Prediction of pharmacist medication interventions using medication regimen complexity.

Area of Science:

• Pharmacology
• Health Informatics
• Critical Care Medicine

Background:

• Critically ill patients require complex medication management for safety and efficacy.
• Critical care pharmacists (CCPs) perform medication interventions to optimize patient care.
• Predicting patient medication management needs is crucial for defining CCP workflow and intervention timeliness.

Purpose of the Study:

• To develop prediction models for the number and intensity of medication interventions in critically ill patients.
• To identify patient-specific factors influencing medication management needs within the first 24 hours of ICU admission.

Main Methods:

• Retrospective observational cohort study of 13,373 adult ICU patients (June 2020 - June 2023).
• Development of regression and machine learning (Random Forest, SVM, XGBoost) models to predict intervention counts.
• Evaluation of models using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Symmetric Mean Absolute Percentage Error (sMAPE).

Main Results:

• The average number of interventions was 4.7 (SD 7.1) and intervention intensity was 24.0 (40.3).
• Machine learning models performed similarly to stepwise logistic regression, outperforming a base model.
• Random Forest model yielded the lowest RMSE (9.26), while Support Vector Machine had the lowest MAE (4.71).

Conclusions:

• Medication intervention quantity can be predicted using patient-specific factors identified within 24 hours of admission.
• Machine learning models did not offer a substantial performance advantage over traditional regression models in this study.
• The developed models provide a framework for institutional workload modeling, adaptable to variations in intervention documentation.