Abstract
Background
Health perception mismatch is common in older adults, yet its longitudinal patterns and prognostic value remain unclear. This study aimed to identify distinct mismatch trajectories and their associations with depression, cognitive decline, and mortality.
Methods
Data from three longitudinal ageing cohorts-the China Health and Retirement Longitudinal Study (CHARLS), English Longitudinal Study of Ageing (ELSA), and Health and Retirement Study (HRS)-were used to identify health perception mismatch trajectories and evaluate their associations with depression, cognitive impairment, and mortality. Baseline SRH-ADL/IADL interplay was examined within these longitudinal cohorts, while the National Health and Nutrition Examination Survey (NHANES) was retained as an independent external cross-sectional comparison. Associations were evaluated using multivariable regression, structural equation modeling, inverse probability of treatment weighting, causal forests, random-effects meta-analysis, and counterfactual prediction.
Findings
Baseline analyses in CHARLS, ELSA, and HRS showed heterogeneous and generally modest SRH-ADL/IADL additive interaction estimates, while NHANES provided an independent external cross-sectional comparison. Three stable perception mismatch trajectories, Mildly pessimistic, Mildly optimistic, and Highly optimistic, were consistently identified across CHARLS, ELSA, and HRS. Despite reporting better SRH, individuals in the Highly optimistic group had poorer baseline health and higher subsequent risks of depression, cognitive decline, and mortality. Counterfactual prediction analyses indicated 13–18% differences in mortality and depression risk between hypothetical trajectory contrasts, with the largest differences observed between the mildly pessimistic and highly optimistic classes, suggesting that discordantly favorable SRH may mark elevated risk.
Interpretation
Health perception mismatch trajectories may help identify older adults at elevated health risk. Discordantly optimistic perceptions may mark underlying vulnerability, suggesting that combining subjective and objective assessments could complement existing approaches to risk stratification. However, these observational findings do not establish that modifying health perceptions or trajectory membership would alter subsequent outcomes.
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1186/s12889-026-28394-x.
Keywords: Health perception mismatch, Self-rated health, Depression, Cognitive function, Mortality
Research in context
Health perception, how individuals evaluate their own health relative to their actual functional status, is a key construct in ageing research. However, the dynamic mismatch between self-rated health (SRH) and objectively measured function remains underexplored. To identify existing evidence, we searched PubMed and Web of Science for English-language studies published up to March 31, 2025, using the terms (‘self-rated health’ OR ‘subjective health’) AND (‘activities of daily living’ OR ‘IADL’ OR ‘functional limitation’) AND (‘mismatch’ OR ‘discrepancy’ OR ‘trajectory’ OR ‘discordance’). We included longitudinal or cross-sectional studies examining the relationship between subjective-objective health discrepancies and outcomes such as depression, cognitive impairment, or mortality.
Previous research has shown that poorer SRH and functional limitation independently predict adverse outcomes including depression, dementia, and premature death. Nevertheless, most studies treated these indicators separately, overlooking the joint implications of their discordance. Previous cross-sectional studies have suggested that favorable self-rated health despite functional impairment may identify individuals with elevated subsequent health risk. However, long-term trajectories of health perception mismatch and their prognostic relevance across populations and cultural contexts remain poorly characterized.
Added value of this study
Using harmonised data from three longitudinal ageing cohorts (CHARLS, ELSA, and HRS), with NHANES retained as an external cross-sectional comparison, we quantified the discrepancy between self-rated health and functional ability and identified three consistent trajectories of health perception mismatch: Mildly pessimistic, Mildly optimistic, and Highly optimistic. Individuals classified as Highly optimistic, those reporting good health despite functional limitations, had higher subsequent risks of depression, cognitive decline, and mortality. These associations were broadly consistent across several analytic approaches, including inverse probability weighting, causal forest modelling, random-effects meta-analysis, and model-based counterfactual prediction. Discordantly favourable self-rated health among individuals with functional impairment may therefore represent a risk-enriched profile for subsequent psychological and physical decline.
Implications of all the available evidence
Health perception mismatch trajectories may provide an additional means of identifying older adults whose subjective health assessments are discordant with their functional status. Combining subjective and objective health measures could complement existing approaches to early risk detection and inform the development of personalised screening strategies. Whether interventions targeting health-perception mismatch improve clinical outcomes requires prospective evaluation.
Introduction
The global population aged 65 and older is projected to nearly double from approximately 727 million in 2020 to about 1.5 billion by 2050, substantially increasing the burden of chronic diseases, functional decline, and mental health disorders worldwide [1]. This demographic transition is set to exert extraordinary economic and social pressure on health systems—driving up care costs, straining long-term care services, and heightening the burden on informal caregivers—while deepening health inequities and eroding overall societal productivity. In response to these profound challenges, rigorous evaluation of older adults’ functional and health status and the delivery of timely, tailored interventions are indispensable for averting adverse outcomes and fostering healthy ageing.
Self-rated health (SRH) and ADL/IADL are two of the most widely used measures in gerontological research. SRH captures an individual’s overall perception of health, while the sum of ADL and IADL assess one’s physical function impairment status [2, 3].
Evidence from longitudinal cohort studies indicates that individuals with poorer SRH are at greater risk of cognitive deterioration, depression, and premature death, irrespective of baseline clinical and sociodemographic characteristics [4]. Likewise, function predict incident dementia, mood disorders and shortened survival in older adults [5].
However, most studies have examined self-perceived health and objective physical functioning as separate predictors in older populations, with limited attention to the potential relevance of discordance between these measures [6]. Although emerging evidence has linked health-perception mismatch with depression, cognitive decline, and premature mortality, existing studies have not established whether the mismatch itself contributes causally to these outcomes or primarily serves as a marker of underlying vulnerability [7–9]. Moreover, an integrated quantitative approach for characterising the magnitude and longitudinal development of this mismatch remains lacking.
This study seeks to bridge the conceptual and empirical gap between subjective and objective dimensions of health by mapping longitudinal patterns of self-perception–function mismatch among older adults. Through examining how these trajectories translate into mental, cognitive, and survival outcomes across different sociocultural contexts, we aim to provide evidence for more equitable and perception-sensitive approaches to ageing research and policy.
