Modeling blood pressure as a continuous variable and the effect of dichotomization on its associated factors among adults in Uganda

Abstract

Modeling patient health indicators require attention to the probability distribution of the outcome measure as well as making use of all information from such measures. Disregarding these two estimation aspects is likely to compromise the precision in the identification of risk factors associated with modifiable diseases like hypertension. The main aim of this study was to model Blood Pressure (BP) as a continuous variable while accounting for its distribution, as gamma and examine the effect of discretization on its risk factors. Data were sourced from the nationwide Non-Communicable Diseases (NCDs) risk factor baseline survey of 2014 which followed the World Health Organization's STEPwise approach. Descriptive statistics and data visualization helped to examine the distribution of hypertension and descriptive characteristics of the adults were generated. This was followed by fitting a simple gamma regression model to assess the association between each covariate and BP, and in turn, identify variables for further analysis at the inferential analysis stage. At the inferential stage, there was a comparison between the gamma and binary logistic regression models to examine the effect of discretization and identify the most suitable model. The study established that the gamma regression model of the BP offered a better precision of identification of risk factors associated with it than the binary logistic regression model. Finally, a gamma model was fitted to examine the determinants of BP. It was found that adults in Uganda aged 45 years and above (Coef = 0.005, SE = 0.001), from Eastern (Coef = 0.023, SE = 0.007), Northern (Coef = 0.021, SE = 0.007), and Western (Coef = 0.019, SE = 0.008) regions with high fasting blood glucose levels (Coef = 0.014, SE = 0.003), and high body mass index (Coef = 0.008, SE = 0.001) were on average associated with higher levels of BP while female (Coef = -0.037, SE = 0.007), and married/cohabiting (Coef = -0.022, SE = 0.009) adults were negatively associated with BP. This study found that age, sex, marital status, fasting blood glucose, body mass index, and the region of residence as the major variables that best explain variations in BP (p < 0.05). The findings proved that attention to distributional analysis and the use of the gamma regression model rather than the binary logistic regression model improves the precision of identification of risk factors associated with BP. Fasting blood glucose, and BMI were independent modifiable risk factors for BP. A future study on the potential determinants of fasting blood glucose/Diabetes on a continuous scale is recommended.