Abstract
BACKGROUND: Hypertension drives pre-mature mortality globally, yet its precursor, pre-hypertension (pre-HTN), remains under-researched in Bangladesh despite offering a critical window for intervention. This study examined the socio-demographic and health-related determinants of pre-hypertension among Bangladeshi adults and evaluated whether machine learning models provide additional predictive value beyond conventional approaches, in support of Sustainable Development Goal (SDG) 3.4. METHODS: Pre-HTN was defined according to JNC 7 guidelines. Using the recent nationally representative sample from BDHS 2022, we integrated survey-weighted bivariate analyses and multivariable logistic regression to identify independent associations. A random forest classification model was implemented to assess variable importance and predictive performance. Model evaluation was conducted using confusion matrices and receiver operating characteristic curves. RESULTS: The study found that age and BMI were the most dominant predictors of pre-HTN, with the condition significantly more prevalent among the elderly and overweight individuals. Crucially, the crude association between diabetes and pre-HTN disappeared after multivariable adjustment. A "reversal" of the social gradient was observed, as higher education significantly increased risk (AOR = 1.26). Adults aged ≥60 years had nearly threefold higher odds compared with those aged < 30 years (AOR = 2.75; 95% CI: 2.32-3.26). Regionally, residents of the coastal region exhibited a significantly higher prevalence of elevated blood pressure (AOR: 1.19; 95% CI: 1.04-1.36) compared to the central region. The random forest model ranked age, body mass index, and sex as the most influential predictors. Discriminatory performance was modest and similar between models (AUC: logistic regression = 0.63; random forest = 0.62). CONCLUSION: Pre-hypertension in Bangladesh is shaped largely by aging, rising body weight, and shifting lifestyle patterns among the highly educated. Both logistic regression and random forest models pointed to the same core predictors, underscoring the stability of these findings. These results highlight the need for focused prevention strategies-supporting weight control in older adults, incorporating movement into the routines of sedentary educated populations, and tailoring programs to regions where risk is consistently higher.