Objectives: This report explores potential moderators and mediators of the effects of risperidone on the main outcome measure. Demographic data (e.g., parent income), ratings on other instruments (e.g., ABC Hyperactivity), medication adherence, and biological data (e.g., leptin levels) were all explored in the analysis.
Methods: Potential moderators and mediators were entered (continuous form where possible) into the original ITT analysis. The MacArthur Foundation Network subgroup (Kraemer et al., 2001, 2002) guidelines for moderation and mediation were employed. As partial correction for numerous comparisons, significance was set at p = 0.01.
Results: Initial severity of ABC Irritability was the only significant moderator of response to risperidone (χ2 = 15.09, p = 0.0001). CeruloplasminRID, a measure of copper, was marginally significant (χ2 = 4.87, p = 0.027). Medication adherence, optimal dose, and weight gain (percent) were significant mediators of response (χ2=7.39, p = 0.0066; χ2=8.19, p = 0.0042; and χ2=19.34, p = 0.0001, respectively). Weight gain appeared to predict improvement better with placebo than active drug. A number of baseline variables (e.g., parent income and education, CGI Severity) were nonspecific predictors, having a significant main effect on outcome, but not a significant triple interaction with time and treatment. Percent change in serum 5’nucleotidase, a zinc marker, had an effect on outcome that was moderated by risperidone [i.e., significant triple interaction with time and treatment (χ2=17.35, p < 0.0001), but not correlated with treatment].
Conclusions: Overall, the paucity of moderators and mediators of risperidone response may be scientifically disappointing but is clinically encouraging; it indicates that risperidone is appropriate for a wide range of children with autism and disruptive behaviors. The relation of high initial severity to greater change is not surprising, and is undoubtedly at least partly regression to the mean, possibly enhanced by treatment. Medication adherence is commonly shown to mediate response to various treatments, and this finding emphasized its role in treatment efficacy. The mediation by dose, although statistically significant, may be an artifact of titration; it was consistent with a physician prescribing increasing doses of placebo in the face of unimproved behavior and no side effects. Perhaps the most important result was that weight gain predicted response in both groups, and seemed to be a somewhat better predictor for placebo than risperidone. This implies that it might be difficult to separate the benefits of weight gain from the benefit of risperidone. Drug moderation of the effect of 5’nucleotidase change was robust enough to survive severe Bonferroni correction and seems worthy of further study, which is required before conclusions are warranted.