Medicare and Medicaid Expenditures
We examine PBPM Medicare expenditures, Medicaid expenditures, and total expenditures over successive six-month intervals from the month of enrollment or NH entry for PACE enrollees and their matched comparison group counterparts (HCBS enrollees and NH entrants). For PACE enrollees, aggregate expenditures during a particular interval are essentially the sum of monthly Medicare or Medicaid capitation payments during that interval, and for HCBS enrollees or NH entrants these are the total FFS expenditures (Medicare or Medicaid) during that interval. PACE enrollees in our sample who leave PACE are therefore lost to followup in our study. Similarly, we also stop following matched comparison group members once they leave FFS Medicare or Medicaid and enroll in a managed care program.12 In other words, unlike in an intent-to-treat analysis where beneficiaries would have been followed until death or till the end of the observation period regardless of their enrollment status, we follow treatment group beneficiaries as long as they are alive and enrolled in PACE, and follow comparison group beneficiaries as long as they are alive and not in managed care. Hence, PBPM expenditures during each six-month interval are calculated as Medicare or Medicaid expenditures during that interval divided by the number of eligible months during that interval, where eligible months are defined as months alive and enrolled in PACE or months alive, enrolled in Medicaid and Medicare, and not in managed care (for the HCBS and NH groups).
We compare the weighted mean capitation payments (separately for Medicare and Medicaid, and the combined total) PBPM, for PACE enrollees in each six-month interval, where the weights are equal to the share of months eligible during that interval, to the weighted mean of their predicted expenditures. To predict what the monthly Medicare and Medicaid expenditures for PACE sample members would have been in the absence of PACE, we estimated linear regression models for these two public programs using the matched comparison group of beneficiaries in each interval. These regressions use weights calculated as the product of share of months eligible and the matching weight for comparison group members. The independent variables in the regression models include all variables used in propensity score matching, as well as the number of months between the end of the pre-enrollment year and the month of sample entry, a binary indicator for NH use in the 90 days prior to sample entry, and indicators for state. We used the estimated regression coefficients for the matched comparison group to predict what the mean monthly expenditures for PACE sample members would have been had they begun using HCBS or entered a NH instead of receiving PACE services, when their followup period began. For each six-month interval of Medicare or Medicaid expenditures, we insert the mean baseline characteristics of PACE sample members into the regression equation that was estimated for the members of the matched comparison group. The difference between actual expenditures and predicted expenditures represents the estimated effect of PACE on Medicare and Medicaid expenditures. Standard errors for the regression-adjusted comparisons of actual and predicted expenditures for PACE enrollees had to be separately computed because they involve a comparison of mean expenditures for PACE participants to a regression-based prediction of mean expenditures that PACE participants would have incurred had they started using HCBS or entered a NH instead of enrolling in PACE. These calculations are described in Appendix B.
Given that the earliest enrollment in our study sample occurs in July 2006, we are able to follow Medicare expenditures for a maximum of 66 months (through 2011) after program entry, but are able to follow Medicaid expenditures for only a maximum of 42 months (through 2009). Thus, we examine Medicare expenditures for a total of 11 six-month intervals and examine Medicaid expenditures for a total of seven intervals.
Our main findings from the expenditures analysis is based on comparing PACE enrollees to the matched comparison group comprised of both HCBS waiver enrollees and NH entrants. However, we repeat the expenditures analysis with the second comparison group comprised of matched HCBS waiver enrollees alone and discuss those results separately.
We examine the cumulative mortality rates for PACE and the matched comparison group (HCBS enrollees and NH entrants) over successive six-month intervals from sample entry. The denominators for the mortality rate calculations were fixed in both groups as the number of treatment and matched comparison group members in the final matched sample for the first six intervals (since all sample members had at least 36 months of followup, with mortality data available through 2011), and restricting the samples in later intervals to those whose period of sample entry would allow us to observe their mortality in a particular interval.13 For each interval, we estimate a separate logistic regression with the mortality indicator as the dependent variable and with similar control variables as for the expenditures analysis, plus a binary indicator identifying the observation as a PACE enrollee or comparison group member. We obtain regression-adjusted predictions of mortality for both PACE enrollees and their matched counterparts in the comparison group. For the mortality analysis, we also compare PACE enrollees to the second matched comparison group comprised of matched HCBS waiver enrollees alone, that is, we present separate mortality results from using both comparison group strategies. Observations for matched comparison group members are weighted using the matching weights only.
Nursing Home Utilization
We also examine use of NH services for both PACE and matched comparison group members, using both matched comparison samples. However, for the NH utilization outcomes, we lead with the results obtained from using the matched comparison group comprised of HCBS waiver enrollees alone. As mentioned above, since NH entrants have high NH utilization by default, using the matched comparison group comprised of both NH entrants and HCBS waiver enrollees would skew the findings in favor of PACE. As a sensitivity test, we do present findings from using this alternative comparison group strategy, and the results are in line with our expectation of being highly favorable for PACE.
For this analysis involving PACE and matched HCBS waiver enrollees, we obtained information on NH utilization from the MDS Timeline file. Specifically, we compare four outcomes--(1) any use of the NH; (2) proportion of days in a NH; (3) at least 30 days (not necessarily consecutive) in a NH; and (4) at least 90 days (not necessarily consecutive) in a NH--for PACE and matched HCBS members over successive six-month intervals from enrollment or sample entry through 2009, that is, for a maximum of 42 months of followup or a total of seven intervals. The last two outcomes (at least 30 or 90 days in a NH) are defined conditional on a sample member being alive during an interval. For each interval, we estimate a separate linear regression for proportion of days in a NH, and separate logistic regressions for the three other binary outcome variables, and obtain regression-adjusted predictions of NH utilizationfor both treatment and matched comparison group members. The control variables in the regression model are same as those used for the mortality analysis. These regressions use the eligibility weights for the treatment group, and eligibility weights multiplied by the matching weights for matched HCBS members, as for the analysis involving expenditures.