Do hierarchical condition category model scores predict hospitalization risk in newly enrolled Medicare advantage participants as well as probability of repeated admission scores?

OBJECTIVES: To compare how well hierarchical condition categories (HCC) and probability of repeated admission (P(RA)) scores predict hospitalization.
DESIGN: Longitudinal cohort study with 12-month follow-up.
SETTING: A Medicare Advantage (MA) plan. PARTICIPANTS: Four thousand five hundred six newly enrolled beneficiaries. MEASUREMENT: HCC scores were identified from enrollment files. The P(RA) tool was administered by mail and telephone. Inpatient admissions were based on notifications. The Mann-Whitney test was used to compare HCC scores of P(RA) responders and nonresponders. The receiver operating characteristic curve provided the area under the curve (AUC) for each score. Admission risk in the top 5% of scores was evaluated using logistic regression.
RESULTS: Within 60 days of enrollment, 45.1% of the 3,954 beneficiaries with HCC scores completed the P(RA) tool. HCC scores were lower for the 1,783 P(RA) respondents than the 2,171 nonrespondents (0.71 vs 0.81, P<.001). AUCs predicting hospitalization with regard to HCC and P(RA) were similar (0.638, 95% confidence interval (CI)=0.603-0.674; 0.654, 95% CI=0.618-0.690). Individuals identified in the top 5% of scores using both tools, using HCC alone, or using P(RA) alone had higher risk for hospitalization than those below the 95th percentile (odds ratio (OR)=8.5, 95% CI=3.7-19.4, OR=3.8, 95% CI=2.3-6.3, and OR=3.9, 95% CI=2.3-6.4, respectively).
CONCLUSION: HCC scores provided to MA plans for risk adjustment of revenue can also be used to identify hospitalization risk. Additional studies are required to evaluate whether a hybrid approach incorporating administrative and self-reported models would further optimize risk stratification efforts.