reports

Validation and uses of the ACG-DX predictive modeling and risk adjustment tool in an Israeli HMO

Clin Med Res 8:51-52.

Maccabi Healthcare Services, Tel Aviv, Israel

BACKGROUND: With an ever increasing population of HMO members with chronic disease and continuing economic pressures to provide good medical care as efficiently as possible, there is a need for methods to stratify populations by health status and their future risk for using limited healthcare resources.

METHODS: We used the ACG-DX program developed at Johns Hopkins University for validation testing on an Israeli PPO population. This program uses same year data to provide a risk adjustment analysis for that year and predict high- users (top 5%) for the following year. The Maccabi Healthcare Services has an extensive demographic and medical database keyed by a unique identifier for each member. We provided age and gender data as well as all diagnoses for year 2006 and total cost for years 2006 and 2007 for each member as input to the ACGDX program. We truncated total costs over the 99.9% for both pharmacy and total costs. The explanatory value for costs in 2006 (R²) and positive predictive value for total costs in 2007 were calculated for the entire population of 1.7 million subscribers. The R² results were compared to results based on American data and reference values (ref.) generated by the program.

RESULTS: The R² explanatory power for same year costs for age 65 and over was 18.7% for total cost (ref. 16%) and 45.5% for pharmacy cost (ref.10%). For those under age 65, the values were 20.5% (ref. 21%) for total cost and 39.1% (ref. 29%) for pharmacy cost. In predicting which members would be in the high (top 5%) cost bracket, we correctly identified 33,479 members (based on 2006 data) as being at risk for high total cost out of the 85,059 members who did have a high total cost in 2007. The positive predictive value was 39.4%.

CONCLUSIONS: We have validated the ACG-DX tool for an Israeli population for both risk management and predictive modeling. This tool can be now used for a large variety of managerial and medical uses, including population, disease and case management, budget allocation and provider payment as well as improving health care equity and quality assurance.