Model-Informed Reinforcement Learning for Enabling Precision Dosing Via Adaptive Dosing.

Area of Science:

• Pharmacology
• Artificial Intelligence
• Computational Biology

Background:

• Precision dosing is crucial for drugs with narrow therapeutic windows and severe side effects.
• Adaptive dosing strategies are needed for time-varying treatments requiring sequential dose adjustments.

Purpose of the Study:

• To investigate coupling reinforcement learning (RL) with population pharmacokinetic/pharmacodynamic (PK/PD) models for precision dosing algorithms.
• To review existing literature and discuss case studies integrating PK/PD models within RL for adaptive dosing.

Main Methods:

• Review of relevant literature on RL and PK/PD models for precision dosing.
• Analysis of case studies (anesthesia, warfarin, anticancer treatments) using PK/PD models as RL simulation engines.
• Formulation of precision dosing problems within the RL framework (state, action, reward).

Main Results:

• Identified heterogeneity in precision dosing approaches, methodologies, and clinical integration.
• Demonstrated the potential of PK/PD models as simulation engines within RL for predicting dosing consequences.
• Highlighted the applicability in diverse clinical scenarios like anesthesia, anticoagulation, and oncology.

Conclusions:

• The integration of PK/PD models into RL frameworks holds significant promise for advancing precision dosing.
• Further research and development are necessary to overcome current limitations and expand applicability to adaptive dosing strategies.