Traditional Risk Indices as Predictors of Future Utilization and Charges in the Context of Population Health for an Uninsured Cohort

• Published in: EGEMS (Washington, DC) DC), 2017-06, Vol.5 (1), p.11-11
• Main Authors: Lubanski, Ethan, Rozario, Nigel, Moore, Charity G, Mulder, Holly Petruso, Dulin, Michael, Ludden, Tom, Rossman, Whitney, Ashby, Avery, McWilliams, Andrew
• Format: Article
• Language: English
• ISSN: 2327-9214; 2327-9214
• DOI: 10.5334/egems.220

The uninsured population presents unique challenges to the application of an integrated approach to population health. Our objective is to compare and test population risk indices for identifying a cohort of uninsured patients at high-risk for avoidable healthcare utilization and costs. Patients who had a least one visit at a safety-net clinic, had a primary address in Mecklenburg County, were aged 18-74, and had the most recent healthcare visit coded as 'uninsured' were identified in the baseline period. The five risk indices used were: the HHS Hierarchical Conditions Category (HCC), the Charlson Comorbidity Index (CCI), Total Cost Index, Total Inpatient Visits Index, and Total Emergency Department Visits Index. First, agreement across the five indices was analyzed. Then, the accuracy of the five risk indices was tested in predicting future utilization and costs for the subsequent 12-month follow-up period. Kappa statistics and percent overlap values showed below average to poor agreement between indices when comparing scorers.The strongest predictors of being in the 90 percentile of total cost during the 12 months follow-up period were the Total Cost Index at baseline (C statistic=0.75) and the HCC (C-statistic=0.73). The CCI and Total Inpatient Visit Index's demonstrated the lowest accuracy for predicting an unnecessary ED visit (C-statistic=0.51, for both). Prior cost and ED utilization were key in predicting their corresponding 12-month metrics. In contrast, the Total Inpatient Visit Index had the worst predictive performance for future hospitalization rates. Some indices were similarly predictive as compared to insured cohorts but others showed contrasting results.