Baseline characteristics influencing placebo response in clinical trials of treatments for fragile X syndrome
Penz, Craig Christopher
MetadataShow full item record
Fragile X Syndrome (FXS) is a disorder caused by a congenital mutation of the FMR1 gene on the X chromosome. FXS is associated with moderate to severe intellectual disability and is one known cause of autism spectrum disorders. There are no approved medications to treat FXS symptoms. In 2013, Seaside Therapeutics completed two Phase 3 studies of an investigational medication, STX209, for treatment of social withdrawal in FXS. Efficacy results for these studies were not positive. Clinical trials of psychoactive drugs often fail to show a statistical difference from placebo controls and a robust response to placebo is often cited as a reason for the failure. Retrospective studies of baseline variables in clinical trials have identified characteristics that were associated with an increased likelihood of responding to placebo. Such information is valuable for the design of future clinical trials and no such studies have been conducted in FXS. This study was a post-hoc analysis of data from Seaside Therapeutics' Phase 3 clinical trials in FXS. Baseline variables for subjects receiving placebo were pooled for analysis. To determine if a subject responded to placebo, the parent-rated ABC-SA and the ABC-IR were used. Clinician-rated assessments, including the CGI-S and CGI-I, were examined as well. Two-sample t-testing, one-way ANOVA testing, and correlation coefficients were calculated to compare the responses of subjects with different baseline characteristics. General linear regression modeling was used to determine if there were multiple baseline variables that could predict placebo response. Logistic regression modeling was used to determine if the baseline variables could predict whether a subject had a higher chance of being a treatment responder. A total of 287 subjects were randomized and completed the Phase 3 studies. Analyses for this study were conducted in a subgroup containing 106 subjects who received placebo. 76% improved during the study on the ABC-SA, indicating that there was a strong placebo effect on the study. None of the dichotomous baseline variables were associated with statistically significant differences in ABC-SA, ABC-IR, CGI-S, or CGI-I scores. Placebo-treated subjects in the 209FX302 study who were taking antipsychotics improved less on the CGI-S than those not on those medications. A similar pattern was observed on the ABC-IR and ABC-SA. Other categorical baseline variables were tested and there was no difference in the mean changes. The CGI-S score at baseline appeared to predict a statistically significant difference in the ABC-IR as more severe subjects were more likely to show a larger change in the ABC-IR. Similar, although not statistically significant results were seen with ABC-SA, CGI-I, and CGI-S changes, in that more severe subjects had greater responses to placebo. ABC-IR score changes were correlated independently with each of the ABC-C subscales but also with parental distress, CGI-S, and VAS-Anxiety. Only one variable, the ABC-IR at baseline, was significantly correlated with the ABC-SA score change, the rest of the variables were not significant. A multiple linear regression model predicting placebo response for the ABC-SA included only the baseline ABC-SA score. When the studies were modeled separately, the 209FX302 model contained additional variables including gender, antipsychotic use, and ABC-stereotypy scores. For the ABC-IR change model, the highest correlation coefficient was found in the 209FX301 study with ABC-IR, gender, Vineland-communication, maternal FMR1 status, and ABC-SL included in the model. 70% of the placebo treated subjects improved on the ABC-SA by at least 25%. Placebo responders were less frequently observed in clinician-rated assessments such as the CGI-I and CGI-S. In logistic regression models, for the ABC-IR response, a higher score on the hyperactivity subscale of the ABC-C was predictive of a lower placebo response. The CGI-S model was statistically significant and included the subject's age, race and ABC-IS score. The ABC-SA response could be modeled only in the 209FX302 study with gender and ADHD medication use remaining in the model. Also in the 209FX302 study, subjects were far less likely to be a responder on the ABC-IR or a total responder, if they were taking antipsychotic medications. Results of this study indicate that the ABC-SA is not recommended in future trials in the FXS patient population. Future trials should also allow ADHD and antipsychotic medication use as they were associated with a lower placebo response in some analyses. In addition, due to their inclusion in regression models, future studies should consider baseline variables such as parental stress and Vineland scores, and when designing study eligibility criteria or stratification variables.