Method
Study design and participants
This multi-cohort study utilized nationally representative data from four comparable datasets, including three longitudinal cohorts-CHARLS, HRS, and ELSA-and one cross-sectional survey, NHANES, which was retained as an external comparison dataset. Detailed study designs and settings have been described previously [10]. Participants aged > 60 years with complete data on ADL/IADL, SRH, depression, cognitive function, and death were included, excluding proxy interviews. The final analytical samples comprised 2,671 individuals (NHANES), 8,124 (CHARLS), 6,879 (HRS), and 3,227 (ELSA). The specific details are shown in Fig. 1 [10].
Fig. 1.
Flowchart of participant selection in three longitudinal ageing cohorts (HRS, ELSA, and CHARLS) and the NHANES external cross-sectional comparison sample
Measurement of cognitive decline
Cognitive function was assessed through standardized measures of episodic memory and executive function/mental status [11], with cohort-specific instruments detailed in Table S1. Scores were standardized to age group-specific z-scores (10-year intervals) within each cohort, with cognitive decline defined as > 1 SD below the age group mean [12].
Measurement of depression
Depression was assessed using validated self-reported instruments: the 10-item CES-D in CHARLS (cutoff > = 10), 8-item CES-D in HRS/ELSA (cutoff > = 3), and PHQ-9 in NHANES (cutoff > = 10). Scores below these thresholds were classified as non-depressed [13]. Details of the cohort-specific definition are in Table S2 [13].
Measurement of all-cause mortality
All-cause mortality was verified through cohort-specific protocols: CHARLS used family-verified exit interviews; [14] ELSA employed similar procedures; [15] HRS and NHANES linked to the National Death Index (NDI), with NHANES utilizing 12 identifiers for probabilistic matching (follow-up to 12/31/2019). Follow-up spanned from baseline to earliest of death, outcome, or administrative censoring (CHARLS Wave 4, ELSA Wave 6, and HRS Wave 13).
Measurement of ADL/IADL
ADL/IADL were evaluated using standardized ADL and IADL [16], with cohort-specific items detailed in Table S3. All items were dichotomously scored (1 = difficulty/unable; 0 = no difficulty), and standardized total scores were combined into a composite function impairment measure of functional limitation burden. The specific formula is as follows:
Measurement of SRH
SRH was assessed using standardized five-point Likert scales across cohorts, with responses ranging from optimal [1] to poor [5] health status.
Measurement of health perception mismatch
We used regression residuals to quantify health perception mismatch. Specifically, we regressed standardized SRH on standardized ADL/IADL. The resulting residuals represent the discrepancy where positive values indicate pessimism (better SRH than predicted by ADL/IADL) and negative values indicate optimism.
Covariates
The study considered several covariates, including demographic, socioeconomic characteristics, as well as health related factors. Specifically, the study included age, sex, marriage, education, annual household income, BMI, drinking history, smoking history, and comorbidity count.
Statistical analysis
Analyses were cohort-specific. To preserve methodological continuity, baseline additive interactions between SRH and ADL/IADL impairment were first examined separately in CHARLS, ELSA, and HRS before trajectory modelling. Additive interaction was assessed using the relative excess risk due to interaction (RERI) for cognitive decline, depression, and death, and pooled estimates were calculated across the three longitudinal cohorts. NHANES was retained as an independent external cross-sectional comparison to assess whether similar SRH-function interplay could be observed in a nationally representative sample with comparable measures. In NHANES, generalized linear models and structural equation modeling (SEM) were used to evaluate associations among SRH, functional impairment, and adverse outcomes, adjusting for demographic, socioeconomic, and health covariates.
For CHARLS, ELSA, and HRS, latent class growth models were fitted using repeated measurements across 3, 3, and 4 time points, respectively. Competing two-, three-, and four-class solutions were compared using information criteria, convergence status, entropy, minimum class size, mean posterior probability, and substantive interpretability. The three-class solution was selected because it improved model fit over the two-class solution while preserving clinically meaningful heterogeneity and provided a more stable and interpretable solution than the four-class model (Supplementary Tables 3 and 4). Participants were assigned to the trajectory class with the highest posterior probability for downstream analyses; trajectory-classification uncertainty was not explicitly propagated in subsequent outcome models. Classification quality was evaluated using entropy and mean posterior probabilities, as summarized in Supplementary Table 4. The Mildly pessimistic trajectory was selected as the reference group because it represented the least-discordant empirical profile and the closest available approximation to concordant health appraisal, rather than because the Mildly pessimistic trajectory was considered clinically optimal or psychologically normative. This approach is consistent with prior health congruence research distinguishing health realists, health pessimists, and health optimists according to the direction and magnitude of discrepancy between subjective and objective health [17–19].
Trajectory-outcome associations were assessed using multivariable-adjusted logistic (depression/ cognitive decline) or cox models (all-cause mortality). Fine–Gray competing-risk sensitivity analyses were additionally performed for depression and cognitive decline, with death before either outcome treated as a competing event. To assess the robustness of observed associations to potential unmeasured confounding, we calculated E-values for each statistically significant odds ratio (OR) or hazard ratio (HR), as well as for the lower bound of their 95% confidence intervals. Cohort-specific SEMs examined pathway consistency, with model fit evaluated by CFI > 0.90 and RMSEA < 0.06.
Across all cohorts, continuous variables were mean-centred. Inverse probability of treatment weighting based on generalised boosted models was used to address measured covariate imbalance across trajectory groups. Causal forest models with doubly robust estimation were used to characterise heterogeneity in trajectory–outcome associations. Counterfactual prediction methods were subsequently used to estimate differences in model-predicted outcome probabilities across trajectory profiles. All tests were two-sided, and p < 0.05 was considered statistically significant. To provide an overall estimate across the three longitudinal cohorts and formally quantify between-cohort heterogeneity, we additionally performed random-effects meta-analyses of the cohort-specific multivariable-adjusted associations between health perception mismatch trajectories and adverse outcomes. Odds ratios (ORs) were pooled for depression and cognitive decline, and hazard ratios (HRs) were pooled for all-cause mortality, using the Mildly pessimistic trajectory as the reference category. Statistical heterogeneity was assessed with the I2 statistic.
