Intrinsic capacity impairment and incident dementia across three population-based cohorts: a harmonized individual participant data analysis.

Intrinsic capacity impairment and incident dementia across three population-based cohorts: a harmonized individual participant data analysis

Objectives

To examine whether intrinsic capacity impairment predicts incident dementia across three countries and to evaluate its value for risk stratification.

Design

: Harmonized multicohort study with individual participant data meta-analysis.

Setting and participants

Community-dwelling adults aged ≥50 years without baseline dementia from CHARLS (n = 17,708), ELSA (n = 9,432), and HRS (n = 22,034).

Measures

Intrinsic capacity was assessed across five domains: cognition, locomotion, vitality, sensory, and psychological functioning. Multidomain impairment was defined as deficits in ≥3 domains.

Results

Pooled hazard ratios rose in graded fashion: 1.31 (95% CI 1.22–1.42) for one impaired domain, 1.65 (1.48–1.85) for two, and 2.41 (2.18–2.68) for ≥3 domains. A Psych-Cognitive impairment pattern approached Multidomain risk (HR 2.15 vs 2.68). Rapid IC decline predicted 42.3% cumulative dementia incidence over six years. Adding IC to a demographic model improved the C-index from 0.69 to 0.76 with net benefit on decision curve analysis. Associations persisted when the cognitive domain was excluded and with death as a competing event; psychological functioning was the most consistent independent domain.

Conclusions

Intrinsic capacity impairment robustly predicts incident dementia across diverse populations. Its low cost and universal applicability support integration into geriatric assessment and primary care for early dementia risk stratification.

1Introduction

Dementia prevention requires scalable tools to identify at-risk older adults before clinical onset. The World Health Organization has proposed intrinsic capacity—the composite of physical, cognitive, sensory, and psychological functions—as the core construct for monitoring healthy ageing [1,2]. IC decline may represent a shared pathway through which multiple risk factors converge to produce late-life disability and dementia, yet its cross-national predictive validity for incident dementia remains unquantified. Prior studies have been predominantly single-country, used variable IC definitions, and focused on disability or mortality rather than dementia. A recent Mexican study found IC predicted 3-year dementia risk mainly among cognitively unimpaired adults, with locomotor function the only independently predictive domain [3]. To address this gap, we harmonized individual-level data from three nationally representative longitudinal ageing cohorts in China, the UK, and the US to examine the dose–response association between IC impairment and incident dementia, identify latent impairment patterns and longitudinal trajectories, and quantify the incremental predictive value of IC beyond conventional risk factors.

2Methods

We analyzed data from the China Health and Retirement Longitudinal Study (CHARLS, baseline 2011, n = 17,708) [4], the English Longitudinal Study of Ageing (ELSA, baseline 2004, n = 9,432) [5], and the Health and Retirement Study (HRS, US, baseline 2010, n = 22,034) [6]. All are biennial, nationally representative surveys of community-dwelling adults aged ≥50. Participants with complete baseline IC and dementia follow-up were included; those with baseline dementia were excluded.

IC was operationalized across five WHO ICOPE domains [7]: cognition (episodic memory, executive function), locomotion (gait speed), vitality (grip strength), sensory (self-reported vision/hearing), and psychological (depressive symptoms). Each domain was dichotomized as impaired below the cohort-specific 25th percentile [8]; a composite IC z-score was also calculated. Multidomain impairment was defined as ≥3 domains impaired. Incident dementia was identified using a harmonized composite definition: physician diagnosis or self-reported memory disease, or cognitive-test performance below 2.25 SD of the cohort-specific baseline mean [9]. Participants with baseline dementia by the same definition were excluded; follow-up was calculated to the first dementia event, death, or last interview.

Cox proportional hazards models estimated hazard ratios (HRs) for IC impairment count and per 1-SD decrease in IC score. A two-stage random-effects meta-analysis pooled cohort-specific estimates [10]. Restricted cubic splines examined nonlinearity. Latent class analysis identified IC impairment patterns; group-based trajectory modelling identified longitudinal trajectories (≥3 waves) [11]. Prediction models compared base demographics versus base + IC count/domains/trajectory, assessed by C-index [12], NRI, IDI [13], and decision curve analysis [14]. Leave-one-cohort-out external validation was performed. Sensitivity analyses included exclusion of early dementia (<2 years), Fine-Gray competing risks models [15], multiple imputation [16], and exclusion of the cognitive domain. Population attributable fractions were calculated [17]. All models adjusted for age, sex, education, marital status, household wealth, residence, smoking, alcohol consumption, physical activity, BMI, and history of hypertension, diabetes, heart disease, and stroke. Domain-specific associations were estimated in mutually adjusted Cox models; analyses stratified by baseline cognitive status examined non-cognitive IC (≥2 of four non-cognitive domains); death before dementia was treated as a competing event using Fine-Gray models [15] and 9-year cumulative-incidence functions; screening performance of IC count thresholds was summarized by sensitivity, specificity, and positive predictive value.

