Simulating clinical trials for model-informed precision dosing: using warfarin treatment as a use case.

Area of Science:

• Pharmacology
• Computational Biology
• Clinical Trials

Background:

• Patient treatment response varies significantly, necessitating individualized dosing for certain drugs like chemotherapies.
• Model-informed precision dosing (MIPD) offers a promising approach to tailor drug regimens and mitigate adverse events.
• Existing MIPD methods, including regression, reinforcement learning (RL), and pharmacokinetic/pharmacodynamic (PKPD) modeling, lack a unified framework for comparison.

Purpose of the Study:

• To develop and utilize a novel framework for simulating clinical MIPD trials.
• To evaluate and compare the efficacy of three distinct MIPD approaches: neural network regression, deep RL, and PKPD modeling.
• To assess the strengths and limitations of each MIPD method in achieving successful treatment individualization.

Main Methods:

• Development of a simulation framework emulating clinical complexities for MIPD trial testing.
• Application of the framework to warfarin dosing, a common anticoagulant with a narrow therapeutic index.
• Comparative analysis of neural network regression, deep RL, and PKPD modeling for predicting optimal dosing strategies.

Main Results:

• The PKPD model achieved the highest success rate (75.1%) in maintaining therapeutic INRs and the highest median time in therapeutic range (TTR) of 74%.
• Regression and deep RL models showed lower success rates (47.0% and 65.8%, respectively) and median TTRs (45% and 68%).
• PKPD and deep RL models demonstrated greater individualization by incorporating monitoring data beyond covariate-explained variability, with PKPD additionally refining predictive individualization.

Conclusions:

• PKPD modeling represents a highly successful and efficient MIPD approach for individualizing warfarin dosing.
• The developed simulation framework provides a valuable tool for cost-effectively evaluating and advancing MIPD strategies.
• Different MIPD approaches offer varying degrees of individualization, highlighting the importance of model selection based on therapeutic goals.