Analyzing Continuous Measures in HIV Prevention Research Using Semiparametric Regression and Parametric Regression Models: How to Use Data to Get the (Right) Answer?

Abstract

Semiparametric regression models based on smoothing splines were used to examine the associations between the risk of HIV seropositivity and a continuous covariate. For example, in the fully parametric logistic regression model, age was associated with a decreased risk of HIV seropositivity [Odds ratio (OR): 0.94 per 5 year increase, 95% Confidence Interval (CI): 0.90-0.98]. This association was not evident when the age was dichotomized at the median [i.e. >26 years vs. ≤26 years (reference)] (OR: 1.06, 95% CI: 0.92-1.21). Understanding the relationship between a continuous covariate and an outcome variable of interest involves determining the shape and strength of that relationship. The choice of the most appropriate approach depends on the specific problem and available data. We showed that using semiparametric regression techniques may be helpful in understanding the best way to do the categorization when it is desired.