Trials that prospectively accrue based on genetics are powerful methods to test hypotheses in precision oncology, but are often difficult to conduct due to the rarity of biomarker-positive cases. This is mitigated in umbrella trials that maximize accrual by testing many hypotheses. However, multiple treatment arms frequently generate conflicting treatment allocations due to simultaneous actionable mutations; the allocation algorithm should be planned based on relative mutation frequencies to accurately estimate sample size and dropout rates. Large sequencing projects like the TCGA could be exploited but proper tools are lacking.
We developed Precision Trial Designer (PTD), a bioinformatic tool that simulates genomically-defined cohorts by iteratively sampling patient data from mutation databases (e.g. TCGA) and provides essential parameters to plan biomarker-driven trials using survival or response-rate endpoints. To show its potential, we simulated a 10-arm imaginary trial on multiple cancers, based on the Molecular Analyses for Personalized medicine (MAP) consensus. We then validated our approach by comparing simulated and real data from the SHIVA01 clinical trial.
In the MAP trial, PTD predicted ≥1 actionable alteration in 73% patients and 32% conflicts. To adequately power each arm, we found the optimal rule that maximizes accrual (ALK inhibitors first, AKT inhibitors last) and propose various designs. In the SHIVA01 simulation, combinatorial point mutations were correctly predicted (18.9%, 95%CI 15.7-22.1 simulated vs 15.3%, 11.8-18.7 real), whereas PTD slightly overestimated copy number alterations (36.3%, 31.6-41 simulated vs 25.6%, 21.4-29.7 real), a predictable gap due to different detection techniques. Overall, 50.8% (46.2-55.7) cases were predicted as biomarker-positive, vs 37% (32.5-41.7) real. The relative contribution of each pathway of treatment (RAS/MEK, PI3K/MTOR) was conserved (47.8%, 41.7-53.9 and 25.8%, 19.9-31.5 simulated vs 53.1%, 95%CI 48.3-57.9 and 31.6%, 27.2-36.1 real, respectively).
PTD predicts combinatorial mutation frequencies with acceptable approximation and overcomes pressing issues in designing precision trials.
Clinical trial identification
Legal entity responsible for the study
This research was funded by an AIRC Investigator Grant to G Curigliano and by grant ANR-10-EQPX-03 from the Agence Nationale de le Recherche (Investissements d’avenir) and SiRIC (Site de Recherche Intégré contre le Cancer).
All authors have declared no conflicts of interest.