Development of Machine Learning Models to Validate a Medication Regimen Complexity Scoring Tool for Critically Ill...

Area of Science:

• Critical care medicine
• Health informatics
• Machine learning in healthcare

Background:

• The Medication Regimen Complexity - Intensive Care Unit (MRC-ICU) is a novel tool for quantifying medication complexity in critically ill patients.
• Assessing medication regimen complexity is crucial for understanding patient outcomes in intensive care settings.

Purpose of the Study:

• To evaluate the effectiveness of machine learning models in predicting inpatient mortality.
• To determine if incorporating medication regimen complexity, specifically using the MRC-ICU score, improves predictive accuracy.

Main Methods:

• A retrospective observational study of 130 adult medical ICU patients.
• Development and testing of six machine learning classifiers (KNN, NB, random forest, SVM, neural network, logistic classifier) using electronic health record data.
• Comparison of three models: demographic data only, demographic data with medication complexity variables, and demographic data with the MRC-ICU score.

Main Results:

• Models incorporating medication regimen complexity variables and the MRC-ICU score demonstrated improved prediction of inpatient mortality.
• The logistic classifier achieved 83% accuracy, with high sensitivity (87%) and positive predictive value (93%).
• The APACHE III score and the 24-hour MRC-ICU score were identified as the most significant predictive variables.

Conclusions:

• The inclusion of the MRC-ICU score enhances the predictive capability for patient outcomes, outperforming the established APACHE III score.
• This study validates the MRC-ICU tool's potential for improving patient outcome predictions in critical care settings.