“Dial-an-ROI?” Changing Basic Variables Impacts Cost Trends in Single-Population Pre-Post (“DMAA Type”) Savings Analysis

Abstract

Disease management (DM) programs claim to achieve cost savings by reducing clinical adverse events. While measuring changes in adverse events is straightforward, plausibly demonstrating savings has been contentious, especially absent an external comparison population. In this situation, a single-population methodology is often used, in which the cost trend for those with no program conditions (“non-chronics”–NC) forms the expected trend for those who have at least 1 program condition (“chronics”–C). The methodology's fundamental assumption is that—absent DM—C and NC trends would be identical (or bear a constant relationship over time). We assessed this assumption by altering the values of key variables used to identify C and NC, and to calculate trend.

We compared C and NC baseline trends for a 23-condition telephonic DM multiemployer program representing nearly 300,000 members. Trends were calculated for 16 combinations of values for 4 key variables: identification look-back frame (12 vs. 24 months); identification threshold (high vs. lower specificity); claims runout (3 vs. 6 months); and minimum required insurance eligibility (any 6 months vs. 12 months continuous eligibility in the measurement year). Identification was performed by annual qualification.

Changes in values for the 4 key variables markedly impacted baseline C and NC trends. C trends varied between 10.1% and 13.1%; NC trends between 5.2% and 12.8%. C-NC trend differences ranged between −1.9% and +7.0%. The combination of 24 months identification look-back, high identification threshold, 6 months runout, and any-6-months eligibility gave the most convergent C and NC trends (10.4% and 10.7%).

Seemingly minor changes in key variables impact C and NC trends in single-population pre-post DM savings methodologies. When a suitable comparison population is not available, at least 1 year of baseline C and NC trends should be reported–as recommended by the DMAA—and values of key variables used should be specified. Plausibility metrics (eg, hospitalizations) should be reported. (Population Health Management 2009;12:17–24)