Development and internal validation of a clinical prediction model for 90-day clinically significant medication-related harm in adults initiating oral anticancer therapy: A multicenter prospective cohort study.

Abstract

IntroductionOral anticancer therapy shifts substantial responsibility for medication administration, interaction management, toxicity recognition, and adherence to patients and outpatient teams. We aimed to quantify early clinically significant medication-related harm and develop an internally validated model to estimate individual 90-day risk at treatment initiation.MethodsA multicenter prospective inception cohort enrolled 610 adults initiating a new oral anticancer therapy across four tertiary oncology centres during 2024-2025. The primary outcome was first clinically significant medication-related harm within 90 days. Thirteen prespecified predictor parameters were entered into a Cox proportional-hazards model. Internal validation used 1000 bootstrap resamples with uniform coefficient shrinkage and baseline-survival recalibration. Discrimination, calibration, Brier score, internal-external centre validation, and decision-curve analysis were evaluated.ResultsDuring 45,048 person-days, 137 participants experienced harm. The Kaplan-Meier estimated 90-day risk was 22.9% (95% CI 19.7-26.5), and median time to first harm was 25 days (IQR 14-39); 60.6% of events occurred by day 30. Higher ECOG performance status, Charlson Comorbidity Index, organ-dysfunction severity, baseline clinically significant drug-drug interactions and medication-regimen complexity were associated with greater hazard, whereas higher health literacy was protective. The optimism-corrected C-index was 0.701, 90-day AUC 0.735 and Brier score 0.157. The shrinkage factor was 0.867. At the illustrative 20% risk threshold, sensitivity was 78.8%, specificity 60.8% and negative predictive value 90.6%.ConclusionsClinically significant medication-related harm was common and concentrated early after initiation of oral anticancer therapy. A model based on routinely available baseline clinical and medication factors provided moderate discrimination and useful risk stratification after internal validation. Independent external validation is required before clinical implementation.