Predicting high-need cases among new Medicaid enrollees.

Objectives: To assess the ability of a short, self-reported health needs assessment (HNA) collected at the time of Medicaid enrollment to predict subsequent utilization and costs.

Study Design: Retrospective cohort study.

Methods: We analyzed individual-level data that included self-reported HNAs, medical care encounter records, and administrative eligibility records for 34,087 childless adult Medicaid enrollees in Wisconsin, covering the period 2009-2010. High need was operationalized using the following outcome variables measured over the first year of program enrollment: having an inpatient admission; membership in the top decile of emergency department (ED) utilization; and membership in the top cost decile. We assessed the ability of the HNA to predict high-need cases using several complementary methods: the C-statistic; integrated discrimination improvement; and sensitivity, specificity, and positive predictive value resulting from multivariate logistic regression estimates.

Results: Using the HNA along with sociodemographic measures met the Hosmer-Lemeshow criterion for adequate predictive performance for the high ED and high cost outcomes (C-statistics of 0.74 and 0.72, respectively). The HNA was associated with large improvements in predictive performance over sociodemographic measures alone for all 3 dependent variables (integrated discrimination improvement of 182%, 413%, and 300% for ED, cost, and inpatient variables, respectively). The HNA also led to considerable improvements in sensitivity and positive predictive value with no resulting decreases in specificity or negative predictive value.

Conclusions: Collecting self-reported health measures for a Medicaid expansion population can yield data of sufficient quality for predicting high-need cases.