Results
Baseline SRH-function interplay and external cross-sectional comparison
Baseline additive interaction analyses showed heterogeneous but generally modest SRH–ADL/IADL interaction patterns across CHARLS, ELSA, and HRS (Supplementary Fig. 1). In pooled longitudinal analyses, RERI estimates were close to the null for cognitive decline (0.05; 95% CI, − 0.18 to 0.28), depression (0.15; 95% CI, − 0.07 to 0.37), and all-cause mortality (0.07; 95% CI, − 0.21 to 0.35). Thus, trajectory modelling was based on baseline SRH–function interplay within the longitudinal cohorts rather than on NHANES as a prerequisite validation dataset. NHANES was retained only as an independent external cross-sectional comparison, in which SRH and functional impairment showed independent associations with adverse outcomes in adjusted models, supporting the broader descriptive relevance of SRH–function discordance.
Three stable mismatch trajectories characterize aging across cohorts
As shown in Fig. 2, a three-class solution was retained as the primary trajectory specification in each cohort, yielding three interpretable trajectories labelled Mildly pessimistic, Mildly optimistic, and Highly optimistic. This solution provided an acceptable balance of model fit, convergence, class size, posterior classification probability, and substantive interpretability across CHARLS, ELSA, and HRS (Supplementary Tables 3 and 4).
Fig. 2.
Latent class growth trajectories of residuals between self-rated and objective health in CHARLS (a), ELSA (b), and HRS (c). Three distinct patterns were identified: Highly optimistic, Mildly optimistic, and Mildly pessimistic, reflecting stable cross-cohort gradients in subjective–objective health alignment
Our analysis identified a dominant Mildly pessimistic trajectory in CHARLS (74.5%), which contrasted with the more balanced distribution of Mildly optimistic (ELSA: 52.4%, HRS: 47.9%) and Mildly pessimistic (ELSA: 43.5%, HRS: 48.1%) trajectories in ELSA and HRS. The trajectories for the vast majority of participants across all cohorts were characterized by remarkable stability, with slopes approximating zero, indicating that an individual’s initial health perception mismatch remained largely unchanged over time. The Highly optimistic group exhibited divergent trajectory patterns across cohorts, with a significant decline in optimism in CHARLS (slope = 0.742), relative stability in ELSA (slope = -1.870), and further strengthening in HRS (slope = -2.060).
Prognostic profile of discordantly favorable self-rated health
Table 1 summarizes the baseline characteristics across the three identified mismatch trajectories in CHARLS, ELSA, and HRS. In all cohorts, participants in the Highly optimistic group-those reporting high SRH despite impaired function-had the most unfavorable health profile at baseline. This group exhibited the highest rates of depression, cognitive decline, and all-cause mortality. In CHARLS, for example, 62.8% had depression, 35.5% showed cognitive decline, and 16.5% died during follow-up. Similar patterns were observed in ELSA and HRS. Members of this group were also older, less educated, and had more chronic conditions. In contrast, the Mildly optimistic group consistently showed the lowest prevalence of adverse outcomes and more favorable socioeconomic and health indicators. The Mildly pessimistic group generally fell in between, though in CHARLS this group had a higher burden of chronic disease than the Mildly optimistic class. These patterns suggest that discordantly high SRH may serve as a potential marker of an adverse health-risk profile among individuals with functional impairment.
Table 1.
Baseline characteristics of study participants by trajectory class
| Characters | level | CHARLS | ELSA | HRS | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Mildly pessimistic | Mildly optimistic | Highly optimistic | p | Mildly pessimistic | Mildly optimistic | Highly optimistic | P | Mildly pessimistic | Mildly optimistic | Highly optimistic | P | ||
| n | 6055 | 1827 | 242 | 1428 | 1717 | 131 | 3312 | 3300 | 266 | ||||
| Depression (%) | No | 3499 (57.8) | 949 (51.9) | 90 (37.2) | < 0.001 | 1209 (84.7) | 1556 (90.6) | 100 (76.3) | < 0.001 | 2453 (74.1) | 2743 (83.1) | 136 (51.1) | < 0.001 |
| Yes | 2556 (42.2) | 878 (48.1) | 152 (62.8) | < 0.001 | 219 (15.3) | 161 (9.4) | 31 (23.7) | 859 (25.9) | 557 (16.9) | 130 (48.9) | |||
| Cognition decline (%) | No | 4959 (81.9) | 1362 (74.5) | 156 (64.5) | < 0.001 | 1189 (83.3) | 1502 (87.5) | 94 (71.8) | < 0.001 | 2784 (84.1) | 2861 (86.7) | 191 (71.8) | < 0.001 |
