Investigating drug plasma levels and clinical response using random regression models.
The reanalysis of the Riesby dataset using a random regression model indicates a significant effect of DMI plasma measurements on HAM-D scores across the four timepoints of the study. This effect is especially strong when the HAM-D change from baseline score is used in place of the actual HAM-D score at the four timepoints. A significant effect of IMI was not found for either the actual HAM-D score or the HAM-D change from baseline score. An endogenous effect was marginally significant when the actual HAM-D score was used; however, this effect was not observed when the HAM-D change score was used as the dependent measure. There also was evidence of a marginally significant effect due to autocorrelation of the residuals, indicating that the residuals at a given timepoint were related to the residuals from previous timepoints according to a first-order autoregressive process. In contrast, when the repeated measures MANOVA was used to analyze these data, a significant effect of DMI was not observed, since subjects without complete data at all timepoints had to be dropped from the analysis. In longitudinal psychiatric studies where missing data are the rule (rather than the exception), the random regression approach provides an attractive alternative to the traditional methods of analyzing longitudinal data.