Using the sample of PACE enrollees matched to the combined pool of NH entrants and HCBS waiver enrollees, we compared the average per beneficiary monthly capitation payments under PACE with predicted expenditures for PACE enrollees had they been in an HCBS waiver program or in a NH in successive six-month intervals since enrollment. Most of the differences between actual and predicted monthly Medicare expenditures for PACE enrollees were statistically insignificant at the 10 percent level (Table 3). In the first six months after sample entry, actual Medicare expenditures for PACE enrollees were significantly lower than predicted expenditures by nearly $2,000 (p-value < 0.001).16 However, in three other intervals (third, fifth, and tenth), actual Medicare expenditures significantly exceeded predicted expenditures by $234 to $445. In the remaining seven intervals, the difference between actual and predicted Medicare expenditures was statistically insignificant.
Actual monthly Medicaid expenditures on PACE enrollees, however, significantly exceeded predicted Medicaid expenditures in all seven intervals. The magnitude of the difference was also quite stable over time--between $546 and $647, and around the $600 mark (all p-values < 0.001). For combined Medicare and Medicaid expenditures, actual costs for PACE enrollees significantly exceeded predicted expenditures by $426 to $917 in five intervals, and the difference, though positive and large, was statistically insignificant in the last interval. However, in the first interval, actual costs were significantly lower by $1,400--a consequence of the large negative Medicare cost difference.
Results from Sensitivity Tests for Medicare and Medicaid Expenditures
We ran two sensitivity tests for the expenditures analysis to confirm that the main findings were robust to changes in the study sample. First, we dropped 606 PACE enrollees, comprising about 16 percent of the treatment group members, who were either enrolled in HCBS waiver services in the six months prior to enrolling in PACE or were in a NH in the 90 days prior to enrolling in PACE, and also dropped around 4 percent of the HCBS waiver enrollees who were in a NH in the 90 days prior to enrolling in the waiver program. The reason for dropping these PACE or HCBS enrollees from the sample prior to matching was to maintain comparability between the two groups by limiting the samples to individuals who were initiating a need for long-term support services at the time of sample entry. We reran matching with the remaining 3,119 PACE enrollees and obtained 2,806 matched comparison group members (1,619 NH entrants and 1,187 HCBS waiver enrollees). The two groups (PACE enrollees and matched HCBS waiver and NH entrants) were well-matched in all baseline covariates, as before (not shown). Further, the results for Medicare, Medicaid, and combined expenditures were similar to the main findings, with few significant differences in Medicare expenditures, actual Medicaid expenditures significantly higher-than-predicted expenditures in all seven intervals, and actual combined expenditures being significantly higher-than-predicted expenditures in all but the first interval, where it was significantly lower by close to $1,500 (Table 4). One difference between the findings from this first sensitivity test and the main findings was that the difference in capitation payments and predicted Medicaid expenditures increased over time from $458 to $964, instead of remaining stable around $600. This is reflected in the findings for combined Medicare and Medicaid expenditures as well.
In another sensitivity test, we excluded the State of New York from the study sample. This test served dual purposes. First and most important, New York being the largest state in the sample and comprising over one-quarter of the full PACE sample, it allowed us to test that the results were not driven by New York. Second, since New York has a number of managed long-term care plans with capitated HCBS but all other Medicaid services provided on an FFS basis, Medicaid expenditures for the matched comparison enrollees from New York could be underestimated, since we calculate FFS expenditures only from the MAX files. Note that the direction of possible bias in the findings from such an underestimation of Medicaid costs for the matched comparison group is quite clear--it underestimates the cost that PACE enrollees would likely have incurred had they remained in FFS, and therefore increases the likelihood of finding that capitated Medicaid payments for PACE exceed predicted expenditures in New York and overall (if New York is included). Given Medicaid expenditure results that do suggest higher Medicaid expenditures under PACE, it was important to test that our main results were unaffected after the exclusion of New York from the study sample.
With the treatment and matched comparison group members from only seven study states, we obtained a new set of findings from the expenditures analysis that closely resembled the main findings (Table 5), further boosting confidence in these impact estimates. For instance, there were only a few significant differences for Medicare expenditures; significantly higher actual Medicaid expenditures for PACE enrollees relative to predicted expenditures in all intervals, and higher actual combined expenditures than predicted expenditures in all but the first interval, as before. One important difference was that the significant gap in Medicaid spending--between actual and predicted expenditures for PACE enrollees--was much higher at $990-$1,323, compared to differences of around $600 in the main findings. This suggests that capitated Medicaid expenditures for PACE enrollees was lower than predicted expenditures in New York, as is borne out by results in the following subsection. More importantly, it suggests that once we rule out any incompleteness in Medicaid expenditures data by excluding New York from the analysis, PACE enrollees still have consistently higher Medicaid capitation payments than predicted expenditures across the remaining seven study states, and in all seven intervals. This also leads to a higher gap between actual and predicted combined expenditures, as reflected in the findings for combined Medicare and Medicaid expenditures in Table 5.
