidemiological relevance for our decision.

2.4.2. Multinomial Logistic Regression on Influencing Factors of Frailty Trajectory

Group‐based trajectory models were estimated using a maximum‐likelihood approach, which allows inclusion of all available observations for each participant and does not require complete data across all follow‐up waves. Participants with partially missing follow‐up data were retained in the analysis under the assumption of missing at random. Categorical variables were analyzed using the chi‐square test. After identifying distinct frailty development trajectory groups, the low and stable trajectory group (LT) was designated as the reference group. Analyses systematically examined differences in sociodemographic factors, health behaviors, social engagement, and mental health among individuals classified into these groups. Subsequently, multinomial logistic regression was applied to estimate odds ratios (ORs) and 95% confidence intervals (CIs), elucidating associations between various factors and frailty trajectory classification. Before analysis, multicollinearity among predictors was assessed using variance inflation factors (VIFs), all below 2, indicating acceptable levels (Elmer et al. 2018). Statistical significance was defined as p < 0.05.

3. Results

3.1. Sample Description

The study cohort consisted of a total of 11,988 older adults, comprising 57.8% females (n = 6925) and 42.2% males (n = 5063). The age distribution showed that the majority were aged 85 years and older (64.6%, n = 7743), followed by those aged 75–84 years (19.7%, n = 2361) and 65–74 years (15.7%, n = 1884). Detailed demographic characteristics are presented in Table 1.

TABLE 1.

Baseline sociodemographic characteristics of the study population(n = 11,988).

Variables	Categories	N	%
Gender	Male	5063	42.23
	Female	6925	57.77
Ethnic	Han	11,147	92.98
	Minorities	841	7.02
Age	65–74	1884	15.72
	75–84	2361	19.69
	≥85	7743	64.59
Education	Illiterate	7856	65.53
	Literate	4132	34.47
Residence	Urban	4094	34.15
	Rural	7894	65.85
Residential status	Living with family	10,022	83.6
	Living alone	1770	14.76
	Institution	196	1.63
Marital status	No spouse	8373	69.84
	Has a spouse	3615	30.16
Economic status	Poverty	2254	18.85
	Medium	8182	68.42
	Affluence	1522	12.73
Insurance	No	2870	23.94
	Yes	9118	76.06
Childhood hunger	Yes	8815	73.53
	No	3173	26.47

3.2. Descriptive Statistics and Correlation Analysis of FI

Table 2 summarizes the median values, interquartile ranges (IQRs), and correlation matrix of the FI across successive follow‐up waves in the elderly population. Pairwise correlations across all four survey waves were statistically significant (p < 0.01). The median FI score demonstrated a gradual decline over time: baseline measurements showed a median of 0.165 (IQR: 0.073–0.305), followed by decreases to 0.152 (IQR: 0.079–0.279) at the first follow‐up, 0.131 (IQR: 0.067–0.256) at the second, and 0.128 (IQR: 0.073–0.226) at the third assessment.

TABLE 2.

Descriptive statistics and correlation analysis of FI.

	2008	2011	2014	2018
2008	1	—	—	—
2011	0.593**	1	—	—
2014	0.468**	0.622**	1	—
2018	0.291**	0.433**	0.542**	1
M	0.165	0.152	0.131	0.128
(p 25, p 75)	(0.073, 0.305)	(0.079, 0.279)	(0.067, 0.256)	(0.073, 0.226)

Note: — indicates repeated data.

^**Indicates p < 0.01.

3.3. Selection of Optimal Subgroups for the Frailty Trajectories

The frailty trajectory analysis utilized the GBTM, testing models with one to four trajectory groups. In the 2‐group model, AvePP values were robust (Group 1: 0.834; Group 2: 0.879), and BIC/AIC values were lowest (BIC = 458.34; AIC = 506.63). However, the highest‐order polynomial term in Group 1 was not statistically significant (p = 0.120), which compromised the validity of its trajectory shape. The 3‐group model showed slightly lower but acceptable AvePP values (approximately 0.82 across groups) and improved BIC/AIC values (BIC = 1347.19; AIC = 1411.58) relative to the 4‐group model (AvePP = 0.78; BIC = 2489.05; AIC = 2561.42). Importantly, all polynomial terms in the 3‐group model were statistically significant (p < 0.001). Considering parsimony, classification reliability (AvePP > 0.7), and parameter significance, the 3‐group model was selected as optimal. Detailed results are presented in Table 3.

TABLE 3.

Comparisons of model fit indices across trajectory group specifications.

