Mid-life antihypertensive initiation and incident dementia: a machine learning-weighted target trial emulation in the health and retirement study.

Abstract

Background: Mid-life hypertension is a treatable vascular risk for dementia, yet trial evidence on whether starting antihypertensives in middle age lowers late-life dementia remains mixed. This study aimed to emulate a randomized trial to estimate whether initiating antihypertensive therapy at mid-life reduces subsequent dementia risk.

Methods: We emulated a randomized target trial of antihypertensive initiation using Health and Retirement Study data (2002-2020; 18-year follow-up). Eligible adults aged 45-65 with incident hypertension and dementia-free at baseline (n = 2,491) were classified in 2002 as antihypertensive treatment initiators or non-initiators. Inverse probability weights combined (I) a Firth-penalized logistic model for treatment initiation and (II) biennial censoring models; weighted Cox regression estimated hazard ratios (HRs) for incident dementia. To probe model dependence, we re-estimated treatment weights with a Super Learner ensemble and repeated all analyses.

Results: Groups differed on several baseline covariates, but weighting brought every standardized mean difference to <0.10. During follow-up, cumulative dementia incidence reached 15% in non-initiators and 13% in initiators. The intention-to-treat risk ratio (RR) was 0.89 (95% confidence interval [CI], 0.57-1.62); biennial RRs were likewise close to the null. Re-estimating treatment weights with a Super Learner ensemble yielded comparable covariate balance and a year-18 RR of 1.05 (0.74-1.70), essentially reproducing the primary result. Analyses accounting for competing risks of death or dropout did not materially change the estimates.

Conclusions: In this nationally representative cohort, starting antihypertensive therapy in mid-life did not meaningfully change dementia risk across 18 years. Agreement between conventional and machine-learning propensity approaches makes a large protective or harmful effect unlikely and underscores the need to explore additional strategies for dementia prevention beyond routine blood pressure pharmacotherapy.