State-Specific Findings for States with Adequate Sample Size
We also looked at state-specific results for the three states with the largest new PACE enrollment over the 2006-2008 period--California, Massachusetts, and New York. However, for these state-specific results--especially for Massachusetts--the sample sizes in most of the later cost intervals, for example, all cost intervals from month 37 of the followup period onwards, were rather small. Hence, the results for these later intervals are likely to be less reliable, especially for Massachusetts.
The state-specific results point towards some interesting differences across states in the expenditure findings. For instance, in California, as for all eight states taken together--there were few significant differences for Medicare expenditures (four significant difference--all negative, suggesting lower actual expenditures than predicted expenditures--for the first four intervals), and significantly higher actual Medicaid and combined expenditures than predicted expenditures for PACE (Table 6). However, the Medicaid spending gap in California was much larger compared to that for the main findings across all eight states, and increased over time from $1,574 in the first interval to $2,672 in the last (and over $3,000 in the fourth and sixth intervals)--due both to an increase in actual PACE capitation payments and a reduction in predicted expenditures over time--a pattern present in the results for combined Medicare and Medicaid expenditures as well.
For Massachusetts, we discuss results through month 36 or the first six intervals only, since the number of PACE enrollees drops to 66 or below in the later intervals. In general, the pattern of findings in Massachusetts for Medicare, Medicaid, and combined expenditures was similar to the main findings (Table 7). However, the Medicaid spending gap decreased over time from $648 in the first interval to $172 in the sixth--due to an increase in predicted expenditures--and the spending gap was statistically significant only in the first three intervals.
The Medicaid expenditure findings for New York differ markedly from the main findings and from those for California and Massachusetts. First, capitated Medicare expenditures were significantly higher under PACE than predicted expenditures in three intervals, with the difference in the first interval continuing to be negative and significant as before. Enrollees' actual Medicaid costs under PACE were significantly lower than their predicted Medicaid FFS costs in all seven intervals by $674-$1,046. Combined Medicare and Medicaid capitated expenditures were lower in all intervals as well, though significantly different from predicted expenditures only in three of the seven intervals (Table 8). Since Medicaid expenditures for the matched comparison group are likely to be underestimated in New York, the magnitude of the negative gap in Medicaid spending would be even larger if Medicaid costs in the comparison group were not potentially incomplete. Hence, our findings show that actual Medicaid expenditures under PACE were significantly lower in New York than expected costs had PACE enrollees been in HCBS waiver programs or in NHs instead.
The Medicaid results in New York seem to be driven by fairly constant or slightly decreasing Medicaid capitation payments over time that are consistently lower than the slightly increasing predicted Medicaid expenditures. For California, the four negative and statistically significant differences on Medicare expenditures was a consequence of both lower actual Medicare capitation payments and higher predicted Medicare expenditures than other states. Finally, for Massachusetts, even though Medicaid capitation payments were lower than all eight states taken together, even lower predicted expenditures resulted in a positive Medicaid spending gap, which diminished as predicted expenditures increased over time. With a longer followup for Medicaid expenditures, it would have been possible to test for favorable Medicaid expenditure findings in Massachusetts as well.
Notably, for both California and Massachusetts--states with a positive Medicaid spending gap--NH entrants comprised the majority of matched comparison group members, especially in California. This suggests that in spite of higher expected Medicaid costs in the comparison group for these states, PACE capitation payments from Medicaid still exceeded predicted Medicaid expenditures. In contrast, the mix of NH entrants and HCBS waiver enrollees was more balanced in New York, with HCBS enrollees slightly outnumbering NH entrants. This could possibly lead to an additional under-prediction of comparison group Medicaid costs in New York, which further established the robustness of our finding of a negative Medicaid spending gap in favor of PACE in New York.
We also tested the state-specific impact estimates for Medicare, Medicaid and total expenditures to determine if estimated impacts were significantly different from each other across states. In general, the impact estimates for Medicaid expenditures were significantly different from each other in all state-to-state comparisons, while the impact estimates for Medicare expenditures tended to differ across states in the first few intervals only (results not shown). Comparing California and Massachusetts, the impact estimates for Medicare expenditures were significantly different from each other at the 10 percent level in the first four intervals, impact estimates for Medicaid expenditures differed significantly in all intervals, and impact estimates for combined expenditures were different in all six intervals reported for Massachusetts. Comparing California and New York, the impact estimates for Medicare expenditures were significantly different from each other at the 10 percent level in the first five intervals, impact estimates for Medicaid expenditures differed significantly in all intervals, and impact estimates for combined expenditures were different in all but the first interval. Finally, comparing Massachusetts and New York, the impact estimates for Medicare expenditures were significantly different from each other at the 10 percent level only in the first interval, impact estimates for Medicaid expenditures differed significantly in all six intervals reported for Massachusetts, and impact estimates for combined expenditures were different in the second to fifth intervals.