Joint blood pressure–body mass index trajectories and incident self-reported heart disease in middle-aged and older adults: a landmark analysis in HRS and ELSA

Background Single blood pressure (BP) or body mass index (BMI) measurements may not reflect cumulative cardiovascular exposure. Objectives To identify joint trajectories of systolic BP, diastolic BP, and BMI and examine their associations with incident self-reported heart disease. Methods We analysed 6,300 Health and Retirement Study (HRS) and 3,483 English Longitudinal Study of Ageing (ELSA) participants aged 50 years or older. Participants had at least two measurements across three assessments over approximately eight years and no self-reported angina, myocardial infarction/heart attack, or heart failure at the third assessment (landmark). Cohort-specific multi-trajectory models identified joint profiles. Cox models estimated associations with post-landmark incident composite heart disease. Results Seven-group solutions had the lowest BIC in both cohorts, but a five-group solution was retained as a stability-informed, parsimonious primary analysis. During follow-up, 430 events occurred in HRS and 152 in ELSA. In fully adjusted models, HRS participants with mid-BP/high-BMI (hazard ratio [HR] 1.63, 95% confidence interval [CI] 1.14–2.33) or high-BP/mid-BMI trajectories (HR 1.65, 95% CI 1.14–2.39) had higher risks than those with low-BP/low-BMI trajectories. In ELSA, estimates for the cohort-specific G4 (mid-BP/high-BMI) and G5 (mid-BP/very-high-BMI) groups were above 1.0 (G4: HR 1.80, 95% CI 0.96–3.35; G5: HR 1.95, 95% CI 0.98–3.91), but both confidence intervals included 1.0 and neither association reached statistical significance. Conclusions Joint BP - BMI trajectories identified clinically interpretable subgroups. In HRS, the mid-BP/high-BMI and high-BP/mid-BMI phenotypes were associated with increased risk of incident self-reported composite heart disease. In ELSA, estimates for the cohort-specific G4 and G5 groups were above 1.0, but their 95% confidence intervals included 1.0 and neither association reached statistical significance. Given that the trajectory classes differ across cohorts, the ELSA findings reflect an independent parallel analysis rather than a formal replication of HRS.