3Results

IC impairment distribution differed markedly across cohorts (Table 1). Sensory impairment was most prevalent (CHARLS 56.5%, ELSA 94.6%, HRS 90.5%), while multidomain impairment (≥3 domains) ranged from 14.7% (CHARLS) to 48.8% (ELSA) (Table A.1; Fig. A.1).Table 1Baseline characteristics and associations of intrinsic capacity impairment with incident dementia in CHARLS, ELSA, and HRS.

Table 1Characteristic CHARLS (n = 17,708) ELSA (n = 9,432) HRS (n = 22,034)

Baseline year 2011 2004 2010

Age, mean (SD) 58.2 (9.5) 65.3 (8.2) 67.5 (10.1)

Female, % 51.4 56.8 51.2

Education, %

Primary or less 45.2 28.5 15.3

Secondary 42.1 48.2 52.6

College or higher 12.7 23.3 32.1

IC impairment count, %

0 (No impairment) 15.7 0.3 1.1

1 (Mild) 44.1 18.7 26.8

2 (Moderate) 25.5 32.2 38.0

≥3 (Multidomain) 14.7 48.8 34.1

Chronic conditions, %

Hypertension 35.2 42.1 48.5

Diabetes 12.8 11.5 18.2

Heart disease 8.5 15.3 22.1

Stroke 2.3 3.2 4.5

Health behaviors, %

Current smoker 25.3 18.7 12.4

Current drinker 32.5 53.2 58.6

IC impairment and incident dementia, HR (95% CI)

1 domain vs 0 1.32 (1.18–1.48) 1.28 (0.99–1.65) 1.35 (1.13–1.61)

2 domains vs 0 1.68 (1.49–1.90) 1.52 (1.20–1.93) 1.78 (1.52–2.08)

≥3 domains vs 0 2.45 (2.15–2.79) 2.20 (1.76–2.75) 2.68 (2.28–3.15)

Per 1 SD decrease in IC score 1.25 (1.18–1.33) 1.22 (1.12–1.32) 1.28 (1.20–1.38)

P for trend <0.001 <0.001 <0.001

Models adjusted for age, sex, education, marital status, household wealth, residence (urban/rural), smoking status, alcohol consumption, physical activity, body mass index, and history of hypertension, diabetes, heart disease, and stroke. Pooled estimates were obtained by two-stage random-effects meta-analysis. Abbreviations: CHARLS, China Health and Retirement Longitudinal Study; ELSA, English Longitudinal Study of Ageing; HRS, Health and Retirement Study; IC, intrinsic capacity; HR, hazard ratio; CI, confidence interval; SD, standard deviation.

Progressively more severe IC impairment was associated with higher dementia risk in a dose–response fashion (Fig. 1A–B). Pooled HRs were 1.31 (95% CI 1.22–1.42) for 1 domain, 1.65 (1.48–1.85) for 2 domains, and 2.41 (2.18–2.68) for ≥3 domains versus none; per 1-SD decrease in IC score, risk increased 24% (HR 1.24, 1.18–1.30). Associations were consistent across cohorts (I² 8.2%–22.7%), and restricted cubic splines confirmed nonlinearity (P < 0.001). Cognitive impairment was the strongest individual predictor (HR 1.45), followed by psychological (HR 1.32) (Fig. A.2).

Latent class analysis identified five impairment patterns (Fig. 1C–D). Dementia risk increased stepwise from the Robust class (15.2% of participants) through Sensory-dominant (22.9%), Locomotion–Vitality (18.3%), and Psych-Cognitive (20.6%) to Multidomain (24.0%): HRs 1.35, 1.52, 2.15, and 2.68, respectively (P < 0.001; Table A.2).

