A machine learning automation system for utilization management

Abstract

We develop high-performance machine learning automation systems for utilization management that are effective across all specialties. We were motivated by the knowledge that current automation systems for utilization management are rules- based and focused on narrow subsets of healthcare specialties. We develop models that can automate nearly 90% of a utilization management team's workload with less than 1% error. We evaluate these models on both historical data and as part of a live system in industry. The performance and efficacy of our models are consistent across both evaluation domains, demonstrating the generalizability of our work.