Subgroups	AvePP				BIC	AIC
	1‐Group	2‐Group	3‐Group	4‐Group
1‐Group	1 (1)				−2496.78	−2472.64
2‐Group	0.834 (0.500−1)	0.879 (0.501−1)			458.34	506.63
3‐Group	0.877 (0.500−1)	0.717 (0.457−0.999)	0.852 (0.467–0.999)		1347.19	1411.58
4‐Group	0.814 (0.390−1)	0.726 (0.351−0.999)	0.721 (0.345–1)	0.841 (0.394–1)	1778.93	1863.43

3.4. FI Scores Across Different Trajectory Groups

The developmental trajectory analysis identified three subgroups comprising 53.0%, 28.4%, and 18.7% of the sample, respectively. Subgroup 1 exhibited a gently increasing trend, with mean FI scores ranging from 0.10 to 0.13, reflecting a slow transition from a healthy state to pre‐frailty; this subgroup was classified as the low and stable trajectory group (LT). Subgroup 2 showed an increase in FI scores from baseline to the second follow‐up, after which the scores stabilized, with mean FI values ranging from 0.24 to 0.38. This pattern indicates progression from pre‐frailty to frailty, and the subgroup was designated the moderate increasing trajectory group (MT). Subgroup 3 demonstrated a continuous upward trend in FI scores from baseline through the third follow‐up, with mean scores ranging from 0.45 to 0.57. The FI consistently remained above 0.25, suggesting a sustained high level of frailty and is therefore categorized as the high trajectory group (HT). Table 4 presents the FI scores of the elderly population across these trajectory subgroups at baseline and each follow‐up. At all time points, the median FI score of the HT group was consistently higher than those of the MT and LT groups, while the MT group’ s median score was consistently higher than that of the LT group.

TABLE 4.

FI scores by trajectory group across follow‐up periods.

Trajectories	2008	2011	2014	2018
	M (p 25, p 75)	M (p 25, p 75)	M (p 25, p 75)	M (p 25, p 75)
LT	0.091 (0.048, 0.146)	0.098 (0.055, 0.152)	0.104 (0.054, 0.164)	0.115 (0.067, 0.183)
MT	0.256 (0.183, 0.304)	0.323 (0.250, 0.402)	0.366 (0.293, 0.467)	0.384 (0.299, 0.470)
HT	0.432 (0.381, 0.518)	0.506 (0.433, 0.574)	0.556 (0.482, 0.631)	0.571 (0.471, 0.652)

3.5. Characteristics of Population Distribution With Different Frailty Trajectories

The three frailty trajectory groups (LT, MT, and HT) exhibited statistically significant differences across most variables (p < 0.05), except for social services (p = 0.264). Significant group differences were observed in relation to gender, ethnicity, age, education, residence, residential status, economic status, marital status, childhood hunger, dietary diversity, physical activity, smoking, alcohol consumption, social activity, anxiety, cognitive activity, and loneliness. Among participants aged ≥85 years, 39.18%, 34.03%, and 26.79% were classified into the LT, MT, and HT groups, respectively, indicating a higher proportion of individuals in the moderate and high frailty trajectories, as detailed in Table 5.

TABLE 5.

Baseline characteristics of three trajectory groups (n = 11,988, %).

