Reinforcement learning and Bayesian data assimilation for model-informed precision dosing in oncology.

Area of Science:

• Pharmacology and Computational Biology
• Oncology
• Biostatistics

Background:

• Model-informed precision dosing (MIPD) aims to optimize drug efficacy and safety through therapeutic drug/biomarker monitoring.
• Current MIPD strategies, like dosing tables or maximum a posteriori estimates, have limitations in quantifying uncertainty and utilizing all patient-specific data.

Purpose of the Study:

• To introduce novel MIPD approaches using Bayesian data assimilation (DA) and/or reinforcement learning (RL).
• To control neutropenia, a critical dose-limiting side effect in anticancer chemotherapy, thereby improving treatment outcomes.

Main Methods:

• Development and application of three novel MIPD strategies integrating Bayesian DA and/or RL.
• Focus on controlling neutropenia incidence, specifically reducing severe (grade 4) and subtherapeutic (grade 0) events.

Main Results:

• The proposed DA and/or RL approaches demonstrate potential for substantial reduction in neutropenia incidence compared to existing methods.
• Reinforcement learning (RL) effectively identifies patient-specific factors influencing dosing decisions.

Conclusions:

• Novel Bayesian DA and RL methods offer advanced MIPD for improved neutropenia control in chemotherapy.
• The flexible DA-RL approach facilitates integration of multiple endpoints for future personalized cancer therapies.