Follow-up analyses evaluated a series of statistical interactions in the DTSA model to determine whether the linear additivity assumption had been violated. This assumption postulates that differences in the value of a predictor correspond to fixed differences in the hazard rate, such that a predictor’s effect does not depend upon the values of other predictors in the model. Following the procedure of Singer & Willett (2003), these analyses were restricted to examine interactions among the robust risk factors of AUDs (LR to alcohol, family history, baseline drinking, age of onset) as described previously. Each of the six possible interactions was tested by creating a cross-product term and adding it to a model that included both main effect covariates. This model was compared to a model excluding the cross-product using LRT to determine whether the inclusion of each interaction significantly improved model fit.