| Yes | 1096 (18.1) | 465 (25.5) | 86 (35.5) | 239 (16.7) | 215 (12.5) | 37 (28.2) | 528 (15.9) | 439 (13.3) | 75 (28.2) | ||||
| Death (%) | No | 5875 (97.0) | 1686 (92.3) | 202 (83.5) | < 0.001 | 1371 (96.0) | 1653 (96.3) | 109 (83.2) | < 0.001 | 2498 (75.4) | 2736 (82.9) | 149 (56.0) | < 0.001 |
| Yes | 180 (3.0) | 141 (7.7) | 40 (16.5) | 57 (4.0) | 64 (3.7) | 22 (16.8) | 814 (24.6) | 564 (17.1) | 117 (44.0) | ||||
| Gender (%) | Male | 3038 (50.2) | 839 (45.9) | 88 (36.4) | < 0.001 | 615 (43.1) | 742 (43.2) | 47 (35.9) | 0.257 | 1475 (44.5) | 1300 (39.4) | 85 (32.0) | < 0.001 |
| Female | 3017 (49.8) | 988 (54.1) | 154 (63.6) | 813 (56.9) | 975 (56.8) | 84 (64.1) | 1837 (55.5) | 2000 (60.6) | 181 (68.0) | ||||
| Age (%) | 61–70 | 3615 (59.7) | 897 (49.1) | 96 (39.7) | < 0.001 | 797 (55.8) | 991 (57.7) | 48 (36.6) | < 0.001 | 2015 (60.8) | 1976 (59.9) | 128 (48.1) | < 0.001 |
| 71–80 | 1993 (32.9) | 660 (36.1) | 91 (37.6) | 529 (37.0) | 556 (32.4) | 51 (38.9) | 1080 (32.6) | 1083 (32.8) | 90 (33.8) | ||||
| > 80 | 447 (7.4) | 270 (14.8) | 55 (22.7) | 102 (7.1) | 170 (9.9) | 32 (24.4) | 217 (6.6) | 241 (7.3) | 48 (18.0) | ||||
| Education (%) | No formal education or primary school | 1823 (30.1) | 798 (43.7) | 155 (64.0) | < 0.001 | 756 (52.9) | 717 (41.8) | 79 (60.3) | < 0.001 | 997 (30.1) | 586 (17.8) | 131 (49.2) | < 0.001 |
| Middle or high school | 2753 (45.5) | 693 (37.9) | 58 (24.0) | 253 (17.7) | 341 (19.9) | 19 (14.5) | 1111 (33.5) | 1099 (33.3) | 80 (30.1) | ||||
| College or above | 1479 (24.4) | 336 (18.4) | 29 (12.0) | 419 (29.3) | 659 (38.4) | 33 (25.2) | 1204 (36.4) | 1615 (48.9) | 55 (20.7) | ||||
| BMI (%, kg/m2) | 18–24 | 3454 (57.0) | 1033 (56.5) | 144 (59.5) | 0.715 | 237 (16.6) | 334 (19.5) | 16 (12.2) | 0.012 | 622 (18.8) | 906 (27.5) | 55 (20.7) | < 0.001 |
| < 18 | 285 (4.7) | 97 (5.3) | 13 (5.4) | 8 (0.6) | 4 (0.2) | 2 (1.5) | 15 (0.5) | 17 (0.5) | 3 (1.1) | ||||
| > 24 | 2316 (38.2) | 697 (38.1) | 85 (35.1) | 1183 (82.8) | 1379 (80.3) | 113 (86.3) | 2675 (80.8) | 2377 (72.0) | 208 (78.2) | ||||
| Drinking history (%) | No | 1593 (26.3) | 513 (28.1) | 53 (21.9) | 0.08 | 175 (12.3) | 144 (8.4) | 28 (21.4) | < 0.001 | 1764 (53.3) | 1454 (44.1) | 196 (73.7) | < 0.001 |
| Yes | 4462 (73.7) | 1314 (71.9) | 189 (78.1) | 1253 (87.7) | 1573 (91.6) | 103 (78.6) | 1548 (46.7) | 1846 (55.9) | 70 (26.3) | ||||
| Smoking history (%) | No | 2017 (33.3) | 560 (30.7) | 71 (29.3) | 0.057 | 515 (36.1) | 692 (40.3) | 58 (44.3) | 0.021 | 1364 (41.2) | 1537 (46.6) | 127 (47.7) | < 0.001 |
| Yes | 4038 (66.7) | 1267 (69.3) | 171 (70.7) | 913 (63.9) | 1025 (59.7) | 73 (55.7) | 1948 (58.8) | 1763 (53.4) | 139 (52.3) | ||||
| Marriage (%) | No | 641 (10.6) | 248 (13.6) | 42 (17.4) | < 0.001 | 91 (6.4) | 98 (5.7) | 12 (9.2) | 0.251 | 312 (9.4) | 343 (10.4) | 22 (8.3) | 0.281 |
| Yes | 5414 (89.4) | 1579 (86.4) | 200 (82.6) | 1400 (98.0) | 1695 (98.7) | 129 (98.5) | 3000 (90.6) | 2957 (89.6) | 244 (91.7) | ||||
| Annual household income (%) | Lowest | 1655 (27.3) | 569 (31.1) | 90 (37.2) | < 0.001 | 404 (28.3) | 407 (23.7) | 35 (26.7) | < 0.001 | 760 (22.9) | 547 (16.6) | 139 (52.3) | < 0.001 |
| Lower-middle | 1549 (25.6) | 474 (25.9) | 72 (29.8) | 425 (29.8) | 441 (25.7) | 62 (47.3) | 1005 (30.3) | 861 (26.1) | 75 (28.2) | ||||
| Upper-middle | 1431 (23.6) | 380 (20.8) | 46 (19.0) | 339 (23.7) | 437 (25.5) | 23 (17.6) | 931 (28.1) | 946 (28.7) | 38 (14.3) | ||||
| Highest | 1420 (23.5) | 404 (22.1) | 34 (14.0) | 260 (18.2) | 432 (25.2) | 11 (8.4) | 616 (18.6) | 946 (28.7) | 14 (5.3) | ||||
| Comorbidity count (mean ± SD) | 2.00 ± 1.63 | 1.83 ± 1.70 | 2.58 ± 1.62 | < 0.001 | 2.06 ± 1.36 | 1.55 ± 1.37 | 2.73 ± 1.46 | < 0.001 | 2.06 ± 1.24 | 1.47 ± 1.17 | 2.77 ± 1.46 | < 0.001 | |
P values were obtained using χ² tests for categorical variables and one-way ANOVA for continuous variables. Comorbidity count represents the number of self-reported chronic conditions, expressed as mean ± standard deviation (SD)
CHARLS China Health and Retirement Longitudinal Study, ELSA English Longitudinal Study of Ageing, HRS Health and Retirement Study
Highly optimistic trajectories are associated with higher psychological and survival risks
As shown in Fig. 3, multivariable-adjusted logistic and Cox regression models showed that the Highly optimistic trajectory was generally associated with higher risks of adverse outcomes across cohorts. The associations were most consistent for cognitive decline and all-cause mortality, whereas the association with depression was evident in CHARLS and HRS but only directionally positive in ELSA (Fig. 3; Supplementary Table 5).
Fig. 3.
