OBJECTIVES Current understanding of the impact of continuous predictors of patient adherence, such as copayment, age, and number of refills prescribed with drug therapy is largely based on the results of parametric multivariate regression analyses. In such a regression, the copayment is usually introduced as a linear effect of continuous risk factors or as a multi- level categorical variable. To assess the potential impact of the a priori assumption of linearity or categorization on the results of parametric analyses, the authors completed an analysis using an assumption-free nonparametric modeling approach. METHODS The data set included 42,221 patients new to PPI therapy. The outcome variable was defined as the total number of 30-day refills that patients obtained during a 6-month follow-up period. The effects of copayment, age, and number of refills prescribed on adherence, adjusted for other common risk factors, were estimated using a smoothing spline method available in the generalized additive model extension of the Poisson regression. RESULTS The generalized additive model fits the data significantly (p <0.01) better than the parametric Poisson regression. The authors conclude that there are strong and practically important nonlinear relationships between adherence and analyzed continuous independent predictors. CONCLUSIONS The linear approximation or categorization of these effects inherent in parametric models may result in biased estimates of the effects of adherence.