Variables	LT n = 6573	MT n = 3179	HT n = 2236	χ2	p
Gender				643.109	<0.001
Male	3442 (67.98)	1051 (20.76)	570 (11.26)
Female	3131 (45.21)	2128 (30.73)	1666 (24.06)
Ethnic				26.924	<0.001
Han	6046 (54.24)	2975 (26.29)	2126 (19.07)
Minorities	527 (62.66)	204 (24.26)	110 (13.08)
Age				2300.000	<0.001
65–74	1728 (91.72)	124 (6.58)	32 (1.70)
75–84	1811 (76.70)	420 (17.79)	130 (5.51)
≥85	3034 (39.18)	2635 (34.03)	2074 (26.79)
Education				569.897	<0.001
Illiterate	3698 (47.07)	2372 (30.19)	1786 (22.73)
Literate	2875 (69.58)	807 (19.53)	450 (10.89)
Residence				26.885	<0.001
Urban	2156 (52.66)	1071 (26.16)	867 (21.18)
Rural	4417 (55.95)	2108 (26.70)	1369 (17.34)
Residential status				205.763	<0.001
Living with family	5320 (53.08)	2662 (26.56)	2040 (20.36)
Living alone	1180 (66.67)	452 (25.54)	138 (7.80)
Institution	73 (37.24)	65 (33.16)	58 (29.59)
Marital status				968.891	<0.001
No spouse	3830 (45.74)	2558 (30.55)	1985 (23.71)
Has a spouse	2743 (75.88)	621 (17.18)	251 (6.94)
Economic status				74.922	<0.001
Poverty	1064 (46.91)	675 (29.76)	529 (23.32)
Medium	4623 (50.50)	2114 (25.84)	1445 (17.66)
Affluence	886 (57.61)	390 (25.36)	262 (17.03)
Social security				29.662	<0.001
No	1452 (50.59)	809 (28.19)	609 (21.22)
Yes	5121 (56.16)	2370 (25.99)	1627 (17.84)
Childhood hunger				24.425	<0.001
Yes	4739 (53.76)	2441 (27.69)	1635 (18.55)
No	1834 (57.80)	738 (23.26)	601 (18.94)
Dietary diversity				13.632	0.001
Good	3643 (53.41)	1881 (27.58)	1297 (19.01)
Poor	2930 (56.71)	1298 (25.12)	939 (18.17)
Smoking				303.574	<0.001
Currently	142 (75.13)	28 (14.81)	19 (10.05)
Previously	2476 (65.06)	855 (22.46)	475 (12.48)
Never	3952 (49.49)	2293 (28.72)	1740 (21.79)
Alcohol consumption				221.886	<0.001
Currently	1172 (70.99)	314 (19.02)	165 (9.99)
Previously	575 (53.44)	327 (30.39)	174 (16.17)
Never	4826 (52.11)	2538 (27.41)	1897 (20.48)
Physical activity				4100.000	<0.001
Low level	797 (21.25)	1154 (30.77)	1799 (47.97)
Medium level	2099 (57.29)	1200 (32.75)	365 (9.96)
High level	3677 (80.39)	825 (18.04)	72 (1.57)
Cognitive activity				1900.000	<0.001
Low level	1353 (31.39)	1429 (33.16)	1528 (35.45)
Medium level	1337 (58.87)	631 (27.79)	303 (13.34)
High level	3883 (71.81)	1119 (20.70)	405 (7.49)
Social activity				876.974	<0.001
Low level	4603 (48.24)	2796 (29.31)	2142 (22.45)
Medium level	1207 (78.22)	262 (16.98)	74 (4.80)
High level	763 (84.40)	121 (13.38)	20 (2.21)
Social services				2.667	0.264
Low level	2394 (55.02)	1121 (25.76)	836 (19.21)
High level	4179 (54.72)	2058 (26.95)	1400 (18.33)
Anxiety				294.095	<0.001
Severe	328 (48.24)	201 (29.56)	151 (22.21)
Mild	3415 (60.93)	1493 (26.64)	697 (12.44)
Never	2830 (49.62)	1485 (26.04)	1388 (24.34)
Loneliness				163.646	<0.001
Severe	451 (49.02)	289 (31.41)	180 (19.57)
Mild	3431 (60.04)	1562 (27.34)	721 (12.62)
Never	3712 (69.33)	1163 (21.72)	479 (8.95)

3.6. Multinomial Logistic Regression Results

To further examine the factors influencing latent frailty categories, this study employed frailty trajectory groups as the dependent variable. Using the LT group as the reference category, multinomial logistic regression was conducted with independent variables that were statistically significant in the univariate analysis. The results identified significant predictors of membership in both the MT and HT groups relative to the LT group (Table 6).

TABLE 6.

Multinomial logistic regression results.

