Study population
Detailed study designs of the CHARLS and ELSA can be found in Supplementary Information. For the current study, participants from these two cohorts were excluded if they met any of the following criteria: (1) were below 50 years of age at baseline; (2) had prevalent CMDs, dementia, or Parkinson’s disease at baseline; (3) lacked follow-up data on cardiometabolic outcomes of interest; or (4) had insufficient data to assess frailty status and cognitive function. Participant selection procedures are presented in Fig. S1.
Assessment of frailty
Frailty was evaluated using the frailty index (FI) and the physical frailty phenotype (PFP) approaches [16, 17]. To harmonize frailty assessments across CHARLS and ELSA datasets, 26 items were selected for FI construction, comprising self-reported health status, chronic diseases (excluding CMDs), physical function, and psychological conditions (Table S1). For all variables included in the FI, a score of 0 indicated the absence of a deficit, and a score of 1 indicated the presence of a deficit. The FI score was calculated as the ratio of the number of deficits present to the total number of deficits considered, with a higher score indicating greater frailty. Based on previous studies [18, 19], FI scores were categorized into three groups: robust (FI score ≤ 0.10), prefrail (FI score > 0.10 and < 0.25), and frail (FI score ≥ 0.25). Participants with more than 20% missing data across FI items (i.e., more than 5 items) were excluded (Fig. S1) [20]. The details of the PFP approach can be found in Supplementary Information.
Assessment of cognitive impairment
For CHARLS participants, cognitive impairment was assessed using a combination of three tests: specifically, the Telephone Interview for Cognitive Status (TICS-10), a word recall test, and a figure drawing test. The composite score of these three tests ranges from 0 to 21, with higher scores indicating better cognitive function [21]. For ELSA participants, cognitive impairment was assessed based on three tests: the word recall test (immediate and delayed recall, with a total score of 20 [10 points each]), the date naming test (maximum score of 4), and the verbal fluency test. In the verbal fluency test, participants were asked to name as many animals as possible within 60 seconds, and the number of animals named was recorded as the test score [22]. Participants scoring more than one standard deviation (SD) below age-appropriate norms were classified as cognitively impaired. Those scoring within one standard deviation of the norm or above were considered cognitively normal [23]. The procedures for these cognitive tests in the two cohorts are described in detail elsewhere [21, 22].
Assessment of cognitive frailty
Participants were categorized according to their frailty and cognitive status into the following groups: normal (non-frail and normal cognition), frailty only, cognitive impairment only, and cognitive frailty (co-occurrence of cognitive impairment and frailty).
Ascertainment of outcomes and endpoints
The primary outcome was incident major cardiometabolic diseases (CMDs), defined as the occurrence of either major cardiovascular diseases (CVDs; including heart diseases and stroke) or diabetes. Incident major CVDs and diabetes were also examined separately as secondary outcomes. In each wave of CHARLS and ELSA, participants were asked whether a doctor had informed them of a diagnosis of diabetes, heart diseases (including angina, heart attack, congestive heart failure, and other heart problems), or stroke. Participants reporting a diagnosis of heart disease or stroke were classified as having incident CVDs, and those reporting a diagnosis of diabetes were classified as having incident diabetes. Follow-up continued until the first occurrence of a major CMD, death, or the end of the follow-up period, whichever came first.
Covariates
The covariates included age (in years), sex (female or male), study region (China or the UK), marital status, education level, smoking status, alcohol consumption, and physician-diagnosed hypertension [13]. To ensure consistency between CHARLS and ELSA, marital status was dichotomized as either married/partnered or other (including separated, divorced, unmarried, or widowed). Education level was classified into two categories: less than high school and high school or above. Smoking status was grouped as current smokers, former smokers, or never smokers. Alcohol consumption was categorized as never drinkers and ever drinkers. Physician-diagnosed hypertension was included as a covariate but was not considered part of the major CMDs definition.
Statistical analysis
Descriptive characteristics were summarized across the four frailty and cognitive groups. Cox proportional hazards regression models were used to investigate the associations of cognitive frailty with the risk of three cardiometabolic outcomes. The proportional hazards assumption was assessed using Schoenfeld residuals, with no violations detected. In the main analysis, frailty was assessed using the FI, and the normal group served as the reference category. For the primary outcome (major CMDs) and two secondary outcomes (major CVDs and diabetes), hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using two models: Model 1 adjusted for age and sex; Model 2 further adjusted for study region (China or the UK), marital status, education level, smoking status, alcohol consumption, and hypertension. To assess potential synergistic effects, the risks in participants with cognitive frailty were compared with those in participants with frailty alone or cognitive impairment alone. Subgroup analyses were conducted by study region (China vs. UK) using separate Cox models based on Model 2 (excluding the stratification variable). Region-specific adjusted survival curves were generated to illustrate time-to-event differences across cognitive–frailty groups. Further stratified analyses were conducted within each cohort by age group (50–65 vs. >65 years) and by sex, with stratification variables excluded from the respective models. Multiplicative interaction terms between cognitive–frailty status and age group or sex were included to evaluate effect modification.
To further examine the role of physical frailty in cardiometabolic risk among individuals with cognitive impairment, we conducted restricted cubic spline (RCS) regression models treating the FI as a continuous exposure, using Cox proportional hazards models (Model 2) with FI = 0.25 set as the reference value. To evaluate the robustness of our findings, two sets of sensitivity analyses were performed. First, all primary and region-stratified analyses were re-estimated using Fine–Gray subdistribution hazard models to account for death as a competing risk [24]. Second, the associations were re-evaluated using the PFP as an alternative frailty measure [17], applying the same modelling strategy and covariate adjustments as in the main analyses.
All statistical analyses were performed using Stata version 18.0 (StataCorp LLC, College Station, TX, USA) and R version 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria). A P-value of less than 0.05 was considered statistically significant.