Care management programs have emerged as a popular response to the continued growth and concentration of health care expenditures. The contribution of effective care management to secondary and tertiary prevention relies in part on efficiently identifying """"""""high-risk"""""""" patients. Prospective identification systems that can recognize characteristics of predictable expense are preferred to contemporaneous or retrospective systems, particularly if system administrative costs are low. Risk- assessment models that use demographic and diagnostic information contained in HMO administrative data sets represent a logical, yet virtually unexplored, prospective identification methodology. Collaborators at the Kaiser Permanente Center for Health Research (CHR), the Center for Health Studies, and elsewhere have created a large administrative data set from 6 HMOs across the U.S. This data set has been created to develop prospective health risk- assessment models for primary use in adjusting Medicare payments to health plans for enrollee health status. These data also present a unique opportunity to study risk-assessment models as preliminary population-based screens for enrollees at relatively higher risk of generating large future medical expenditures, i.e., becoming """"""""high-cost."""""""" We assert that early identification of these enrollees can promote effective secondary and tertiary preventive measures. The HERALD project leverages this opportunity to provide useful information for health plan decisionmakers (e.g., care management program directors) concerned with reducing the illness burden in their populations as efficiently as possible.
The specific aims of HERALD are to: 1. Compare the ability to forecast future """"""""high-cost"""""""" status of the Global Risk-Assessment Model (GRAM) Adjusted Clinical Groups (ACGs, formerly called Ambulatory Care Groups) Diagnostic Cost Groups (DCGs) Chronic Disease Scores A logistic model; and A prior-expense model. 2. Evaluate the """"""""high-cost"""""""" forecasting performance of GRAM: Across various enrollee population sizes (e.g., 50,000, 100,000) Within particular population subgroups (e.g., elderly, children, and older dependents) For multiple categories of risk of high-cost status (e.g., extreme, moderate, low). 3. Evaluate the temporal stability of GRAM's forecasting performance (e.g., distance in time - 1995 risk factors forecasting 1997 expense). We expect the results from HERALD (1) to serve as a base for future explorations of risk-assessment models such as GRAM enhanced by data on secondary diagnoses, functional health status, and behavioral risk factors; and (2) inform a future test of GRAM's """"""""value-added"""""""" as a preliminary screen for an administered health status survey.