High intrinsic capacity

Proportion

Intrinsic capacity is an important outcome indicator of healthy ageing. Monitoring intrinsic capacity provides critical insights into the aging process and informs effectiveness of public health interventions implemented to promote healthy aging and well-being.

Intrinsic capacity comprises all the mental and physical capacities that a person can draw on and includes locomotor capacity, cognitive capacity, psychological capacity, vison capacity, hearing capacity and vitality capacity. "Cognitive capacity" includes attributes of mental functions such as memory, attention, executive function, perception, orientation, and language abilities. "Locomotor capacity" includes a broad set of attributes of the musculoskeletal system that encompasses endurance, balance, muscle strength, muscle function, muscle power and a joint function of the body. "Vison capacity" is the clarity and sharpness of vision, crucial for daily functioning and independence. "Hearing capacity" is the ability to perceive sound, which is critical for communication and social interaction. "Psychological capacity" includes broader attributes such as self-efficacy, resilience, hope, optimism, mood, sense of coherence, autonomy, resourcefulness, identity, hope, religiosity/spirituality and life valuation. "Vitality capacity" reflects the physiological state of energy and metabolism.

Age (five-year age band), sex, income level, education level, place of residence (administrative region, e.g. cities, towns, semi-dense areas, and rural areas), setting ( residential care facility, at home in the community), disability status, nationally relevant population groups

This indicator is measured using data from nationally representative population surveys of older people that assess intrinsic capacity domains using standardized protocols. Intrinsic capacity is operationalised as a latent construct measured by 26 indicators spanning five subdomains in the analytic dataset: vitality (Forced Expiratory Volume in one second, handgrip strength, and additional biomarkers where available, including insulin-like growth factor 1, dehydroepiandrosterone sulfate, and haemoglobin), psychological capacity (four symptom items and a reverse-coded mood item), sensory capacity (two self-reported items plus objective distance and near vision scores and hearing measures), locomotor capacity (chair-stand, balance, and Timed Up-and-Go capacity items), and cognitive capacity (memory, naming, semantic memory, verbal fluency, orientation, and immediate and delayed recall measures). All items are oriented so that higher values reflect higher capacity. To enable estimation with polytomous item response models, selected continuous indicators are discretised into ordered categories using prespecified cut-points or empirical quantiles (for example, handgrip strength categorised into quartiles; distance vision, near vision, and hearing scores categorised using 0–25–50–75–100 thresholds; verbal fluency categorised using cut-points at 5 and 15), while binary indicators are retained as dichotomous. A bifactor multidimensional item response theory model is fitted using the mirt package in the R statistical computing environment, specifying one general intrinsic capacity factor loading on all 26 items and five subdomain-specific group factors loading on their respective item sets; graded response models are used for ordered items and two-parameter logistic models for dichotomous items. Model parameters are estimated using the Metropolis–Hastings Robbins–Monro algorithm. Person-level latent scores are derived using expected a posteriori scoring with quasi–Monte Carlo integration, yielding a general intrinsic capacity score and subdomain scores. For reporting, scores are rescaled to a 1–100 metric using a min–max transformation within the analytic sample and merged back to the respondent dataset using the unique identifier. High intrinsic capacity is defined using an outcome-anchored cut-point derived from external validation against care dependence. The five rescaled intrinsic capacity subdomain scores (locomotor, vitality, psychological, cognitive, and sensory; each ranging from 0 to 100) are entered simultaneously as predictors in a weighted logistic regression model with care dependence as the binary outcome, implemented as a binomial generalised linear model with the final survey weights applied as frequency weights. Model-based predicted probabilities of care dependence are generated for respondents with complete data on the predictors and are used to construct a weighted receiver operating characteristic curve and to compute a weighted area under the curve. The optimal probability threshold is selected by maximising Youden’s index (sensitivity minus one minus specificity) across receiver operating characteristic thresholds, and the sensitivity and specificity at the selected threshold are reported. Respondents are classified as having high intrinsic capacity if their predicted probability of care dependence is below the selected receiver operating characteristic threshold, thereby anchoring the “high” category to a clinically meaningful outcome rather than to sample-dependent percentiles. For interpretability on the intrinsic capacity metric, the selected probability threshold is also mapped to an approximate cut-point on the 0–100 intrinsic capacity scale using the fitted logistic regression equation; because the multivariable model includes separate subdomain predictors, this mapping requires an explicit rule for the subdomain profile, and the implemented approach approximates the cut-point by setting all subdomain scores to a common value across the 0–100 range and identifying the value at which the predicted probability equals the receiver operating characteristic-derived threshold. Because this is a Tier III indicator and global standards are evolving, these definitions require formal testing and comparison across settings, ideally against external outcomes including care dependence, functional ability, falls, hospitalisation, long-term care use, and mortality, and across population subgroups to identify the approach that best balances feasibility, reliability, measurement equivalence, comparability, and predictive validity. Alternative operational definitions are evaluated alongside the latent-score approach. First, each intrinsic capacity domain is classified as “no clinically significant decline” versus “decline” using prespecified clinical cut-offs for the relevant domain measures, and high intrinsic capacity is defined as meeting the “no decline” criterion in all domains, or under a less stringent rule, in at least a prespecified minimum number of domains. Second, domain measures are standardised to a 0–100 scale and combined to form a composite intrinsic capacity score, calculated as the mean of available domain scores with equal weights unless an alternative weighting scheme is prespecified and empirically validated, and high intrinsic capacity is defined using an absolute composite threshold (for example ≥80) together with a domain floor rule to avoid classifying respondents as high when one domain indicates marked impairment.

amuthavallithiya@who.int

AMUTHAVALLI THIYAGARAJAN, Jotheeswaran

Float

10244

This indicator depends on the availability, harmonisation, and standardisation of measures across intrinsic capacity domains, and estimates can vary when different instruments, categorisation rules, clinical cut-points, weighting schemes, or scoring approaches are used. In the latent-score approach, results are also influenced by modelling choices, including the selection of items, the grouping of domains, the treatment of continuous measures as ordered categories, and the assumptions of the bifactor multidimensional item response theory model. Objective performance measures can be affected by testing conditions, interviewer practice, equipment, and temporary health states, whereas self-reported measures are subject to recall error, reporting preferences, and cultural norms. Missing data across domains can introduce bias if handled inconsistently, and analyses that require complete or near-complete data may systematically exclude older people with poorer health or greater impairment. In addition, the definition of "high" intrinsic capacity may remain partly sample-dependent when scores are rescaled within the analytic sample or when thresholds are derived empirically from the same dataset. The outcome-anchored threshold is derived against care dependence and therefore reflects the predictive relation between intrinsic capacity and that specific external outcome in the study population; its transportability to other populations, settings, and outcomes may be limited. The probability threshold selected using receiver operating characteristic analysis depends on the observed outcome distribution and on the balance chosen between sensitivity and specificity, and the subsequent mapping of that threshold back to a single 0–100 intrinsic capacity cut-point requires simplifying assumptions about the profile of subdomain scores. Alternative definitions based on domain cut-offs or composite scores may be easier to implement, but they can also yield different prevalence estimates and classification patterns. Because this is a Tier III indicator and global standards are still evolving, all proposed definitions require further validation and comparison across countries, subgroups, and longitudinal studies. Finally, when measured cross-sectionally, the indicator provides only a snapshot of intrinsic capacity and does not capture trajectories of decline, recovery, or resilience over time.