nonadherence

nonadherence predicts low levels of 6-thioguanine (6-TG) and 6-methylmercaptopurine (6-MMP) and detects thiopurine nonadherence in pediatric inflammatory bowel disease (IBD) patients using a modified random forest model.

Key Features:

• Predictive Modeling: Employs a modified random forest algorithm with cross-validation and resampling to predict low 6-TG, low 6-MMP, and nonadherence in patients treated with AZA.
• Model Performance: Low 6-TG prediction: AUC 0.87, sensitivity 81%, specificity 80%; nonadherence prediction: AUC 0.94, sensitivity 82%, specificity 86%.
• Clinical Definitions: Low AZA dosing defined as 6-TG <125 pmol/8×10^8 erythrocytes and 6-MMP <5700 pmol/8×10^8 erythrocytes; nonadherence characterized by undetectable levels of both metabolites, specifically <240 pmol/8×10^8 erythrocytes.
• Data Utilization: Developed and evaluated on 332 observations from 88 pediatric IBD patients split into training and testing datasets.

Scientific Applications:

• AZA dosing optimization: Predicts metabolite-based low dosing to inform azathioprine (AZA) dose adjustments in pediatric IBD.
• Nonadherence detection: Identifies patients with undetectable thiopurine metabolites to flag potential nonadherence.
• Therapeutic monitoring: Uses routine thiopurine metabolite measurements (6-TG, 6-MMP) to support metabolite-guided clinical decisions in pediatric IBD.

Methodology:

Uses a modified random forest algorithm with cross-validation and resampling; models were trained and tested on a dataset of 332 observations from 88 pediatric IBD patients, and the 6-TG prediction yielded an RMSE of 110 on the testing dataset.