Longitudinal trajectories of questionnaire-derived health burden and clinical correlates among community-dwelling older adults: a retrospective cohort study in Deyang, China

Community health-management records may help describe heterogeneity in older adults’ long-term health states, but repeated questionnaire-derived information is seldom evaluated alongside routine clinical measures. Objective: To identify longitudinal patterns of a standardized multidomain constitution index and describe their clinical correlates among community-dwelling older adults. Methods: We analyzed routine records from Deyang, China, collected from 22 February 2017 to 9 April 2024. After same-day deduplication, 49,472 records from 17,912 adults were available; 8,862 participants with at least three assessments and non-zero follow-up entered trajectory modeling. Linear and quadratic Gaussian finite-mixture models with two to six classes were compared using 100 independent starts. Baseline correlates were estimated with multinomial logistic regression, repeated objective measures with generalized estimating equations, and robustness with posterior pseudo-class draws, high-confidence analyses, inverse-probability weighting and a raw-score model. Results: The lowest-BIC acceptable solution was a five-class linear model: very low (7.2%), low (28.4%), intermediate (28.0%), moderately high (27.1%) and high (9.3%) index. Predicted values increased modestly within all groups over six years. Baseline-index quartiles reproduced 54.3% of modal class assignments and had an adjusted Rand index of 0.247. The highest-index group was associated with older age, higher BMI, more chronic disease categories, larger waist circumference, higher LDL-C and lower HDL-C, while blood-pressure findings were lower rather than uniformly adverse. Conclusions: The selected classes were primarily persistent severity-pattern groups with modest temporal heterogeneity. This descriptive index may complement conventional assessment but should not be interpreted as a validated global-health measure, diagnostic category or causal clinical phenotype. We analyzed routine community health-management records collected in Deyang, China, from 22 February 2017 to 9 April 2024. After same-day deduplication, 49,472 records from 17,912 adults aged 60 years or older were available; 8,862 participants with at least three assessments and non-zero follow-up entered trajectory modeling. The primary measure was an equally weighted index formed by standardizing and averaging eight questionnaire domains from the traditional Chinese medicine constitution assessment used in local health management. Linear and quadratic Gaussian finite-mixture models with two to six classes were compared using 100 independent starts. Baseline correlates were estimated with multinomial logistic regression, and repeated objective measures were analyzed using generalized estimating equations. Posterior pseudo-class draws, high-confidence analyses, inverse-probability weighting, and a raw-score model assessed robustness. Among 8,862 participants in the trajectory cohort, the lowest-BIC acceptable solution was a five-class linear model: very low (7.2%), low (28.4%), intermediate (28.0%), moderately high (27.1%) and high (9.3%) burden. Predicted values rose modestly within all groups over six years. Compared with the very-low group, high-burden membership was associated with older age (odds ratio [OR] 2.25 per 5 years, 95% confidence interval [CI] 2.01–2.52), higher body mass index (OR 1.39 per kg/m2, 95% CI 1.34–1.45), and more chronic disease categories (OR 1.95, 95% CI 1.56–2.45); men had lower odds of high-burden membership (OR 0.48, 95% CI 0.38–0.62). After adjustment, the high-burden group had greater waist circumference and LDL cholesterol and lower HDL cholesterol at baseline. Classification was robust in the raw-score sensitivity analysis (84.9% exact agreement; weighted kappa 0.939). In routine community health-management records, questionnaire-derived health burden was organized mainly into persistent severity strata with modest within-stratum change. The high-burden stratum was associated with less favourable waist and lipid profiles. These profiles may complement, but should not replace, conventional clinical assessment in community healthy-ageing services.