Associations of trajectory classes with depression, cognitive decline, and death in CHARLS (a, b), ELSA (c, d), and HRS (e, f). Panels a, c, and e present unadjusted models, while panels b, d, and f show inverse probability of treatment weighting (IPTW)–adjusted models. Estimates are odds ratios or hazard ratios with 95% confidence intervals, using Mildly pessimistic as the reference category
Further sensitivity analyses supported the robustness and reliability of these findings, as demonstrated by inverse probability treatment weighting (IPTW), causal forest modeling (Fig. 4), and structural equation modeling (SEM; Table 2).
Fig. 4.
Heterogeneity in adjusted associations between trajectory profiles and depression, cognitive decline, and death in CHARLS (a–c), ELSA (d–f), and HRS (g–i), estimated using causal forest models
Table 2.
Structural equation modeling coefficients linking trajectory classes with depression, cognitive decline, and mortality
| Database | Outcome1 | Outcome2 / Predictor | Estimate | Std.Error | Z.value | P value | Std.All |
|---|---|---|---|---|---|---|---|
| CHARLS | Depression | Mildly optimistic | 0.094 | 0.035 | 2.719 | 0.007 | 0.038 |
| Depression | Highly optimistic | 0.376 | 0.083 | 4.5 | < 0.001 | 0.061 | |
| Cognitive decline | Mildly optimistic | 0.138 | 0.04 | 3.449 | 0.001 | 0.051 | |
| Cognitive decline | Highly optimistic | 0.274 | 0.088 | 3.113 | 0.002 | 0.041 | |
| Death | Mildly optimistic | 0.42 | 0.06 | 6.959 | < 0.001 | 0.145 | |
| Death | Highly optimistic | 0.84 | 0.109 | 7.712 | < 0.001 | 0.118 | |
| Depression | Cognitive decline | 0.141 | 0.021 | 6.775 | < 0.001 | 0.141 | |
| Depression | Death | -0.096 | 0.035 | -2.754 | 0.006 | -0.096 | |
| Cognitive decline | Death | 0.028 | 0.039 | 0.717 | 0.474 | 0.028 | |
| ELSA | Depression | Mildly optimistic | -0.275 | 0.06 | -4.561 | < 0.001 | -0.131 |
| Depression | Highly optimistic | 0.226 | 0.126 | 1.794 | 0.073 | 0.042 | |
| Cognitive decline | Mildly optimistic | -0.128 | 0.057 | -2.229 | 0.026 | -0.061 | |
| Cognitive decline | Highly optimistic | 0.378 | 0.123 | 3.074 | 0.002 | 0.071 | |
| Death | Mildly optimistic | -0.01 | 0.091 | -0.112 | 0.911 | -0.005 | |
| Death | Highly optimistic | 0.65 | 0.146 | 4.458 | < 0.001 | 0.114 | |
| Depression | Cognitive decline | 0.215 | 0.04 | 5.32 | < 0.001 | 0.215 | |
| Depression | Death | -0.068 | 0.059 | -1.147 | 0.251 | -0.068 | |
| Cognitive decline | Death | 0.027 | 0.055 | 0.465 | 0.642 | 0.027 | |
| HRS | Depression | Mildly optimistic | -0.018 | 0.039 | -0.465 | 0.642 | -0.009 |
| Depression | Highly optimistic | 0.247 | 0.090 | 2.752 | 0.006 | 0.045 | |
| Cognitive decline | Mildly optimistic | -0.269 | 0.038 | -7.152 | 0.000 | -0.120 | |
| Cognitive decline | Highly optimistic | 0.381 | 0.085 | 4.469 | 0.000 | 0.066 | |
| Death | Mildly optimistic | -0.266 | 0.036 | -7.412 | 0.000 | -0.126 | |
| Death | Highly optimistic | 0.454 | 0.083 | 5.497 | 0.000 | 0.083 | |
| Depression | Death | 0.030 | 0.027 | 1.134 | 0.257 | 0.030 | |
| Depression | Cognitive decline | 0.110 | 0.026 | 4.240 | 0.000 | 0.110 | |
| Cognitive decline | Death | 0.158 | 0.024 | 6.633 | 0.000 | 0.158 |
Standardized estimates (Std.All) are presented. Models were adjusted for age, sex, education, marital status, household income, BMI, drinking, smoking, and comorbidity count
CHARLS China Health and Retirement Longitudinal Study, ELSA English Longitudinal Study of Ageing, HRS Health and Retirement Study
The associations between the mildly optimistic trajectory and adverse outcomes were heterogeneous across cohorts (Fig. 3; Supplementary Table 5). In CHARLS, this trajectory was associated with higher risks of depression, cognitive decline, and all-cause mortality. In contrast, it showed inverse associations with depression in ELSA and HRS, no significant association with cognitive decline in either cohort, and a lower risk of all-cause mortality in HRS. Sensitivity analyses using IPTW, causal forest modelling, SEM, and Fine–Gray competing-risk models showed broadly similar patterns, although association estimates varied by cohort and outcome (Fig. 4; Table 2; Supplementary Table 6). E-values for statistically significant associations ranged from approximately 1.7 to 9.0, with the largest values observed for mortality outcomes in CHARLS (Supplementary Table 5). Exploratory interaction analyses indicated that trajectory-outcome associations may differ according to BMI, comorbidity burden, smoking, drinking, sex, education, and age. These findings were considered hypothesis-generating because the interaction patterns were not consistent across cohorts and outcomes (Supplementary Table 7).Random-effects meta-analyses across CHARLS, ELSA, and HRS further supported the adverse profile of the Highly optimistic trajectory (Supplementary Fig. 3). Compared with the Mildly pessimistic trajectory, the Highly optimistic trajectory was associated with higher pooled risks of depression and cognitive decline, with little between-cohort heterogeneity. The pooled association with all-cause mortality was also elevated, although substantial heterogeneity indicated that survival-related estimates should be interpreted cautiously.
