The final analysis sample using the first comparison group strategy consisted of 3,725 PACE and 3,264 matched comparison group members, of whom 1,834 were NH entrants and 1,430 were HCBS waiver enrollees. Table 1 shows the distribution of the final, matched study sample across the eight states in the study, by treatment status. The number of PACE enrollees in our final matched sample ranged from 118 in New Mexico to 952 in New York, with New York, Massachusetts, and California being the three largest states in terms of sample size or the number of treatment and matched comparison group members.14 For four states (California, Massachusetts, Michigan, and Pennsylvania) NH entrants comprised the majority of the matched comparison group, while for the other four states (Colorado, New Mexico, New York, and Oregon), there were a greater number of HCBS waiver enrollees in the matched comparison group. Table 1 also shows the sample breakdown for the second matched comparison group, comprised of HCBS waiver enrollees alone. The second matched comparison group is somewhat smaller, with 2,745 HCBS waiver enrollees.
Figure 2a and Figure 2b present histograms of the estimated propensity score in the treatment and matched comparison groups, for the two matched samples. In each of these figures, the distribution of the score across the treatment and matched comparison groups, that is, the percentage of treatment and matched comparison group enrollees in each bin of the probability distribution of being in PACE, was similar. In other words, the treatment and matched comparison groups were well-matched on the probability distribution of being in treatment status.
In Table 2, we present the results from testing for baseline equivalence in the matched sample when PACE enrollees are matched to the combined pool of NH entrants and HCBS waiver enrollees (A) or to HCBS enrollees alone (B). We find that PACE and the matched comparison group A were well-matched in baseline characteristics, chronic conditions, and in Medicare service use and costs. For instance, the average age of beneficiaries in both groups was 79 years with around 70 percent being female; more than a third of beneficiaries in both groups had Alzheimer's or dementia, around a third of the enrollees had an inpatient admission in the calendar year prior to the year of sample entry, and their average, annualized Medicare expenditures in the prior year were over $14,000. The only statistically significant difference between the two groups was in the prior use of SNFs, with the treatment group at a slightly higher rate of use (13 percent) relative to the matched comparison group (11 percent). An F-test reported at the bottom of Table 2 did not reject the null hypothesis of equality in the joint distribution of all matching variables across the two groups. Also, when PACE enrollees were matched to HCBS waiver enrollees alone (matched comparison group B in Table 2), the two groups were nearly identical in their mean baseline characteristics, chronic conditions, and Medicare service use and costs--with none of the differences being statistically significant.
Table 2 also reports means for two other variables that were not used in matching --whether a beneficiary was enrolled in Medicaid in the 12 months prior to sample entry and whether a beneficiary was in a NH in the 90 days prior to sample entry. We find that a similar percent of treatment and matched comparison group members had Medicaid enrollment in the prior 12 months. Specifically, 79 percent of PACE enrollees, 77 percent of matched comparison group members in sample A, and 81 percent of matched HCBS waiver enrollees in sample B were enrolled in Medicaid in the prior 12 months, with the difference between PACE enrollees and matched comparison group members in sample A being statistically significant. In general, a small percentage of members in the treatment and matched comparison groups--7 percent, 2 percent, and 5 percent respectively in PACE, mixed comparison sample (A), and HCBS-only comparison sample (B)--were in a NH in the prior 90 days, with the treatment-comparison difference statistically significant in both matched samples. This percentage is especially low (2 percent) in the matched comparison sample A, since we excluded NH entrants with any NH stay in the prior 90 days from the study sample, and therefore, we did not use this as a matching variable in our analysis.15