Randomized Matched-Control Design for Disease Modification Drugs
*Qing Liu, QRMedSci, LLC 

Keywords: Biomarkers, Data-mining, Double-randomized delayed start design, False discovery rate, Mediation analysis, Randomized matched control design, Randomized registry trial

Development of disease modification drugs is extremely challenging for neurodegenerative diseases. Often it is necessary to show clinical evidence for disease medication via a randomized delayed-start design, especially when there are no established biomarkers for use to directly measure the biological or physiological progression of the disease. There has been a broad regulatory support for this design in areas of Alzheimer’s disease and Parkinson disease. Following this design, a trial would consist of two periods where in the first period patients are randomized to a new drug or a placebo, and in second period, placebo patients would switch to treatment with the new drug. The objectives of the trial are to establish that the new drug slows down the disease progression in the first period and that patients who have delayed-start with the new drug would not achieve the level of clinical response compared to those who started with the new drug early. While this design is conceptually sound, it is extremely difficult to implement due to high dropout rates. In a recent advisory committee meeting on a Parkinson’s disease drug, it was concluded that the second period is essentially an observational study and any analysis intended for establishing disease modification is not interpretable.

To resolve this problem, we propose a randomized matched-control design. In addition to the two periods of the randomized delayed-start design, the new design also includes a prospective run-in period where both static and treatment related outcomes are used to classify patients into matched-cohorts. Upon completion of the delayed-start period, a matched-cohort analysis is performed to establish design modification. The matched cohorts are defined with a data-mining methodology involving establishing frequent prognostic sets and their associations with clinical outcomes for which false discovery rates are controlled. The matched-cohort analysis is a part of the clinical efficacy analysis that controls multiple type 1 error rates. The proposed design is best implemented as a randomized registry trial for which the run-in period and frequent prognostic sets are already established.