Model-based hypothetical contrasts between health-perception trajectory profiles
Counterfactual prediction models (Fig. 5) showed that predicted outcome risks were generally higher under the Highly optimistic profile than under the Mildly pessimistic profile. In CHARLS, the predicted probabilities of depression, cognitive impairment, and mortality differed by approximately 13–15% between these hypothetical profiles. In ELSA, the corresponding difference in predicted depression risk was approximately 18%, accompanied by smaller differences in predicted cognitive decline and mortality. In HRS, the Highly optimistic profile was associated with an approximately 15% higher predicted probability of depression and a 14% higher predicted probability of mortality, whereas the difference in predicted cognitive decline was smaller but directionally consistent. These estimates represent model-based hypothetical contrasts conditional on measured covariates. They do not represent observed within-person transitions or the effects of interventions that move individuals between trajectory classes.
Fig. 5.
Model-based counterfactual predictions of outcome differences across hypothetical trajectory profiles in CHARLS (a-c), ELSA (d-f), and HRS (g-i). Mean predicted differences were estimated for trajectory contrasts conditional on the included demographic, socioeconomic, and health covariates. These estimates should not be interpreted as the effects of observed within-person transitions
Discussions
Although SRH and ADL/IADL have been widely studied separately, few studies have longitudinally examined their discordance and its associations with depression, cognitive decline, and mortality [20]. Using longitudinal data from CHARLS, ELSA, and HRS, we identified distinct trajectories of health-perception mismatch. Individuals classified as Highly optimistic, who reported good health despite functional limitations, had higher risks of adverse outcomes. Considerable heterogeneity was observed across trajectory groups and cohorts. These findings support further investigation of the processes underlying health-perception mismatch and its potential relevance to risk stratification and future intervention research, but they do not establish that modifying mismatch trajectories would change clinical outcomes.
The pooled analyses add an important layer of evidence to the cohort-specific findings. Across the three longitudinal cohorts, the Highly optimistic trajectory showed consistent associations with depression and cognitive decline, with no observed heterogeneity for either outcome, and remained associated with higher mortality risk after pooling. By contrast, the Mildly optimistic trajectory did not show significant pooled associations and displayed marked between-cohort heterogeneity. These findings suggest that the most discordantly optimistic profile may represent a more reproducible adverse-risk profile, whereas milder optimism may be more context dependent across populations. The substantial heterogeneity observed for mortality also indicates that pooled survival estimates should be interpreted cautiously.
The identification of the same three health-perception mismatch trajectories in CHARLS, ELSA, and HRS, despite differences in participant characteristics, is consistent with previous observations of relatively stable appraisal patterns between subjective and objective health. The Highly optimistic group, representing approximately 2% of participants in CHARLS and 8–10% in ELSA and HRS, reported favourable health despite functional limitations. Membership in this group was associated with approximately 1.5–2.0-fold higher odds of depression and cognitive impairment and markedly higher mortality estimates, although the magnitude of these associations varied across cohorts (Fig. 3). Roehr et al. reported a similar pattern and suggested that reduced awareness of physical decline may occur during the early stages of cognitive impairment [18]. Denial, benefit finding, social comparison, cultural norms of stoicism, and health literacy may contribute to differences in symptom appraisal across cohorts [19]. These factors may partly account for why discordantly favorable SRH marks a subgroup with higher observed psychological and physical risk.These findings highlight the importance of combining subjective assessments with functional measures in geriatric risk screening [21, 22]. Individuals in the Highly optimistic trajectory showed approximately 1.5–2.0-fold higher odds of depression and cognitive impairment, although associations with cognitive decline and mortality were not consistent across cohorts (Table 2; Fig. 4, and Fig. 5). One possible explanation is that reduced awareness of functional decline may accompany early cognitive or neurodegenerative changes, leading to discordance between perceived and measured health [23]. Stress-related biological and psychological pathways may also be relevant, although these mechanisms could not be directly tested in the present study [24]. Behavioral coping strategies such as denial and positive reappraisal can obscure emerging mood disturbances, reducing treatment seeking and intensifying neuroinflammation linked to depression and dementia [25]. Differences in health literacy and healthcare access may further shape how early symptoms are recognized, reported, and managed [26]. These factors may partly explain why discordantly favourable SRH identified a subgroup with elevated psychological and physical risks. The associations for the Mildly optimistic trajectory were less consistent, with higher risks observed in CHARLS but neutral or inverse associations in ELSA and HRS. This variation may reflect differences in cultural context, healthcare access, and the interpretation of self-rated health across populations. The variable cognition-mortality associations across cohorts may reflect differences in cultural context, healthcare systems, measurement, or residual confounding. In contrast, the Mildly optimistic trajectory was associated with higher risks in CHARLS but showed neutral or inverse associations in ELSA and HRS, suggesting that cohort-specific social, cultural, healthcare, and measurement contexts may modify these associations [27]. Recognizing such cohort-specific heterogeneity may improve culturally sensitive risk assessment and inform future intervention research.
Associations between optimistic mismatch profiles and adverse outcomes appeared stronger among participants with higher BMI or a greater burden of chronic disease, whereas weaker associations were observed among some participants with higher educational attainment or stronger social support (Table S8) [28]. Counterfactual predictions further illustrated this heterogeneity. The predicted mortality probability differed by up to 13.7% between the Highly optimistic and Mildly pessimistic profiles in one cohort, while the corresponding difference in predicted depression probability exceeded 18% in another cohort (Fig. 5). Smaller differences were observed among healthier or better-educated participants. These patterns are consistent with previous evidence that common risk factors may be associated with different outcomes across population subgroups. In clinical and public health settings, combining SRH, functional performance, BMI, comorbidity burden, and education may therefore provide a more detailed risk profile. Such stratification could help inform the development and evaluation of tailored screening, rehabilitation, and psychosocial support strategies. Recognising heterogeneity in trajectory–outcome associations may support a more nuanced approach to risk assessment in clinical and public health settings. Reliance on either subjective or objective health assessments alone may overlook vulnerable individuals, particularly those reporting favourable health despite functional limitations [29]. Combining SRH with physical performance measures, such as gait speed or grip strength, may help identify individuals at risk of functional decline who do not report corresponding difficulties [30]. Targeted screening for cognitive and emotional problems may therefore be particularly relevant for individuals with discordantly optimistic health perceptions. The observed variation according to BMI, chronic disease burden, and educational attainment also suggests that these characteristics may be useful when identifying higher-risk subgroups. Multidimensional assessment could consequently complement existing approaches to risk stratification and guide the design of future personalised prevention programmes. Prospective evaluation is needed to determine whether these strategies improve subsequent health outcomes.
