Our quantitative outcomes analysis consists of two parts: (1) a comparison of average outcomes before and after the start of the SASH program, and between SASH program participants and beneficiaries in the two comparison groups; and (2) estimating the impact of the SASH program on outcomes in a regression framework. The statistical model we use to estimate the program impact is a difference-in-differences (DID) model. It is estimated separately for the two SASH comparison groups: the non-SASH/MAPCP Demonstration group and the non-SASH/non-MAPCP Demonstration group (see Appendix C).
For the Medicare expenditure outcomes, we use a linear version of the DID model. In this case, the impact estimate is the (regression-adjusted) difference between SASH program participants and the comparison group in the change in outcome levels between the baseline and intervention periods. As such, we will refer to this estimate as a DID estimate, which can be considered the average program effect across the entire period of SASH participation to date. A negative DID estimate indicates that, between the baseline and intervention periods, average outcomes among SASH program participants either increased by a smaller amount or decreased by a larger amount, relative to the comparison group. Thus, negative DID estimates are indications that the SASH program was successful in reducing the expenditure trend among intervention beneficiaries, relative to the comparison group. Positive DID estimates reflect the opposite.
For the utilization outcomes, we use a non-linear (negative binomial) version of the DID model. In this case, the impact estimate shows whether during the intervention period the (regression-adjusted) utilization rate increased or decreased among SASH program participants, relative to the comparison group. The estimate does not have a DID interpretation, so for utilization outcomes we will simply refer to the "impact estimate" or "SASH program effect." Positive numbers indicate that the SASH program was associated with increased utilization relative to the comparison group, whereas negative numbers indicate a decrease in utilization.