Statistical prediction of high-cost claimants using commercial health plan data

Abstract

Among Cigna’s claimant population with at least one year of continuous medical or pharmacy eligibility over 2014–2015 (N = 2.7 million), our objective was to accurately identify high-cost claimants and identify clinical and demographic cost drivers among individuals with commercial health plan benefits. High-cost claimants were defined as those with annual costs over $100,000. We collected 800+ potential risk factors and utilized multivariable weighted logistic regression on an oversampled model dataset to estimate odds ratios for clinical and demographic factors available in claims data. We used decision tree methodology to assist in variable selection/reduction. High-cost claimants (n = 17,702) comprised only 0.6% of the 2015–2016 population, but accounted for over 20% of 2015–2016 total costs. Our optimized maximum likelihood estimation model identified cost drivers including: actuarial prospective episode-related group (ERG) risk score, prior-year medical claim costs, prior-year pharmacy claim costs, gaps in care/noncompliance score, hemophilia, short stature, and end-stage renal disease. Our findings show that weighted logistic regression modeling with oversampling techniques can be used to identify high-cost claimants in the upcoming year more accurately than traditional maximum likelihood estimation. Managed care decision makers should use prospective claims data analyses to target and implement intervention programs, with the goal of managing care among those at risk for incurring catastrophic costs.