Variables (ref.)		MT (ref.: LT)					HT (ref.: LT)
		OR	SE	Z	p	95%CI	OR	SE	Z	p	95%CI
Gender (male)	Female	1.701	0.121	7.450	<0.001	1.479–1.955	2.027	0.221	6.480	<0.001	1.637–2.51
Ethnic (Han)	Minorities	0.671	0.067	−4.010	<0.001	0.552–0.815	0.435	0.069	−5.210	<0.001	0.318–0.595
Age (65–74)	75–84	2.661	0.305	8.550	<0.001	2.126–3.331	2.320	0.588	3.320	0.001	1.412–3.813
	≥85	5.996	0.663	16.200	<0.001	4.827–7.446	6.444	1.525	7.880	<0.001	4.053–10.246
Education (illiterate)	Literate	1.001	0.066	0.010	0.993	0.879–1.139	0.547	0.046	−7.190	<0.001	0.465–0.645
Residence (urban)	Rural	0.747	0.043	−5.110	<0.001	0.667–0.835	0.378	0.053	−6.900	<0.001	0.287–0.498
Residential status (living with family)	Alone	0.740	0.056	−3.970	<0.001	0.637–0.858	1.146	0.309	0.510	0.613	0.675–1.945
	Institution	1.263	0.261	1.130	0.260	0.842–1.895	0.720	0.085	−2.780	0.005	0.571–0.908
Marital status (no spouse)	Has a spouse	0.928	0.066	−1.060	0.291	0.807–1.066	0.568	0.060	−5.360	<0.001	0.461–0.698
Economic status (poverty)	Medium	0.735	0.052	−4.330	<0.001	0.639–0.845	0.511	0.076	−4.490	<0.001	0.381–0.685
	Affluence	0.761	0.076	−2.750	0.006	0.626–0.924	1.378	0.122	3.630	<0.001	1.159–1.639
Social security (No)	Yes	1.229	0.076	3.320	0.001	1.088–1.388	1.042	0.093	0.460	0.646	0.875–1.24
Childhood hunger (Yes)	No	0.874	0.054	−2.190	0.029	0.775–0.986	1.308	0.108	3.240	0.001	1.112–1.538
Dietary diversity (good)	Poor	1.119	0.063	1.990	0.046	1.002–1.25	1.035	0.368	0.100	0.922	0.515–2.079
Smoking (currently)	Previously	1.563	0.382	1.830	0.068	0.968–2.524	1.077	0.380	0.210	0.833	0.54–2.149
	Never	1.557	0.379	1.820	0.069	0.966–2.51	1.415	0.265	1.850	0.064	0.98–2.043
Alcohol consumption (currently)	Previously	1.521	0.172	3.700	<0.001	1.218–1.899	1.478	0.205	2.810	0.005	1.126–1.941
	Never	1.293	0.111	2.980	0.003	1.092–1.53	0.130	0.012	−21.660	<0.001	0.108–0.156
Physical activity (low level)	Medium level	0.533	0.037	−9.100	<0.001	0.465–0.61	0.026	0.004	−23.280	<0.001	0.019–0.036
	High level	0.301	0.022	−16.170	<0.001	0.261–0.348	0.605	0.064	−4.750	<0.001	0.492–0.744
Cognitive activity (low level)	Medium level	0.710	0.052	−4.670	<0.001	0.615–0.82	0.473	0.047	−7.480	<0.001	0.389–0.576
	High level	0.686	0.046	−5.570	<0.001	0.601–0.783	0.628	0.103	−2.830	0.005	0.455–0.867
Social activity (low level)	Medium level	0.740	0.063	−3.560	<0.001	0.627–0.873	0.378	0.104	−3.530	<0.001	0.22–0.649
	High level	0.604	0.069	−4.390	<0.001	0.482–0.756	0.539	0.078	−4.250	<0.001	0.406–0.717
Anxiety (severe)	Mild	0.773	0.085	−2.330	0.020	0.622–0.96	0.427	0.068	−5.340	<0.001	0.313–0.584
	Never	0.746	0.088	−2.480	0.013	0.592–0.941	0.589	0.079	−3.930	<0.001	0.453–0.767
Loneliness (severe)	Mild	0.787	0.077	−2.460	0.014	0.65–0.953	0.548	0.084	−3.910	<0.001	0.405–0.741
	Never	0.675	0.074	−3.590	<0.001	0.544–0.836	0.914	0.440	−0.190	0.852	0.356–2.349

3.6.1. Demographic Characteristics

Females exhibited significantly higher odds of belonging to both the MT group (OR = 1.701, 95% CI: 1.479–1.955) and the HT group (OR = 2.027, 95% CI: 1.637–2.510) compared to males (p < 0.001). A pronounced age gradient was observed: compared to adults aged 65–74, those aged 75–84 had 2.66‐fold higher odds of MT membership (OR = 2.661, 95% CI: 2.126–3.331) and 2.32‐fold higher odds of HT membership (OR = 2.320, 95% CI: 1.412–3.813). Adults aged ≥85 years and older showed substantially elevated odds for both MT (OR = 5.996, 95% CI: 4.827–7.446) and HT (OR = 6.444, 95% CI: 4.053–10.246) groups (p < 0.001).

3.6.2. Socioeconomic Indicators

Rural residency was associated with significantly lower odds of both MT (OR = 0.747, 95% CI: 0.667–0.835) and HT (OR = 0.378, 95% CI: 0.287–0.498) membership compared to urban residency (p < 0.001). Although literacy showed no significant association with MT membership (OR = 1.001, 95% CI: 0.879–1.139, p = 0.993), it was strongly associated with a reduced likelihood of HT membership (OR = 0.547, 95% CI: 0.465–0.645, p < 0.001) compared to illiteracy. Medium economic status (vs. poverty) decreased the risks for both MT (OR = 0.735, 95% CI: 0.639–0.845) and HT (OR = 0.511, 95% CI: 0.381–0.685) membership (p < 0.001), whereas affluence was associated with lower odds of MT (OR = 0.761, 95% CI: 0.626–0.924, p = 0.006) but significantly higher odds of HT (OR = 1.378, 95% CI: 1.159–1.639, p < 0.001). Institutional residence predicted a lower likelihood of HT membership (OR = 0.720, 95% CI: 0.571–0.908, p = 0.005) compared to living with family. Compared with those who experienced childhood hunger, individuals without such experience had lower odds of belonging to the MT group (OR = 0.874, 95% CI: 0.775–0.986, p = 0.029) but higher odds of belonging to the HT group (OR = 1.308, 95% CI: 1.112–1.538, p = 0.001).

