PI-033 - MULTISTATE MODEL TO INVESTIGATE OVERALL SURVIVAL (OS) FROM COPANLISIB PIVOTAL TRIALS.

J. Moss¹, J. Paget¹, J. Veasy¹, B. Childs², P. Morcos², D. Garmann³; ¹BAST Inc. Limited, Leicester, UK, ²Bayer HealthCare Pharmaceuticals, Inc., Whippany, NJ, USA, ³Bayer HealthCare Pharmaceuticals, Inc., Wuppertal, Germany.

Background: Copanlisib is a pan-class I phosphatidylinositol 3‐kinase (PI3K) inhibitor with sub-nanomolar potency against ‐α and ‐δ isoforms. Pivotal studies investigated copanlisib in relapsed/refractory follicular lymphoma as a single agent (CHRONOS-1 [NCT01660451]) supporting its accelerated approval and in combination with rituximab (CHRONOS-3 [NCT02367040]) or chemoimmunotherapy (CHRONOS-4 [NCT02626455]). The confirmatory trial, CHRONOS-4, failed its primary endpoint resulting in US FDA approval withdrawal and termination of further copanlisib development. Further, many PI3K inhibitors have shown a detrimental OS effect [FDA ODAC 2022]. We investigated OS using pharmacometrics.

Methods: The 3 CHRONOS studies were pooled, and a dataset curated to describe patient disposition from study start to death. A semi-Markov multistate model described transitions between 6 states; modeled as 13 independent probability density functions. Exponential, Weibull, Gompertz and cubic spline models were fitted to each transition. Models were selected based on AIC and on visual inspection of simulated and observed number of cumulative transitions. Covariates were investigated on relevant transition(s). Visual predictive checks evaluated model performance. Separately, parametric time-to-event (TTE) analyses investigated serious adverse events (SAEs) and Grade ≥3 AEs.

Results: 1109 patients receiving copanlisib (707) or placebo (402) were included in the analyses. The multistate model adequately described patients in each state, transitions, and OS across studies and treatment (Figure). Discontinuation (d/c) due to AE to death showed an approximate eightfold higher hazard within the first 50 days compared with other transitions to death. Covariate analysis for time from start to d/c due to AE, d/c due to AE to death, SAEs, Grade ≥3 AEs identified several covariates including copanlisib treatment, however no specific copanlisib subset was at higher risk compared to remaining copanlisib treated patients.

Conclusion: Multistate modeling characterized hazards to death in copanlisib pivotal trials. Copanlisib treatment was identified as a covariate on some transitions, however no specific copanlisib treated subset was identified at higher risk which could be excluded from therapy.