Risk screening in a medicare/medicaid population administrative data versus self report

To compare the abilities of two validated indices, one survey-based and the other database-derived, to prospectively identify high-cost, dual-eligible Medicare/Medicaid members.

A longitudinal cohort study.

A Medicaid health maintenance organization in Philadelphia, Pa.

HMO enrollees (N = 558) 65 years and older eligible for both Medicare and Medicaid.

Two hundred ninety six patients responded to a survey containing the Probability of Repeat Admission Questionnaire (Pra) between October and November 1998. Using readily available administrative data, we created an administrative proxy for the Pra. Choosing a cut point of 0.40 for both indices maximized sensitivity at 55% for the administrative proxy and 50% for the survey Pra. This classification yielded 103 high-risk patients by administrative proxy and 73 by survey Pra. High-cost patients averaged at least 2.3 times the resource utilization during the 6-month follow-up. Correlation between the two scores was 0.53, and the scales disagreed on high-cost risk in 78 patients (54 high-cost by administrative proxy only, and 24 high-cost by survey Pra only). These two discordant groups utilized intermediate levels of resources, $2,171 and $2,794, that were not statistically significantly different between the two groups (probability > chi2 =.66). Receiver operating characteristic curve areas (0.68 for survey Pra and administrative proxy for respondents, and 0.67 by administrative proxy for nonrespondents) revealed similar overall discriminative abilities for the two instruments for costs.

The Medicaid/Medicare dual-eligible population responded to the survey Pra at a rate of 53%, limiting its practical utility as a screening instrument. Using a cut point of 0.40, the administrative proxy performed as well as the survey Pra in this population and was equally applicable to nonrespondents. The time lag inherent in database screening limits its applicability for new patients, but combining database-driven and survey-based approaches holds promise for targeting patients who might benefit from case management intervention.