Optimal policy determination in sequential systemic and locoregional therapy of oropharyngeal squamous carcinomas: A patient-physician digital twin dyad with deep Q-learning for treatment selection.

摘要: Abstract Purpose Currently, selection of patients for sequential vs. concurrent chemotherapy/radiation regimens lacks evidentiary support, and it is based on locally-optimal decisions each step. We aim to optimize the multi-step treatment head neck cancer predict multiple patient survival toxicity outcomes, we develop, apply, evaluate a first application deep-Q-learning (DQL) simulation this problem. Patients methods The decision DQL digital twin patient’s were created, trained evaluated dataset 536 oropharyngeal squamous cell carcinoma (OPC) with goal of, respectively, determining optimal respect metrics, predicting outcomes patient. models subset 402 (split randomly) separate set 134 patients. Training evaluation dyad was completed in August 2020. includes 3-step complete relevant history cohort treated at MD Anderson Cancer Center between 2005 2013, radiomics analysis performed segmented primary tumor volumes. Results On validation set, 87.09% mean 90.85% median accuracy outcome prediction, matching clinicians’ improving (predicted) rate by +3.73% (95% CI: [-0.75%, +8.96%]), dysphagia +0.75% (CI: [-4.48%, +6.72%]) when following decisions. Conclusion Given prediction predicted improvement medically yielded approach, patient-physician dynamic problem has potential aiding physicians course assessing its outcomes.