3.6.3. Lifestyle and Health Behaviors

Moderate alcohol consumption (compared to excessive consumption) was associated with increased risks for both MT (OR = 1.521, 95% CI: 1.218–1.899) and HT (OR = 1.478, 95% CI: 1.126–1.941) membership (p < 0.01). In contrast, abstinence was linked to higher odds of MT (OR = 1.293, 95% CI: 1.092–1.530, p = 0.003) but significantly lower odds of HT (OR = 0.130, 95% CI: 0.108–0.156, p < 0.001). Physical activity demonstrated a strong protective gradient: compared to low activity, medium activity reduced MT odds by 47% (OR = 0.533, 95% CI: 0.465–0.610) and HT odds by 97% (OR = 0.026, 95% CI: 0.019–0.036), while high activity reduced MT risk by 70% (OR = 0.301, 95% CI: 0.261–0.348) and HT risk by 40% (OR = 0.605, 95% CI: 0.492–0.744) (p < 0.001). Higher levels of cognitive activity (medium/high versus low) were consistently associated with lower odds of both MT and HT membership (e.g., HT: medium OR = 0.473, 95% CI: 0.389–0.576; high OR = 0.628, 95% CI: 0.455–0.867; p < 0.01).

3.6.4. Social Engagement and Mental Health

Higher levels of social activity (medium/high vs. low) significantly decreased the odds of membership in both the MT and HT groups (e.g., HT: high OR = 0.539, 95% CI: 0.406–0.717; p < 0.001). Reduced anxiety severity (mild/none vs. severe) was protective against both trajectories (e.g., HT: mild OR = 0.427, 95% CI: 0.313–0.584; none OR = 0.589, 95% CI: 0.453–0.767; p < 0.001). Mild loneliness (vs. severe) lowered the odds of MT (OR = 0.787, 95% CI: 0.650–0.953) and HT (OR = 0.548, 95% CI: 0.405–0.741), while the absence of loneliness reduced the risk of MT (OR = 0.675, 95% CI: 0.544–0.836).

4. Discussion

This study utilized longitudinal data from the CLHLS spanning the period from 2008 to 2018. It employs GBTM to identify overall developmental trends and heterogeneous trajectories of frailty among older adults. The results revealed a general linear decline in frailty levels over time, distinguishing three distinct trajectory subgroups: low and stable (LT; 53.0%), moderate increasing (MT; 28.4%), and high‐level (HT; 18.7%) frailty groups. Additionally, this study investigated key predictors of frailty progression, including demographic factors, socioeconomic status, health behaviors, social participation, and mental health. These findings provide valuable insights into the natural history of frailty in aging populations and serve as a foundation for tailored prevention and intervention strategies.

This study found that frailty exhibits stronger correlations between adjacent time points, with correlations weakening as the interval between assessments increases. The lowest correlation occurred between 2008 and 2018, indicating that early frailty status has substantial predictive value for short‐term outcomes but significantly diminished predictive power over extended periods (10 years). This highlights the necessity of dynamic frailty reassessment, consistent with previous research (Chun‐xia et al. 2017). The findings suggest a gradual improvement in frailty levels, albeit at a markedly slower pace over time. After 2014, frailty appeared to enter a “stable plateau phase,” suggesting increased challenges in achieving intervention effects during later stages. Although the median FI decreased, the FI in the HT group continued to worsen. This demonstrates that improvements in group averages may obscure deteriorating trends within high‐risk subpopulations. This finding contrasts with a Korean cohort study (Jung et al. 2022), which also identified three trajectory groups; however, its “high trajectory” group started with a higher FI, increased over time, and then plateaued during later follow‐up. The underlying mechanisms for these differences remain unclear but may relate to differences in the average age of the study samples. Furthermore, the FI is not a directly observed measure; some components of the FI may saturate in late life and cease to decline (Kim and Rockwood 2024; Fried et al. 2021; Jung et al. 2022). Importantly, the observed stabilization and percentile shifts at later waves should be interpreted with caution, as they may partly reflect changes in the underlying analytic sample due to attrition and selective survival of relatively healthier participants, rather than true slowing of frailty progression at the population level. The IQR carries dual implications. On the positive side, reduced disparities in frailty among the elderly. On negative, the 25th percentile value in 2018 remained unchanged from baseline, while the 75th percent significantly decreased. This pattern may therefore be influenced by survivor bias, whereby individuals with more severe frailty are less likely to remain under long‐term observation, potentially contributing to an apparent compression of the frailty distribution over time. Accordingly, the proposed interpretation of a characteristic pattern of “rapid initial progression followed by slower deterioration,” as well as the identification of a “golden intervention window,” should be regarded as hypothesis‐generating rather than causal. Therefore, based on the characteristic pattern of “rapid initial progression followed by slower deterioration,” we propose a “golden intervention window” Phase 1 (baseline to Year 3), emphasizing early interventions targeting the MT group to halt progression; and Phase 2 (after Year 7), focusing on disability prevention within the HT group (Kim and Rockwood 2024).