Longitudinal trajectory analysis revealed four distinct groups (Fig. 2A–B). Six-year cumulative dementia incidence rose from 5.2% (Stable high, 27.2% of participants) to 12.8% (Moderate stable, 33.1%), 24.5% (Gradual decline, 22.2%), and 42.3% (Rapid decline, 16.7%); corresponding HRs were 1.58, 2.12, and 3.25 versus Stable high (Table A.3).

Adding IC to prediction models improved discrimination and clinical utility (Fig. 2C–D). The C-index increased from 0.69 (base model) to 0.76 with IC trajectory; NRI and IDI were significantly positive, calibration was satisfactory, and leave-one-cohort-out validation confirmed robust external performance (Table A.4; Fig. A.3; Fig. A.4; calibration in Fig. A.2).

Sensitivity analyses yielded consistent HRs for multidomain impairment (range 2.15–2.51; Table A.5). Associations persisted across demographic subgroups, with stronger effects at age ≥75 and ≥2 chronic diseases (Table A.6). Overall, 41.8% of incident dementia cases were attributable to any IC impairment (Table A.7).

In mutually adjusted models, psychological functioning was the most consistent independent domain (pooled HR 1.29, 1.17–1.43), followed by locomotion (1.09, 0.96–1.24); cognition was strongly but inconsistently associated (2.47, 1.61–3.79; I² = 97.2%), consistent with partial circularity with cognitive-test-based ascertainment, whereas sensory and vitality were not independently associated (Fig. A.5; Table A.8). Excluding cognition, non-cognitive IC remained associated with dementia (CHARLS 1.10 [1.00–1.21]; HRS 1.14 [1.02–1.28]).

In analyses stratified by baseline cognitive status, non-cognitive IC (≥2 of four non-cognitive domains) was associated with dementia in both strata, with a non-significant trend toward stronger associations among cognitively unimpaired participants (pooled HR 1.08 [0.96–1.21] vs 1.05 [0.95–1.17]; interaction P = 0.09, I² = 0%; Table A.9).

Associations persisted when death was treated as a competing event (CHARLS Fine-Gray HR 1.74 [1.46–2.09] for ≥3 vs 0 domains; HRS 1.82 [1.18–2.80]; Table A.10); over 9 years, cumulative dementia incidence rose from 11.8% (IC count 0) to 28.1% (IC count ≥3) in CHARLS and from 11.8% to 16.2% pooled (Fig. A.6).

4Discussion

In this harmonized analysis of 49,174 older adults, IC impairment robustly predicted incident dementia in a dose-dependent manner across three nations, with consistency across sensitivity analyses. This represents the largest cross-national validation of the WHO ICOPE construct for dementia [18]. The consistent dose–response across countries at different stages of epidemiological transition suggests IC captures universal brain health vulnerability.

The domain-specific analyses refine this picture: psychological functioning was the most consistent independent predictor, locomotion showed a modest cross-cohort concordant association partially supporting the Mexican study [3], cognition showed marked heterogeneity reflecting circularity with cognitive-test-based ascertainment, and sensory and vitality domains were not independently associated. The prognostic value of IC thus reflects additive contributions across functional systems, with psychological and motor domains warranting particular clinical attention.

Among the five impairment patterns, the Psych-Cognitive class carried a dementia risk approaching that of the fully impaired Multidomain class, consistent with evidence linking late-life depression and subjective cognitive decline to cerebrovascular damage, hippocampal atrophy, and chronic inflammation. Clinically, screening for depression and subjective memory complaints may identify a high-risk group missed by conventional cardiovascular algorithms. The trajectory analysis further supports the value of serial IC monitoring: rapid functional decline predicted a 42.3% cumulative dementia incidence [19,20], suggesting that the velocity of functional decline, rather than a single cross-sectional assessment, carries the strongest prognostic signal. In practice, biennial IC screening in primary care could flag rapid decliners for targeted evaluation.

Adding IC to a basic model improved the C-index from 0.69 to 0.76, with positive net benefit on decision curve analysis—an increment comparable to some blood-based biomarkers, yet IC is non-invasive and cost-free [21]. For geriatricians managing older adults with multimorbidity, serial IC assessment may distinguish functional decline signalling neurodegeneration from stable age-related deficits. At the population level, 41.8% of incident dementia cases were attributable to IC impairment, underscoring the potential magnitude of benefit if interventions preserving IC prove effective [[22], [23], [24]].

