Towards Advanced Therapeutic Drug Monitoring: A Safety and Performance Evaluation of a Clinical Decision Support System for Imatinib (Preprint)

UNSTRUCTURED Background: Therapeutic Drug Monitoring (TDM) aims to optimize clinical outcomes but is hindered by logistical delays and susceptibility to data errors. Integrating Model-Informed Precision Dosing (MIPD) with Point-of-Care testing, fortified by rule-based validation and alerts, forms a transformative Clinical Decision Support System (CDSS) essential for securing this complex workflow. Methods: This in silico study rigorously evaluated a rule-based CDSS for imatinib against a human TDM expert. We simulated a set of 60 virtual patients using a previously developed population pharmacokinetic model, and deliberately enriched with data quality errors (e.g., timing inaccuracies, reporting errors) to stress-test system resilience. We compared CDSS and TDM expert performances based on error detection capabilities and clinical impact on therapeutic target attainment and safety. Results: The CDSS and the expert demonstrated comparable error detection capabilities. Although the CDSS showed numerically higher sensitivity (53% CI95% [35–70] vs 45% CI95% [30–61]) and lower specificity (83% CI95 [63–95] vs 100% CI95 [81–100]). Clinically, the CDSS significantly outperformed the expert in target attainment (55% vs 37%; p=0.046). However, while the expert generated no detrimental interventions, the CDSS resulted in four detrimental outcomes, where patients initially within the target range were shifted outside it. Conclusion: The CDSS demonstrated superior efficacy in optimizing imatinib exposure while serving as a robust digital gatekeeper. While active optimization drives higher efficacy, a few observed detrimental cases highlight that algorithmic safety remains contingent on data integrity. This proof-of-concept study validates the CDSS architecture as a prerequisite for future prospective clinical trials.