A Proposal for Post Hoc Subgroup Analysis in Support of Regulatory Submission.

BACKGROUND: In clinical trials, it is not uncommon that the primary analysis fails to achieve the study objective for demonstrating the safety and efficacy of a test treatment under investigation, while a specific sub-population analysis shows a significant positive result. In this case, whether the observed positive sub-population analysis results can be used in support of regulatory submission of the test treatment under investigation is an interesting question to both the investigator(s) and the regulatory medical/statistical reviewers. METHODS: In this article, several statistical evaluations for confirming the integrity and validity of the observed sub-population analysis results were proposed in support of the regulatory submission. Selection bias caused by looking at one subgroup is adjusted before all statistical evaluations, including reproducibility, consistency between sub-population and the entire population, generalizability between the promising sub-population and other sub-populations, and sensitivity index when there are shifts in mean and/or variability between sub-populations. The multiplicity issue is also addressed in measuring generalizability. RESULTS: A numerical example of a global (multi-regional) clinical trial was presented for illustration purposes. The choice of applying which estimation approach relies on the scale of test statistics. Recommendations for incorporating statistical evaluations in measuring sub-population analysis are provided. Finally, we proposed possible solutions such as real-world data and real-world evidence for regulatory concerns, which may increase the insufficient power. CONCLUSION: Sub-population analysis can contribute to regulatory submission if it passes the evaluation. This analysis can also support hypothesis generation and the planning of future clinical trials, though it fails to pass the measurement process.

Duke Scholars

Author Shein-Chung Chow Biostatistics & Bioinformatics, Division of Biostatistics

Author Jiajun Liu