Joint regression modeling of blood pressure and associated factors among adults in Uganda: Implications for clinical practice

Abstract

Blood pressure (BP) is a repeated measurement data as multiple measurements of both systolic blood pressure (SBP) and diastolic blood pressure (DBP) are simultaneously obtained on a patient to determine a raised blood pressure (hypertension). In examining factors associated with hypertension, BP is measured either as a binary outcome leading to information loss and reduced statistical efficiency or as a continuous outcome based on the average of one of the measurements or a combination of the two but independently thus ignoring possible correlation. We simultaneously modeled the risk factors for increased SBP and DBP among adults in Uganda and tested the difference in the effect of certain determinants on SBP versus DBP. We analyzed the 2014 nationwide non-communicable disease risk factor baseline survey data of Ugandans aged 18–69 years. We considered SBP and DBP as two continuous outcomes and conducted multivariate linear regression to jointly model SBP and DBP accounting for their distribution as bivariate normal. Of 3,646 participants, 950 (26.1%) had hypertension based on SBP (BP ≥ 140 mmHg) and DBP (BP ≥ 90 mmHg), 631 (17.3%) based on SBP alone, and 780 (21.4%) based on DBP alone. The study found that an increase in age (ranging from 18–69 years), obesity, income, being centrally obese, and hypercholesterolemia were significantly associated with higher SBP levels. Living in eastern, northern, and western Uganda regions was significantly associated with lower SBP, whereas increasing age, obesity, and hypercholesterolemia were significantly associated with higher DBP. Adults who rarely added salt to their meals were on average associated with higher DBP levels than those who never added salt to their meals. We found a strong residual correlation between SBP and DBP (r = 0.7307) even after accounting for covariates at the marginal level. This study presents a statistical technique for joint modeling of blood pressure, enabling the estimation of correlation between two outcomes and controlling family-wise error rate by testing the effect of a risk factor across both outcomes simultaneously.