Stratification by baseline cognitive status showed a non-significant trend toward stronger effects among cognitively unimpaired participants, consistent with IC as an early vulnerability marker [3]. These results support a two-tier, ICOPE-aligned screening pathway in primary care—an initial five-domain IC screen, with individuals showing multiple impairments or rapid decline proceeding to focused cognitive assessment and a multidomain care plan (Fig. A.7); at an IC count threshold of ≥3, approximately 29% would be flagged (sensitivity 42.2%, specificity 72.3%). This pathway is proposed for risk stratification and service planning, not dementia diagnosis, and thresholds require prospective validation.

Limitations include potential circularity from including cognition within the IC construct (partly addressed by sensitivity analyses excluding the cognitive domain), cross-cohort measurement heterogeneity, unmeasured confounding, and limited generalizability to low-income settings. Because several IC domains have a single harmonized indicator, a five-factor confirmatory factor analysis could not be identified and measurement equivalence cannot be assumed; we provide measurement transparency data instead (Table A.8). Loss-to-follow-up flags were not consistently available; we therefore compared participants who died before dementia ascertainment with survivors (Table A.11) and treated death as a competing event. The observational design precludes causal inference. In conclusion, IC impairment is a strong, dose-dependent predictor of incident dementia. Its low cost and universal applicability support integration into routine geriatric assessment for early risk stratification.

CRediT authorship contribution statement

Yulong Zhou: Conceptualization, Methodology, Software, Validation, Formal Analysis, Investigation, Visualization, Supervision, Project Administration. Yingzheng Guo: Writing – original draft, Writing – review & editing.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Ethics statement

The studies involving human participants were reviewed and approved by the Institutional Review Boards of Peking University (CHARLS), the National Research Ethics Service (ELSA), and the University of Michigan (HRS). The patients/participants provided their written informed consent to participate in these studies. This secondary analysis of publicly available, de-identified data was exempt from additional ethical approval by the Institutional Review Board of Naval Medical University.

Declaration of competing interests

The authors declare that they have no competing interests

Data availability statement

Publicly available datasets were analyzed in this study. The CHARLS data can be found at http://charls.pku.edu.cn. The ELSA data can be found at the UK Data Service (https://ukdataservice.ac.uk). The HRS data can be found at https://hrs.isr.umich.edu. The GBD 2023 data can be accessed via the Global Health Data Exchange (https://ghdx.healthdata.org/). Further inquiries can be directed to the corresponding author.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the authors used Claude (Anthropic) to assist with language editing and formatting. The AI tool was used for approximately 10–15% of the writing process, primarily for sentence-level refinement of syntax, grammar, and adherence to journal-specific formatting requirements. All scientific content, data analysis, interpretation, and conclusions were generated entirely by the authors. The authors reviewed and edited all AI-assisted content and take full responsibility for the final manuscript.

References

1World Health OrganizationWorld report on ageing and health2015 https://www.who.int/publications/i/item/9789241565042/

2Beard J.R. Officer A. de Carvalho I.A. Sadana R. Pot A.M. Michel J.-P. The World report on ageing and health: a policy framework for healthy ageing Lancet. 387 2016 2145 2154 10.1016/S0140-6736(15)00516-4 26520231 PMC4848186

3López-Teros M.T. Yeverino-Castro S.G. Yocupicio-Medrano F. Mimenza-Alvarado A.J. Gutiérrez-Robledo L.M. Bello-Chavolla O.Y. Association between intrinsic capacity and dementia risk in older Mexicans Alzheimers Dement. 22 2026 e7157810.1002/alz.71578 PMC13275326 42309981

4Zhao Y. Hu Y. Smith J.P. Strauss J. Yang G. Cohort profile: the China Health and Retirement Longitudinal Study (CHARLS) Int J Epidemiol. 43 2014 61 68 10.1093/ije/dys203 23243115 PMC3937970

5Steptoe A. Breeze E. Banks J. Nazroo J. Cohort profile: the English longitudinal study of ageing Int J Epidemiol. 42 2013 1640 1648 10.1093/ije/dys168 23143611 PMC3900867

6Sonnega A. Faul J.D. Ofstedal M.B. Langa K.M. Phillips J.W.R. Weir D.R. Cohort profile: the Health and Retirement Study (HRS) Int J Epidemiol. 43 2014 576 585 10.1093/ije/dyu067 24671021 PMC3997380