This study has several limitations that warrant consideration. Although harmonisation efforts were made across CHARLS, ELSA, and HRS, differences in measurement instruments, follow-up periods, and cultural interpretations of health may have affected cross-cohort comparability. The trajectory classifications were derived from the available longitudinal data and may not fully capture latent heterogeneity or dynamic changes in mismatch states; classification uncertainty was also not propagated through all outcome models. Although multivariable adjustment, IPTW, causal forest modelling, and counterfactual prediction provided complementary assessments, residual confounding from unmeasured factors, such as personality traits, health literacy, healthcare access, and disease severity, cannot be excluded. Reverse causation is also possible, as early cognitive impairment, depressive symptoms, or subclinical disease may influence health appraisal before the corresponding outcomes become clinically apparent. Counterfactual estimates should therefore be interpreted as model-based comparisons across trajectory profiles rather than effects of actual transitions. Certain relevant mediators and moderators could not be harmonised across datasets. Future studies should examine these associations in broader populations, incorporate biomarkers and repeated health assessments, and evaluate whether addressing health-perception mismatch can improve subsequent outcomes.
Conclusion
Across CHARLS, ELSA, and HRS, we identified three consistent health-perception mismatch trajectories: mildly pessimistic, mildly optimistic, and highly optimistic. Membership in the Highly optimistic trajectory was associated with higher risks of depression, cognitive decline, and mortality despite favourable SRH. Incorporating both subjective and functional health measures into geriatric assessment may help identify risk-enriched profiles, inform targeted screening, and support the allocation of health and community-care resources.
Supplementary Information
Acknowledgements
We thank the participants and research teams of NHANES, CHARLS, ELSA, and HRS. We also acknowledge the Gateway to Global Aging Data and collaborating institutions for compiling harmonized resources for CHARLS, ELSA, and HRS that facilitate cross-cohort comparative ageing research.
Abbreviations
- ADL
Activities of Daily Living
- BMI
Body Mass Index
- CES-D
Center for Epidemiologic Studies Depression Scale
- CHARLS
China Health and Retirement Longitudinal Study
- CI
Confidence Interval
- ELSA
English Longitudinal Study of Ageing
- HRS
Health and Retirement Study
- HR
Hazard Ratio
- IADL
Instrumental Activities of Daily Living
- IPTW
Inverse Probability of Treatment Weighting
- LCGA
Latent Class Growth Analysis
- MMSE
Mini-Mental State Examination
- NHANES
National Health and Nutrition Examination Survey
- OR
Odds Ratio
- SD
Standard Deviation
- SEM
Structural Equation Modeling
- SE
Standard Error
- SRH
Self-Rated Health
Authors’ contributions
XYZ conceived and designed the study, supervised the analyses, and drafted the manuscript. KHX and XHH performed all statistical analyses and contributed to manuscript revision. ZHW and YDF assisted with data sourcing and analytic methods. DHZ and QZ provided supervision and critical review of the manuscript. KHX and XHH verified the underlying data. XYZ had full access to all the data and had final responsibility for the decision to submit the manuscript. All authors read and approved the final manuscript.
Funding
This work was supported by the Zhejiang Health Science & Technology Program — Commission–Province Joint Key Project (WKJ-ZJ-2517). The funder had no role in the study design, data collection, data analysis, data interpretation, manuscript preparation, or the decision to submit the manuscript for publication.
Data availability
The datasets analyzed during the current study are publicly available.NHANES data are available from the U.S. Centers for Disease Control and Prevention; CHARLS data are available from Peking University; ELSA data are available via the UK Data Service; and HRS data are available from the University of Michigan. No additional individual-level data beyond these publicly accessible sources were generated or analyzed for the present study.
Declarations
Ethics approval and consent to participate
CHARLS received approval from the Ethics Review Committee of Peking University. HRS received approval from the University of Michigan. ELSA received approval from London Multi-Centre Research. NHANES received approval from the National Center for Health Statistics and Ethics Review Board. All individuals signed informed consent. All procedures performed in the studies involving human participants were in accordance with the Declaration of Helsinki-Ethical Principles for Medical Research.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Xuanhan Hu, Zhihan Wu and Xinyu Zhang contributed equally to this work.
Contributor Information
Dechao Feng, Email: dechao.feng@ucl.ac.uk.
Yuchen Bai, Email: baiyuchen125@outlook.com.
Qi Zhang, Email: zhangqi1@hmc.edu.cn.
References
- 1.Strong K, Noor A, Aponte J, Banerjee A, Cibulskis R, Diaz T, et al. Monitoring the status of selected health related sustainable development goals: methods and projections to 2030. Glob Health Action. 2020;13(1):1846903. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Kaplan G, Baron-Epel O. What lies behind the subjective evaluation of health status? Soc Sci Med. 2003;56(8):1669–76. [DOI] [PubMed] [Google Scholar]
- 3.Chen K, Tan Y, Lu Y, Wu J, Liu X, Zhao Y. Effect of Exercise on Quality of Life in Parkinson’s Disease: A Systematic Review and Meta-Analysis. Parkinsons Dis. 2020;2020:3257623. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Bond J, Dickinson HO, Matthews F, Jagger C, Brayne C. Self-rated health status as a predictor of death, functional and cognitive impairment: a longitudinal cohort study. Eur J Ageing. 2006;3(4):193–206. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Zheng X, Yin Y, Yang L, Zhang X, Xiao S, Liang X, et al. Socioeconomic status and depression in later life: longitudinal mediation effects of activities of daily living. BMC Psychiatry. 2024;24(1):625. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Pedersen AK, Ernstsen L. Instrumental activities of daily living and self-rated health in community-dwelling older adults: cross-sectional findings from the HUNT Study (HUNT4 Trondheim 70+). BMC Geriatr. 2025;25(1):334. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Liu PL, Cohen HJ, Fillenbaum GG, Burchett BM, Whitson HE. Association of Co-Existing Impairments in Cognition and Self-Rated Vision and Hearing With Health Outcomes in Older Adults. Gerontol Geriatr Med. 2016;2. [DOI] [PMC free article] [PubMed]
- 8.Du M, Tao L, Liu M, Liu J. Trajectories of health conditions and their associations with the risk of cognitive impairment among older adults: insights from a national prospective cohort study. BMC Med. 2024;22(1):20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Peng MM, Wang P, Liang Z, Chen Z, Niroumand Sarvandani M. Moderating Effects of Age on the Longitudinal Associations Between IADL Disability, Self-Rated Health, and Depression Among Middle‐Aged and Older Adults. Health & Social Care in the Community. 2025;2025.
