Machine learning model predicted FOLFIRINOX dose modifications in pancreatic ductal adenocarcinoma cohort

Background: FOLFIRINOX is a cornerstone regimen for eligible patients with pancreatic ductal adenocarcinoma (PDAC), but its clinical benefit is limited by substantial toxicity and frequent dose modification. In real-world practice, dose modifications are often individualized, and the clinical factors associated with these decisions remain incompletely characterized. Objective: To develop and evaluate an electronic medical record (EMR)-based machine-learning framework for modeling cycle-specific FOLFIRINOX dose modification decisions in patients with PDAC. Methods: We included patients with PDAC who received FOLFIRINOX at UCSF oncology clinics between November 2011 and December 2023. Predictors included demographic, clinical, laboratory, and treatment variables derived from the EMR. Logistic regression, random forest, and XGBoost models were trained using group-based 5-fold cross-validation to predict cycle-specific dose modifications for 5-fluorouracil, irinotecan, and oxaliplatin. Model performance was evaluated using area under the receiver operating characteristic curve. Results: The cohort included 514 patients receiving FOLFIRINOX across 5,041 treatment cycles. The mean age was 59 years, 60% of patients were White, 41% had a history of smoking, and patients received a median of 6 chemotherapy cycles. More than 60% of patients required at least one dose modification during treatment. XGBoost demonstrated the highest performance across component drugs, with AUCs ranging from 0.53 to 0.70. Clinically plausible predictors of irinotecan and oxaliplatin dose modification included hepatic and renal function markers, cumulative drug exposure, treatment-related symptoms, and demographic or behavioral characteristics. Conclusion: We developed an EMR-based machine-learning framework to model real-world FOLFIRINOX dose modification and identified clinically plausible, routinely available predictors, particularly for irinotecan and oxaliplatin. Variable model performance suggests that dosing decisions are only partially captured by structured EMR data, highlighting both the limitations of current data-driven approaches and clinical domains where ML-based models may support individualized dosing and toxicity surveillance. Future informatics efforts should incorporate dose-modification rationale, patient-reported and functional outcomes, and validation across diverse practice settings.