Modeling Hypertension Risk Factors: Continuous Versus Binary Blood Pressure Scales

Abstract

Background Risk factors for hypertension are commonly identified with blood pressure measured and modeled on a binary rather than continuous scale leading to information loss and reduced statistical efficiency.

Methods We analyzed data from a nationally representative survey of adult Ugandans (18-69 years), focusing on non-communicable disease (NCD) risk factors. We report the accuracy of modeling risk factors for hypertension with blood pressure measured on a binary scale using binary logistic regression versus a continuous scale using gamma regression. We built a binary logistic regression model and two gamma regression models: one for the average systolic blood pressure (SBP) and another for the average diastolic blood pressure (DBP). Model accuracy was compared using Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC).

Results Of the 3,987 participants, the median age was 33 years (interquartile range: 25-45). Gamma regression model for the average SBP identified age, sex, region, marital status, income, and body mass index (BMI) as risk factors for hypertension (deviance = 212,196; AIC = 32,777.70; BIC = 179,331). Gamma regression model for the average DBP showed age, region, marital status, and BMI as risk factors for hypertension (deviance = 230,745; AIC = 35,438.24; BIC = 197,880). The binary logistic regression model indicated age, marital status, and BMI as risk factors for hypertension (deviance = 14,973,578; AIC = 37,556.11; BIC = 14,940,713).

Conclusion Risk factors for hypertension are more accurately modeled when blood pressure is on a continuous than binary scale. These differences have important public health and clinical significance in treating hypertension. We recommend using gamma regression for modeling the risk factors for hypertension.