A simulated trial with reinforcement learning for the efficacy of Irinotecan and Ifosfamide versus Topotecan in relapsed, extensive stage small cell lung cancer

Objectives Synthetic data may proxy clinical data. At the absence of direct clinical data, this study aimed to compare Irinotecan and Ifosfamide (II) with Topotecan in synthetic, recurrent small cell lung cancer (SCLC) patients within a simulated clinical trial. Materials and methods Two simulation stages were conducted. Initially, 200 recurrent SCLC cases were simulated to replicate a previous phase 3 trial, testing the utility of Cox proportional hazards model and simulation methodology together, where patients were randomized to receive Cyclophosphamide, Adriamycin, Vincristine (CAV) or Topotecan. In the second stage, 600 recurrent SCLC patients were simulated and randomized to compare Topotecan versus II in terms of overall survival (OAS), using Reinforcement Learning (RL) and Cox proportional hazards model. Results CAV versus Topotecan comparison showed no statistical difference in overall survival (hazard ratio (HR): 0.89, 95% CI: 0.67-1.18, P = 0.418), aligning with the original clinical trial. For the Topotecan versus II comparison, the RL framework significantly favored the II arm (mean reward points: 193.43 versus - 251.82, permutation P < 0.0001). Likewise, II arm exhibited superior median OAS compared to Topotecan arm (11.12 versus 6.30 months). HR was 0.44 (95% CI: 0.38-0.52) with P < 0.0001, in favor of II. Conclusion Artificial trial results for CAV versus Topotecan matched the original trial, confirming indifference of OAS. Additionally, II yielded superior overall survival compared to Topotecan in recurrent SCLC patients. These demonstrate the potential of RL and simulation in conjunction with Cox modelling for similar studies. However, definitive conclusions necessitate a randomized clinical trial between II and Topotecan in this patient cohort.