AI-driven subgroup discovery for hypertension control in multi-ethnic populations

Abstract Hypertension disproportionately affects ethnic minority populations in Europe, with prevalence rates up to five times higher than in the majority population. Despite adequate treatment and adherence, blood pressure (BP) control remains poor. Current clinical guidelines-primarily based on European populations-fail to reflect the biological, behavioral, and environmental diversity of minority groups. Artificial intelligence (AI) offers opportunities to integrate complex health data, uncover hidden non-linear patterns, and improve prediction of treatment outcomes. We conducted the first AI-driven subgroup discovery analysis in hypertension to identify patients achieving optimal BP control and key predictors of treatment success in diverse populations. We analyzed 1,580 hypertensive participants from the prospective HELIUS cohort (2015-2021) in Amsterdam, representing Dutch, South-Asian Surinamese, African Surinamese, Turkish, Moroccan, and Ghanaian backgrounds. Three complementary AI-driven algorithms-PRIM, SSD ++, and APRIORI-SD-were applied to identify subgroups with superior BP outcomes based on demographics, lifestyle, medication use, clinical and biological markers, chronic conditions, and treatment adherence. Subgroups were evaluated using support, coverage, accuracy, interpretability, and expert review. We identified 21 biologically and clinically meaningful subgroups, with SSD ++ demonstrating the best performance. Surprisingly, subgroups were primarily defined by antihypertensive medication type and baseline BP at treatment initiation, rather than by ethnicity, lifestyle, or comorbidities. A multi-ethnic subgroup characterized by lercanidipine use alone achieved 96% BP control. AI-driven subgroup discovery revealed that baseline BP and specific medication types are strongest predictors of BP control than ethnicity or lifestyle factors.