Development and Validation of a Hypertension Prediction Model for Community-Dwelling Older Adults Using Routine Health Examination Data: Focus on Clinical Utility and Geriatric Relevance

Background Hypertension is a leading cause of cardiovascular morbidity and mortality in older adults, but routine screening tools tailored to community-dwelling seniors are limited. Older adults often have reduced screening adherence due to mobility issues, and hospital-based prediction models lack generalizability to primary care settings. This study aimed to develop and validate a hypertension prediction model for community-dwelling older adults using routine health examination data, with a focus on geriatric-specific clinical utility. Methods A cross-sectional study was conducted using data from 10,639 community-dwelling adults aged ≥ 60 years who underwent routine health examinations at the Community Health Service Center of Jiangling Sub-district (Suzhou, China) between January 2023 and December 2023. Missing data were handled via multiple imputation (mice package, R 4.3.0) under the assumption of missing at random (MAR), with 10 imputed datasets generated. The dataset was split into a training set (70%, n = 7,447) and test set (30%, n = 3,192) using stratified random sampling to maintain hypertension prevalence balance. Logistic regression (backward elimination, AIC criterion) and random forest models (1000 trees) were developed. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive/negative predictive values (PPV/NPV), calibration curves (Hosmer-Lemeshow test), and decision curve analysis (DCA). A sensitivity analysis compared model performance between imputed and complete-case datasets (n = 3,682). Results The mean age of participants was 68.2 ± 6.1 years (range: 60–98 years), and 57.38% (n = 6,100) had hypertension. The logistic regression model demonstrated moderate but clinically meaningful discriminative ability in the test set: AUC = 0.6382 (95% CI: 0.6214–0.6550), sensitivity = 0.682, specificity = 0.564, PPV = 0.663, and NPV = 0.585. Calibration was good (Hosmer-Lemeshow test: χ² = 12.36, P = 0.18), indicating agreement between predicted and observed hypertension probabilities. DCA showed the model provided a positive net benefit (0.08–0.12) across a geriatric-relevant threshold range (0.2–0.5): at a threshold of 0.3 (optimal for seniors aged 65–70 years), it identified 10 additional true cases per 100 participants compared to a “treat-all” strategy. Sensitivity analysis confirmed no significant bias from imputation (complete-case AUC = 0.621, 95% CI: 0.602–0.640). The top predictors were age (OR = 1.042, 95% CI: 1.038–1.046), high-density lipoprotein cholesterol (HDL-C; OR = 0.708, 95% CI: 0.663–0.756), and low-density lipoprotein cholesterol (LDL-C; OR = 1.169, 95% CI: 1.121–1.219). Conclusion This model uses routine health examination data to provide a cost-effective, interpretable tool for hypertension screening in community-dwelling older adults. Its moderate discriminative ability is offset by strong clinical utility (valid calibration, positive net benefit) and alignment with geriatric care needs. It can be integrated into community health services to improve screening efficiency and reduce hypertension-related burden in aging populations.