Background: In a typical clinical trial treatment effects will not be expected to be the same on all of the study participants. As a result, investigators are often tempted to look at the effects of a given treatment in subgroups of patients in order to determine who will benefit the most or the least, especially when the treatment effect in the total sample is insignificant or borderline. This paper aims at demonstrating the application of random effect models as one approach to identify subpopulations suitable for subgroup analysis in a longitudinal study.
Methods: Data collected from a double-blind randomized controlled trial were used to demonstrate how multilevel modeling using random effects can be used to identify subgroups of postmenopausal women who benefit the most from nonhormonal treatment (gabapentin) of their hot flashes.
Results: We estimated subject-specific treatment effects and correlated these effects with patient characteristics at baseline. We found that women with a higher severity of hot flashes score at baseline were more likely to have the greatest reduction in hot flashes score from the treatment. Also, women who had a serum creatinine level higher than the median level at baseline demonstrated a greater response to gabapentin compared to the placebo group.
Conclusion: Our proposed method can help researchers identify patient factors that are associated with differential effect. Those factors are potential areas for further clinical investigation or for constructing subgroups for sub-analysis.