- 10.Banks J, Muriel A, Smith JP. Attrition and health in ageing studies: Evidence from ELSA and HRS. Longit Life Course Stud. 2011;2(2). [DOI] [PMC free article] [PubMed]
- 11.Crimmins EM, Kim JK, Langa KM, Weir DR. Assessment of cognition using surveys and neuropsychological assessment: the Health and Retirement Study and the Aging, Demographics, and Memory Study. J Gerontol B Psychol Sci Soc Sci. 2011;66(1):i162–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Levy R. Aging-associated cognitive decline. Working Party of the International Psychogeriatric Association in collaboration with the World Health Organization. Int Psychogeriatr. 1994;6(1):63–8. [PubMed] [Google Scholar]
- 13.Lu X, Yao Y, Jin Y. Corrigendum to Digital exclusion and functional dependence in older people: findings from five longitudinal cohort studies eClinical Medicine 54 (2022) 101708. EClinicalMedicine. 2023;64:102239. [DOI] [PMC free article] [PubMed]
- 14.Zhao Y, Hu Y, Smith JP, Strauss J, Yang G. Cohort profile: the China Health and Retirement Longitudinal Study (CHARLS). Int J Epidemiol. 2014;43(1):61–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Steptoe A, Breeze E, Banks J, Nazroo J. Cohort profile: the English longitudinal study of ageing. Int J Epidemiol. 2013;42(6):1640–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Katz S, Ford AB, Moskowitz RW, Jackson BA, Jaffe MW, STudies of illness in the aged. The index of ADL: A standardized measure of biological and psychosocial function. JAMA. 1963;185:914–9. [DOI] [PubMed] [Google Scholar]
- 17.Hong TB, Zarit SH, Malmberg B. The role of health congruence in functional status and depression. J Gerontol B Psychol Sci Soc Sci. 2004;59(4):P151–7. [DOI] [PubMed] [Google Scholar]
- 18.Ruthig JC, Chipperfield JG, Payne BJ. A five-year study of older adults’ health incongruence: consistency, functional changes and subsequent survival. Psychol Health. 2011;26(11):1463–78. [DOI] [PubMed] [Google Scholar]
- 19.Zhong Y, Cao J. Risk perception bias, lifestyle, and chronic disease risk in the middle-aged and older Chinese population: results of a national prospective longitudinal study. BMC Public Health. 2026;26(1):434. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Liao CC, Yeh CJ, Lee SH, Liao WC, Liao MY, Lee MC. Providing instrumental social support is more beneficial to reduce mortality risk among the elderly with low educational level in Taiwan: a 12-year follow-up national longitudinal study. J Nutr Health Aging. 2015;19(4):447–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Röhr S, Pabst A, Riedel-Heller SG, Jessen F, Turana Y, Handajani YS, et al. Estimating prevalence of subjective cognitive decline in and across international cohort studies of aging: a COSMIC study. Alzheimers Res Ther. 2020;12(1):167. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Idler EL, Benyamini Y. Self-rated health and mortality: a review of twenty-seven community studies. J Health Soc Behav. 1997;38(1):21–37. [PubMed] [Google Scholar]
- 23.Brewer R, Murphy J, Bird G. Atypical interoception as a common risk factor for psychopathology: A review. Neurosci Biobehav Rev. 2021;130:470–508. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Lei AA, Phang VWX, Lee YZ, Kow ASF, Tham CL, Ho YC et al. Chronic Stress-Associated Depressive Disorders: The Impact of HPA Axis Dysregulation and Neuroinflammation on the Hippocampus-A Mini Review. Int J Mol Sci. 2025;26(7). [DOI] [PMC free article] [PubMed]
- 25.Elshaikh U, Sheik R, Saeed RKM, Chivese T, Alsayed Hassan D. Barriers and facilitators of older adults for professional mental health help-seeking: a systematic review. BMC Geriatr. 2023;23(1):516. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Soldevila-Domenech N, Ayala-Garcia A, Barbera M, Lehtisalo J, Forcano L, Diaz-Ponce A, et al. Adherence and intensity in multimodal lifestyle-based interventions for cognitive decline prevention: state-of-the-art and future directions. Alzheimers Res Ther. 2025;17(1):61. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Huang D, Yang LH, Pescosolido BA. Understanding the public’s profile of mental health literacy in China: a nationwide study. BMC Psychiatry. 2019;19(1):20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Qu Y, Hu HY, Ou YN, Shen XN, Xu W, Wang ZT, et al. Association of body mass index with risk of cognitive impairment and dementia: A systematic review and meta-analysis of prospective studies. Neurosci Biobehav Rev. 2020;115:189–98. [DOI] [PubMed] [Google Scholar]
- 29.Kernick D, Chew-Graham CA, O’Flynn N. Clinical assessment and management of multimorbidity: NICE guideline. Br J Gen Pract. 2017;67(658):235–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Muhammad T, Hossain B, Das A, Rashid M. Relationship between handgrip strength and self-reported functional difficulties among older Indian adults: The role of self-rated health. Exp Gerontol. 2022;165:111833. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The datasets analyzed during the current study are publicly available.NHANES data are available from the U.S. Centers for Disease Control and Prevention; CHARLS data are available from Peking University; ELSA data are available via the UK Data Service; and HRS data are available from the University of Michigan. No additional individual-level data beyond these publicly accessible sources were generated or analyzed for the present study.