We fitted two multivariate Cox regression models to the data (Table 1). Model I included the main effects of AGRS, medication condition, gender, the FTND score, and the interaction between AGRS and FTND. Model II included all predictors except the interaction term, thus serving as a test comparative model that assumes no higher-order effects. Submodels were compared by examining the difference in the (-2LL) estimates that are distributed as a Chi-square with degrees of freedom equivalent to the differences in degrees of freedom between the models. Models were also compared using their Akaike Information Criterion (AIC).25 The results indicated that model I (−2 LL) = 368.20 degrees of freedom (df) = 5, AIC = 378.20) provided a better fit to the data than model II (−2 LL = 373.97, df = 4, AIC = 381.99) (ie, Δχ2 = 5.79, Δdf = 1, P < 0.05; ΔAIC = 3.79); that is, the removal of the interaction term worsened the fit of the model. The results of the best fitting model (Model I) indicated that increasing levels of genetic risk (as measured
