Assessing treatment switch among patients with multiple sclerosis: A machine learning approach.
Jieni Li1, Yinan Huang2, George J Hutton3
1Department of Pharmaceutical Health Outcomes and Policy, College of Pharmacy, University of Houston, TX, USA.
Exploratory Research in Clinical and Social Pharmacy
|August 9, 2023
Summary
Machine learning models can predict disease-modifying agent (DMA) switching in multiple sclerosis (MS) patients. Random forest models showed better discrimination than logistic regression, though overall predictive performance was similar.
Area of Science:
- Medical Informatics
- Machine Learning in Healthcare
- Multiple Sclerosis Therapeutics
Background:
- Multiple sclerosis (MS) patients often switch disease-modifying agents (DMAs) due to efficacy and safety concerns.
- Predicting treatment switching is crucial for optimizing patient care and outcomes.
Purpose of the Study:
- To develop and compare machine learning (ML) models, specifically random forest (RF) and logistic regression (LR), for predicting DMA switching in MS patients.
- To identify key factors influencing DMA switching behavior.
Main Methods:
- A retrospective longitudinal study utilizing the TriNetX electronic medical records (EMR) network.
- Identification of 7258 adult MS patients with at least one DMA prescription between 2010 and 2017.
- Development and comparison of RF and LR models using 72 baseline characteristics, evaluated by AUC, accuracy, recall, G-measure, and F-1 score.
Main Results:
- 16% of MS patients switched DMAs within two years.
- The RF model demonstrated significantly better discrimination (AUC=0.65) compared to the LR model (AUC=0.63).
- RF and LR models showed similar predictive performance based on F- and G-measures; key predictors included age, index medication type, and year.
Conclusions:
- Random forest models offer improved discrimination for predicting DMA switching in MS patients compared to logistic regression.
- Despite AUC differences, RF and LR models exhibit comparable predictive utility based on F- and G-measures.
- Further research is warranted to explore ML's role in predicting treatment outcomes and guiding clinical decision-making for optimal MS management.
