Bayesian and Probit comparative analysis of risk factors for high blood pressure among adults in Kassena-Nankana Districts, Ghana: An AWI-Gen sub-study

Abstract Background: Hypertension prevalence is rising in rural Ghana, but risk factor identification is constrained by methodological limitations in sparse data. This study comparatively applied frequentist (Probit) and Bayesian models in identifying hypertension risk factors and evaluating consistency and uncertainty quantification. Methods: This was a cross-sectional analysis of 2,010 adults in the Kassena-Nankana districts. Ordered Probit and Bayesian models were used to assess associations between socio-demographic, behavioral and blood pressure variables. Model fit was compared via AIC/BIC. Results: The prevalence of hypertension in the study population was 21.9Model comparison showed the ordered Probit model had superior fit (AIC/BIC: 4644.8/4734.2) to the Bayesian model (AIC/BIC: 5137.5/5165.2). Both models consistently identified male sex (Probit β=0.309, p<0.001; Bayesian mean β=0.264, 95% CrI: 0.15–0.34), older age (55–60 years: Probit β=0.514, p<0.001; Bayesian β=0.499, 95% CrI: 0.41–0.58), and higher BMI (Probit β=0.052, p<0.001; Bayesian β=0.052, 95% CrI: 0.04–0.07) as significant risk factors. However, the Bayesian model showed greater uncertainty for marital status (β=0.272, 95% CrI: -0.08–0.56) and smoking (β=-0.260, 95% CrI: -0.29 – -0.01) compared to the Probit model. Both models revealed an inverse association of smoking with hypertension, while pesticide exposure showed conflicting directions and this warrants further investigation. Conclusion: Male sex, older age, and higher BMI were consistent predictors of hypertension. Inverse associations with smoking and conflicting effects of pesticide exposure warrant further investigation. These findings highlight the need for targeted, context-specific interventions in rural Ghana. While the Probit model demonstrated a better fit, the Bayesian approach provided deeper insight into uncertainty within sparse subgroups, supporting their complementary use in hypertension research, especially in resource-limited settings.