Reifenberg, J. | Search bioRxiv

bioRxiv2025

Model-informed Deep Q-Networks to Guide Infliximab Dosing in Pediatric Crohn's Disease

Model-informed precision dosing (MIPD) utilizes pharmacokinetic/pharmacodynamic (PK/PD) models to optimize drug therapy. However, conventional MIPD often requires manual simulation and regimen selection, which are time-consuming and demand specialized expertise. Reinforcement learning (RL), in which an agent learns optimal decisions through iterative interactions with an environment, offers a scalable and automated alternative. In this study, we developed a model-informed Deep Q-Network (DQN) to personalize infliximab dosing for patients with Crohns disease. The DQN was trained in a simulation environment incorporating a population PK model, inter-individual variability, and assay error. Virtual patients with randomly sampled covariates were used to explore dosing strategies at infusions 1, 3, and 4. Doses ranged from 1 to 10 mg/kg at infusion 1 and from 1 to 20 mg/kg thereafter, with intervals of 4 to 12 weeks. The reward function prioritized achieving trough concentrations of 18-26 {micro}g/mL before infusion 3 and 5-10 {micro}g/mL before infusions 4 and 5, while penalizing overtreatment and additional infusions. The DQN policy converged after 80,000 episodes, yielding target attainment probabilities (PTAs) of 92.9% and 98.4% at infusions 4 and 5, respectively, in 1,000 virtual patients. High doses (11-20 mg/kg) were selected in only 0.2% of cases. At infusion 4, 66.8% of patients received an 8-week interval, and 57.3% at infusion 5. Retrospective real-world validation showed that patients whose actual doses matched DQN recommendations had trough levels significantly closer to target ranges. These findings support the feasibility of using DQN-based agents to enhance and automate infliximab individualized dosing in pediatric populations.

pharmacology and toxicology↗

bioRxiv2025

Hybrid Population PK-Machine Learning Modeling to Predict Infliximab Pharmacokinetics in Pediatric and Young Adult Patients with Crohn's Disease

Population pharmacokinetic (PK) model-based Bayesian estimation is widely used for dose individualization, particularly when sample availability is limited. However, its predictive accuracy can be compromised by factors such as misspecified prior information, intra-patient variability, and uncertainties in PK variations. In this study, we developed a hybrid approach that combines machine learning (ML) with population PK-based Bayesian methods to improve the prediction of infliximab concentrations in children with Crohns disease. We calculated prediction errors between Bayesian-estimated and observed infliximab concentrations from 292 measurements across 93 patients. Incorporating clinical patient features, we explored various ML algorithms, including linear regression, random forest, support vector regression, neural networks, and XGBoost to correct the Bayesian-based prediction errors. The predictive performance of these ML models was assessed using root mean square error (RMSE) and mean prediction error (MPE) with 5-fold cross-validation. For Bayesian estimation alone, the RMSE and MPE were 4.8 {micro}g/mL and -0.67 {micro}g/mL, respectively. Among the ML algorithms, the XGBoost model demonstrated the best performance, achieving an RMSE of 3.78 {+/-} 0.85 {micro}g/mL and an MPE of -0.03 {+/-} 0.69 {micro}g/mL in 5-fold cross-validation. The ML-corrected Bayesian estimation significantly reduced the absolute prediction error compared to Bayesian estimation alone. This hybrid population PK-ML approach provides a promising framework for improving the predictive performance of Bayesian estimation, with the potential for continuous learning from new clinical data to enhance dose individualization. Key pointsO_LIA new hybrid model combining population pharmacokinetic model-based Bayesian estimation and machine learning significantly improved the accuracy of infliximab concentration predictions in young adult and pediatric patients with Crohns disease. C_LIO_LIThe developed hybrid model can facilitate infliximab individualized dosing by accounting for changes in clinical conditions and patient-specific factors that the conventional Bayesian estimation approach may not address, and can be integrated into precision dosing dashboards, such as RoadMAB, for real-world clinical application. C_LIO_LIThis study indicates that model predictive accuracy can be enhanced by combining the Bayesian method with machine learning, even with a relatively small amount of clinical data. This is particularly encouraging for specific populations, such as pediatric patients, where obtaining rich clinical data is challenging. C_LI

pharmacology and toxicology↗