The study identifies three distinct developmental trajectories of frailty in the elderly and demonstrates that baseline characteristics can predict trajectory group membership. Frailty trajectories are categorized into three classes: LT, MT, and HT, consistent with findings from Japan (Bae et al. 2023), Korea (Jung et al. 2022), and the United States (Howrey et al. 2020). Differences compared to a French population survey (Tchalla et al. 2023) likely arise from smaller sample sizes and greater regional bias in those studies. Analysis of the LT and MT groups reveals that over half of older adults maintained a low and relatively stable frailty level throughout the extended observation period, while nearly one‐third experienced a period of clinically significant progression before subsequently stabilizing. In contrast, the HT group comprises nearly one‐fifth of older adults who are at high risk of frailty and experience continuous deterioration, consistent with findings from China (Wu et al. 2022) and Canada (Singh et al. 2025). The FI clearly confirms that these three trajectory groups represent distinct populations with increasing baseline frailty levels and faster progression rates. These findings highlight three key characteristics of elderly frailty trajectories: (1) Heterogeneity—frailty development is not uniform but exhibits significant variability; (2) Baseline predictability—the baseline FI level strongly predicts future frailty trajectory; and (3) Trajectory pattern differences—the groups differ not only in initial frailty levels but also in progression speed and pattern. These results emphasize the necessity for personalized management strategies (Kim and Rockwood 2024; Fried et al. 2021; Cesari et al. 2017). First, targeted intervention models and prevention strategies for elderly frailty must be developed (Bhattarai et al. 2024). Integrated care for older adults should be implemented (World Health Organization 2024; Braude et al. 2025), disparities in access to frailty care reduced (Geiger et al. 2016), and approaches such as Comprehensive Geriatric Assessment (CGA) (Kim and Rockwood 2024; Dent et al. 2017), multidisciplinary management (Cesari et al. 2017; Walston et al. 2018), and digital health technologies (Chen et al. 2022) employed to slow frailty progression. Second, early‐stage frailty is potentially reversible. Early identification and active intervention targeting modifiable factors can delay progression from health or pre‐frailty to frailty and disability (Cesari et al. 2017). Given the trajectory‐specific characteristics, differentiated intervention plans are recommended for the three groups. The LT group, although relatively stable, shows a slight upward trend (from health toward pre‐frailty); interventions for this group should focus on health maintenance, including regular health screening and intensive guidance on healthy lifestyles within community settings. The MT group demonstrates significant frailty progression in the early phase (approximately the first 6 years), representing a critical “golden intervention window.” For this group, comprehensive programs should be implemented, encompassing dietary diversity interventions, regular physical activity, and cognitive training, aiming to prevent further deterioration. The HT group, characterized by rapid and continuous frailty progression, requires the earliest and most intensive interventions. Multidisciplinary team care—including medical treatment, rehabilitation, psychological support, and disability prevention strategies—should be prioritized to mitigate further functional decline.

