32379
The Utility of Omnibus Statistical Approaches in Go/No-Go Decision Making in a Phase 2 Study in Adolescents with Autism Spectrum Disorder: A Case Example
The informativeness of Phase 2 clinical trials in autism spectrum disorder (ASD) has been diminished by a lack of gold standard, syndrome-specific outcome measures that render clear, easy-to-interpret results. This shortcoming has undermined confident go/no-go decision-making when deciding whether a drug candidate is worthy of further clinical development, or not.
Objectives:
One potential remedy involves better statistical approaches that improve confidence in results analyses, especially when the size of the study is relatively small. In this example, two omnibus statistical approaches were applied to a recent set of ASD Phase 2 clinical trial results, to determine the likelihood of the pattern of affirming efficacy findings in the study being a reflection of false discovery (i.e. a false positive finding).
Methods:
The efficacy results were derived from a study involving AMO-02/tideglusib. This orally administered new chemical entity is a GSK-3β enzyme inhibitor with multiple converging lines of preclinical evidence that indicate that it may represent a potential therapeutic agent for individuals with ASDs.
In this study, 83 adolescents between 12 and 16 y.o. were blindly randomized 1:1 to placebo or AMO-02/tideglusib which was up-titrated from 400 mg to 1000 mg once-daily, weight-adjusted, during a 12-week active treatment period.
A step-wise omnibus approach was utilized, commencing with a concordant trend analysis that simultaneously considered mean results from 7 pre-specified outcome measures utilized in the study (including the ABC, VABS, RBS, OARS, and Parent Chief Complaint rating scales as well as a biomarker assessment that assessed pAkt levels in blood). The second step utilized permutation testing which involved 1000 computerized simulations in which study participants were randomly re-assigned to a treatment group, although the entire results data set of each subject was left intact.
Results:
The standard, pre-specified results analyses revealed that most measures trended for greater improvement with AMO-02/tideglusib than placebo including measures of social withdrawal (ABC-Social) and repetitive behaviors (RBS-R), as well as daily living skills (Vineland), memory (NEPSY) and sleep quality (CSHQ), with several measures reaching nominal significance.
The concordant trends analysis confirmed efficacy in 3 of the 7 outcome measures/endpoints involved in the analysis, with the remaining endpoints showing no worsening on AMO-02/tideglusib, and a pattern was evident in which AMO-02/tideglusib consistently outperformed placebo. The probability that this pattern occurred by chance alone was lower than 0.09; the exact value of the false-positive rate could be calculated using permutation testing. Two scenarios were run using the permutation testing approach and these revealed p-values of 0.07 and 0.007 in favor of AMO-02/tideglusib over placebo.
Conclusions:
An initial concordant trend analysis and subsequent permutation testing provided reiterative confidence about the affirming study findings that effectively established positive proof-of-concept for AMO-02/tideglusib as a potential treatment for ASD. Accordingly, it can be concluded that AMO-02/tideglusib merits further progression in clinical development in ASD.