Validation of latent health behaviour subtypes in young and middle-aged adults with hypertension: a machine learning-augmented LCA approach with external cohort replication

AIMS: To validate behavioural subtypes among young and middle-aged hypertensive patients using latent class analysis (LCA) and assess their generalizability across populations through external validation. METHODS AND RESULTS: We applied LCA to a derivation cohort of 2000 adults with hypertension to classify nine health behaviour indicators. Model robustness was examined with bootstrap stability testing. External validation was conducted in NHANES 2009-2010 (n = 541) using confirmatory LCA. A reproducible three-class solution emerged: Active (32.6%), Selective (28.1%), and Passive (39.3%). Demographic and clinical factors-including age, sex, education, insurance type, adverse medication symptoms, hypertension severity, time since diagnosis, number of medications, and comorbidities-were associated with class membership. Notably, the Passive group showed higher medication adherence (mean 46.4% ± 10.7%), whereas the Selective group had a greater risk of intermittent discontinuation of antihypertensive therapy. CONCLUSION: These findings support the development of individualized intervention strategies targeting health behaviour in young and middle-aged hypertensive patients, providing a scalable framework for stratified hypertension management in diverse populations.