The results indicate that, compared to men, women have a significantly higher risk of belonging to both the MT and HT groups (p < 0.001), consistent with previous research findings (Jung et al. 2022; Bae et al. 2023). Age‐related declines in organ function and reduced estrogen levels in women contribute to the onset of frailty (Reproductive Aging Research as a Gateway to Health and Wellbeing 2016). This study identified a significant age gradient effect on frailty, with advancing age associated with an accelerated divergence in frailty trajectories, aligning with the cumulative deficit model of aging (Hoogendijk et al. 2018). In age‐stratified analyses, although baseline frailty levels varied across age groups, the HT group exhibited a similar upward trend in frailty progression across all age brackets. Although some studies report relatively stable frailty changes across age groups (Stolz et al. 2017; Hoogendijk et al. 2018), increasing evidence supports a nonlinear increase in frailty with age (Welstead et al. 2021; Jang et al. 2022). This underscores that an individual's rate of frailty progression depends on their initial frailty level‐those starting at higher frailty levels progress more rapidly. The subgroup analysis among ultra‐elderly individuals (≥85 years) further highlights the vulnerability of this population. A higher proportion of participants in this age group were classified into the moderate and high frailty trajectory groups, indicating an increased risk of adverse frailty progression in advanced age. This finding is consistent with the decline in physiological reserve and the cumulative accumulation of health deficits in late life. Given that the ultra‐elderly represent the fastest‐growing segment of the aging population, these results underscore the importance of early screening and targeted frailty prevention strategies for this particularly high‐risk group. Unexpectedly, rural residence was associated with a reduced risk of belonging to both the MT and HT frailty trajectory groups, which is contrary to conventional expectations. One potential explanation relates to environmental sensitivity in older adults. As individuals age, they may become more responsive to environmental stressors. Urban settings often present higher levels of environmental stress, reduced opportunities for social integration, and other challenges that could elevate frailty risk (Chen and Yu 2024). In contrast, rural older adults may maintain higher levels of physical activity through daily labor or household tasks and sustain stronger community ties, which could contribute to resilience against frailty. Another possible explanation is the survivor effect, whereby healthier individuals are more likely to remain in rural areas over time, creating a selective sample with lower apparent frailty risk. It should be emphasized that this finding does not imply that rural older adults generally enjoy better health; rather, it suggests the existence of distinct frailty development patterns between urban and rural populations. Although stratified analyses by urban and rural residence or mediation analyses examining the regulatory roles of lifestyle, social engagement, and health behaviors could provide deeper mechanistic insight, such analyses were beyond the scope of the current study. Therefore, this observation is presented as hypothesis‐generating, highlighting potential avenues for future research to clarify the core driving factors underlying the lower frailty risk observed in rural settings. Education level also predicted frailty risk, consistent with a 10‐year follow‐up study demonstrating that lower education increases frailty likelihood (Stolz et al. 2017). Interestingly, higher economic status was associated with a greater likelihood of belonging to the high frailty trajectory group. Although higher socioeconomic status is often considered protective for health, this finding may reflect the complex and stage‐specific relationship between economic resources and frailty progression. Individuals with higher economic status may be more likely to adopt sedentary lifestyles and engage in social drinking and frequent dining activities, which may increase metabolic burden and elevate the risk of chronic diseases. Additionally, greater access to healthcare resources may lead to more frequent or excessive medical interventions. Although improved medical care can prolong survival following major illnesses, it may also contribute to the gradual accumulation of physiological deficits over time. From a life‐course perspective, prolonged survival with multiple chronic conditions may ultimately lead to higher levels of frailty despite socioeconomic advantage. These findings suggest that the relationship between socioeconomic status and frailty is not uniformly protective and may vary across different stages of frailty development. This finding contrasts with results from Guo et al. (2022) for the Chinese population. Compared to individuals who experienced childhood hunger, those without such experience had a lower risk of belonging to the MT group but a higher risk of belonging to the HT group. In the early stages of frailty, adequate childhood nutrition may provide a metabolic advantage that reduces frailty risk. However, in advanced stages, the long‐term physiological resilience potentially developed through early‐life nutritional adversity may become insufficient to counteract cumulative age‐related vulnerabilities, thereby increasing the likelihood of severe frailty (Liu 2018). This suggests that the same exposure factor can have opposing effects at different points along the frailty continuum, underscoring the need for life‐course‐oriented and stage‐specific frailty management strategies. Therefore, efforts should prioritize follow‐up for older adults with high‐risk factors, such as advanced age and female sex, and promote adaptive frailty management across the life course. Given the gender and educational disparities in frailty among Chinese elders, it is essential to combine targeted screening and follow‐up with continuous health education to enhance health literacy, thereby preventing or delaying frailty onset.

Interestingly, moderate alcohol consumption among elderly men was associated with an increased risk of belonging to both the MT and HT groups, whereas alcohol abstinence significantly reduced the risk of HT group membership. This finding aligns with a meta‐analysis (Kojima et al. 2018). However, a 3‐year longitudinal study reported no association between heavy alcohol consumption and a reduced risk of frailty (Seematter‐Bagnoud et al. 2014). Poor dietary diversity was associated with a higher likelihood of MT group membership, consistent with previous research (Huang et al. 2025). Nutrition plays a crucial role in both the development and progression of frailty (Ni Lochlainn et al. 2021). This study found that physical activity has a strong protective effect against frailty in older adults. The 2019 International Clinical Practice Guidelines for Sarcopenia and Frailty, developed by the International Conference on Frailty and Sarcopenia Research (ICFSR), recognize physical activity as a key strategy to prevent and delay frailty (Gosselink 2022). Cognitive activity similarly reduced the risk of belonging to the HT group, with both moderate and high intensity showing significant effects. Research based on the CHARLS similarly reports that frequent participation in cognitive activities lowers the risk of frailty onset among Chinese elderly individuals (Ma et al. 2022). Engaging in cognitive activities may help preserve white and gray matter integrity, thereby reducing the risk of cognitive decline (Ribeiro et al. 2024). Furthermore, higher levels of social engagement were associated with a lower likelihood of membership in higher frailty trajectory groups. Participation in structured social activities may attenuate frailty risk through multiple pathways: fostering mutual support networks, enhancing a sense of belonging, and mitigating social isolation (Singh et al. 2025). Overall, it is recommended to guide older adults toward adopting healthy lifestyles by increasing dietary diversity, reducing alcohol consumption, engaging in regular physical activity, and enhancing cognitive stimulation. Efforts should also actively encourage participation in diverse social activities and develop built environments that facilitate physical activity, thereby reducing the risk of frailty.

