Multiple sclerosis risk stratification and healthcare cost prediction using machine learning

Abstract

BACKGROUND: The increasing availability of healthcare data offers an opportunity to address chronic diseases such as multiple sclerosis (MS) more proactively. We aimed to develop a machine learning (ML) approach to identify high-risk MS patients and predict their healthcare spending. METHODS: We conducted a retrospective analysis of de-identified commercial insurance claims from over 267,000 individuals (631 with MS), spanning January 2016 to June 2018. Monthly claims data were aggregated, and 72 ML regression and 63 classification models were trained to predict which patients would be in the top decile of healthcare spending over the subsequent four months. Model performance was compared with predictions based on four-month and one-month historical spending assessments. RESULTS: MS patients comprise less than 0.3% of the study population yet account for over 2.5% of total healthcare expenditures. In a four-month evaluation dataset, our ML models capture an average of 76.0% of the actual top-decile spending, surpassing the four-month (43.5%) and one-month (36.5%) historical methods. Notably, the ML approach identifies more individuals transitioning into high-cost status, suggesting potential utility in guiding earlier clinical decisions. CONCLUSIONS: Our proof-of-concept ML-driven framework predicts imminent high-cost MS patients more accurately than simpler, retrospectively focused approaches. These findings may inform proactive risk stratification and resource allocation strategies, though further investigation is needed to determine how best to integrate these predictions into clinical practice.

Ähnliche Arbeiten