This methods paper overviewed the challenges in statistical analyses of clinical trials with continuous scale outcomes measures. The currently used statistical methods for this type of data were identified from clinical trials published between September 16, 2001 and September 15, 2002 in three respected journals: one from psychology, psychiatry, and medicine were reported. The strengths and limitations of the commonly used statistical models were examined. To address some problems that plague commonly used statistical methods for this type of data, an intrinsically nonlinear function was developed and implemented on a clinical trial dataset using nonlinear growth curve methodology. The results of the nonlinear growth curve model were compared to those of repeated measures ANOVA, the mixed model for repeated measures, and a polynomial linear growth curve model.
For a clinical trial that evaluated outcomes from pharmacological and behavioral interventions for treating chronic tension-type headaches, the nonlinear growth curve model provided a better fit to the data based on Schwarz’s Bayesian Information Criteria and Akaike’s Information Criteria, more reasonable subject-specific and interpolated predicted values, and more clinically meaningful coefficients than the competing models. The four coefficients for the nonlinear function represented the baseline symptom level, the amount of change, and two different aspects of the rate of change. Despite the increased complexity in estimation, this nonlinear growth curve model appears to be viable alternative for analyzing clinical trial data.