Lower anxiety levels were significantly associated with a reduced risk of adverse frailty trajectory outcomes, supporting established evidence linking psychological distress to accelerated functional decline (Ye et al. 2024; Singh et al. 2025). Chronic anxiety may exacerbate frailty through two primary mechanisms: (1) dysregulation of immune and inflammatory responses, which amplifies somatic symptoms, and (2) hypervigilance toward health concerns, which may paradoxically reduce adherence to preventive behaviors. Loneliness emerged as an independent predictor of frailty trajectory escalation, consistent with global epidemiological trends (Lopez et al. 2012). Mechanistically, loneliness may impair physical function by diminishing motivation for activity engagement, thereby accelerating sarcopenia and mobility limitations‐hallmarks of frailty (Hwang et al. 2021; Jang et al. 2022). These findings underscore the importance of community‐ and family‐based initiatives that provide accessible psychological counseling services to the elderly, aiming to effectively alleviate diverse forms of emotional distress.

In summary, older adults at high risk of frailty experience a “cumulative disadvantage”, characterized by the combined effects of socioeconomic deprivation, behavioral risk factors, and psychological vulnerability. The typical profile of this high‐risk group includes the following characteristics: female sex; age ≥85 years; rural residence; poor economic status; experience of childhood hunger; moderate alcohol consumption; poor dietary diversity; physical inactivity; low cognitive activity; severe anxiety; pronounced loneliness; and low social participation. Therefore, it is recommended to prioritize targeted frailty screening for individuals with these risk factors and to develop risk assessment tools that incorporate socioeconomic indicators. Interventions should adopt a life‐course perspective, emphasizing prevention during the pre‐frailty stage and providing multidimensional support for moderately frail individuals to delay the progression of disability.

5. Limitations

First, analyses were based on baseline measurements of predictors, without accounting for changes over time. Although time‐varying analyses could provide deeper insight into frailty dynamics, inconsistencies in follow‐up measurements constrained such approaches. Second, substantial attrition over the 10‐year follow‐up may have introduced selective survival bias, as participants remaining in later waves were likely healthier, potentially affecting later‐stage trajectory estimates. Third, some variables were self‐reported or proxy‐reported, which may have introduced information bias, particularly among participants with poorer health or cognitive function. Fourth, the FI used equally weighted indicators for comparability, which may not fully capture region‐specific disease patterns. Fifth, the sample primarily consisted of community‐dwelling older adults, with underrepresentation of ultra‐elderly (≥85 years) and ethnic minority groups, limiting generalizability. Finally, as an observational study, causal inferences cannot be drawn; trajectory‐specific intervention implications should be considered hypothesis‐generating. Future research using repeated, standardized measurements, objective data, and more diverse populations is warranted to further validate and extend these findings. Although certain subpopulations were underrepresented, the large sample size and the stability of the trajectory modeling process support the robustness of the three identified frailty trajectories. Because trajectory classification is derived from internal longitudinal patterns rather than external population weighting, this limitation is unlikely to substantially alter the overall trajectory structure. Instead, it primarily affects the generalizability of subgroup‐specific interpretations.

6. Conclusions

This study identified distinct frailty trajectories among older adults and demonstrated that baseline characteristics are associated with different patterns of frailty progression. Older adults experiencing adverse frailty trajectories tend to face cumulative disadvantages, including socioeconomic deprivation, behavioral risks, and psychological vulnerability. These findings suggest that frailty prevention and management should move beyond a uniform approach and instead adopt personalized, life‐course‐oriented strategies. Early screening, timely identification of pre‐frailty, and targeted multidimensional interventions are essential to delay frailty progression and reduce the risk of disability. Particular attention should be given to the oldest‐old population, who may be more vulnerable to unfavorable frailty trajectories and require proactive, tailored preventive care.

Author Contributions

C.S. and G.M.W. contributed equally to this work. C.S., G.M.W., and Z.H.W. contributed to the study conception and design. Data analysis was performed by C.S. and G.M.W. The first draft of the manuscript was written by C.S. and G.M.W., and all authors commented on previous versions. All authors read and approved the final manuscript.

Funding

None to declare.

Ethics Statement

Ethical approval for the survey was obtained from the Institutional Review Board of Peking University (Approval No. IRB00001052‐13074).

Consent

All participants provided written informed consent after receiving comprehensive information about the study's objectives and methods.

Consent for Publication

Not applicable.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting Information: phn70134‐sup‐0001‐SupMat.docx

Supporting Information: phn70134‐sup‐0002‐STROBE‐checklist.pdf

Data Availability Statement

The dataset described in this paper can be obtained after sending a data user agreement to the data team on the website of Peking University Open Research. Data: https://opendata.pku.edu.cn/dataset.xhtml?persistentId = doi:10.18170/DVN/WBO7LK&version = 2.0.