Finding future high-cost cases: comparing prior cost versus diagnosis-based methods.

Abstract

OBJECTIVE: To examine the value of two kinds of patient-level dat a (cost and diagnoses) for identifying a very small subgroup of a general population with high future costs that may be mitigated with medical management. DATA SOURCES: The study used the MEDSTAT Market Scan (R) Research Database, consisting of inpatient and ambulatory health care encounter records for individuals covered by employee- sponsored benefit plans during 1997 and 1998. STUDY DESIGN: Prior cost and a diagnostic cost group (DCG) risk model were each used with 1997 data to identify 0.5-percent-sized "top groups" of people most likely to be expensive in 1998. We compared the distributions of people, cost, and diseases commonly targeted for disease management for people in the two top groups and, as a bench mark, in the full population. PRINCIPAL FINDINGS: the prior cost- and DCG-identified top groups overlapped by only 38 percent. Each top group consisted of people with high year-two costs and high rates of diabetes, heart failure, major lung disease, and depression. The DCG top group identified people who are both somewhat more expensive ($27,292 vs. $25,981) and more likely ( 49.4 percent vs. 43.8 percent ) th an the prior-cost top group to have at least one of the diseases commonly targeted for disease management. The overlap group average cost was $46,219. CONCLUSIONS: Diagnosis-based risk models are at least as powerful as prior cost for identifying people who will be expensive. Combined cost and diagnostic data are even more powerful and more operation ally useful, especially because the diagnostic information identifies the medical problems that may be managed to achieve better out comes and lower costs.