7World Health OrganizationIntegrated care for older people (ICOPE): guidance for person-centred assessment and pathways in primary care2nd ed.2024 https://www.who.int/publications/i/item/9789240103726

8Beard J.R. Si Y. Liu Z. Chenoweth L. Hanewald K. Intrinsic capacity: validation of a new WHO concept for healthy aging in a longitudinal Chinese study J Gerontol A Biol Sci Med Sci. 77 2022 94 100 10.1093/gerona/glab226 34343305

9Jorm A.F. The Informant Questionnaire on cognitive decline in the elderly (IQCODE): a review Int Psychogeriatr. 16 2004 275 293 10.1017/s1041610204000390 15559753

10Riley R.D. Higgins J.P.T. Deeks J.J. Interpretation of random effects meta-analyses BMJ. 342 2011 d549 10.1136/bmj.d549 21310794

11Nagin D.S. Group-based trajectory modeling: an overview Ann Nutr Metab. 65 2014 205 210 10.1159/000360229 25413659

12Harrell F.E. Califf R.M. Pryor D.B. Lee K.L. Rosati R.A. Evaluating the yield of medical tests JAMA. 247 1982 2543 2546 7069920

13Pencina M.J. D’Agostino R.B. D’Agostino R.B. Vasan R.S. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond Stat Med. 27 2008 157 172 10.1002/sim.2929 discussion 207-21217569110

14Vickers A.J. Holland F. Decision curve analysis to evaluate the clinical benefit of prediction models Spine J. 21 2021 1643 1648 10.1016/j.spinee.2021.02.024 33676020 PMC8413398

15Austin P.C. Fine J.P. Practical recommendations for reporting Fine-Gray model analyses for competing risk data Stat Med. 36 2017 4391 4400 10.1002/sim.7501 28913837 PMC5698744

16Sterne J.A.C. White I.R. Carlin J.B. Spratt M. Royston P. Kenward M.G. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls BMJ. 338 2009 b2393 10.1136/bmj.b2393 19564179 PMC2714692

17Rockhill B. Newman B. Weinberg C. Use and misuse of population attributable fractions Am J Public Health. 88 1998 15 19 10.2105/ajph.88.1.15 9584027 PMC1508384

18Sánchez-Sánchez J.L. Lu W.-H. Gallardo-Gómez D. Del Pozo Cruz B. de Souto Barreto P. Lucia A. Association of intrinsic capacity with functional decline and mortality in older adults: a systematic review and meta-analysis of longitudinal studies Lancet Healthy Longev. 5 2024 e480 92 10.1016/S2666-7568(24)00092-8 38945130

19Zhou Y. Wang G. Li J. Liu P. Pan Y. Li Y. Trajectory of intrinsic capacity among community-dwelling older adults in China: the China health and retirement longitudinal study Arch Gerontol Geriatr. 124 2024 10545210.1016/j.archger.2024.105452 38728820

20Li Y. Zhang T. Li M. Shen R. Wang X. Zhuo C. Trajectories of intrinsic capacity decline and related factors in old persons: a 15-year community-based cohort study in Beijing J Nutr Health Aging. 29 2025 10052610.1016/j.jnha.2025.100526 PMC12180025 40048878

21Lin H.-R. Tsuji T. Kondo K. Imanaka Y. Development of a risk score for the prediction of incident dementia in older adults using a frailty index and health checkup data: the JAGES longitudinal study Prev Med. 112 2018 88 96 10.1016/j.ypmed.2018.04.004 29626553

22Kivipelto M. Mangialasche F. Ngandu T. Lifestyle interventions to prevent cognitive impairment, dementia and Alzheimer disease Nat Rev Neurol. 14 2018 653 666 10.1038/s41582-018-0070-3 30291317

23Livingston G. Huntley J. Sommerlad A. Ames D. Ballard C. Banerjee S. Dementia prevention, intervention, and care: 2020 report of the Lancet Commission Lancet 396 2020 413 446 10.1016/S0140-6736(20)30367-6 32738937 PMC7392084

24Sum G. Lau L.K. Jabbar K.A. Lun P. George P.P. Munro Y.L. The World Health Organization (WHO) Integrated Care for Older People (ICOPE) framework: a narrative review on its adoption worldwide and lessons learnt Int J Environ Res Public Health. 20 2022 154 10.3390/ijerph20010154 36612480 PMC9819593