Tutorial in biostatistics: Analyzing associations between total plasma homocysteine and B vitamins using optimal categorization and segmented regression

Data analysts consider standard regression models (e.g., generalized linear model) or nonparametric smoothing techniques (e.g., loess or splines) when examining the association between two variables. Before this step, a quantile-based summarization is typically used for exploring the exposure-response relationship. Unfortunately, these exploratory approaches may not be optimal or efficient for guiding the formal analysis in many biological and nutritional data settings. We suggest a recently developed method for selection of cutpoints as a tool of data summary and segmented regression as a modeling approach in the analysis of plasma total homocysteine and related vitamins. These methods are often complementary in discovering the underlying complex pattern of association.