Support and Services at Home (SASH) Evaluation: First Annual Report. APPENDIX C: Quantitative Data and Methods


C.1. Data

RTI receives Medicare claims on a quarterly basis from the Actuarial Research Corporation (ARC). ARC takes prospective TAP claims from CMS on a monthly basis and quarterly creates netted claims files for analytic purposes. With each new quarter of data, ARC updates past files with claims processed after the previous cutoff date. This update process covers a two-year run-out period. The ARC files contain the Medicare claims for demonstration and comparison beneficiaries from January 2010 forward. Prior to 2010, Medicare claims are pulled from the Data Extract System by RTI analysts.

Property-level data used in this report come from two separate HUD data bases. TRACS is the data base for all multi-family properties (Section 202, Section 236, Section 8, etc.), and the PIC is the data base for all public properties. Medicare claims and HUD records were merged using the person's SSN. If SSN was missing, merges were attempted using first and last name and date of birth.

RTI also received permission from the State of Vermont to receive the Medicare identification number for beneficiaries participating in the SASH program. Along with the start date of their particular SASH property, this information was then merged with their Medicare claims data through the use a cross-referenced Health Insurance Claim number.

C.2. Comparison Groups

The analysis considers the following two between-group comparisons:

  • SASH/MAPCP Demonstration beneficiaries versus non-SASH/MAPCP Demonstration beneficiaries: this comparison yields estimates of the SASH program effect (among MAPCP Demonstration beneficiaries).

  • SASH/MAPCPDemonstrationbeneficiaries versus non-SASH/non-MAPCP Demonstration beneficiaries: this comparison yields estimates of the combined SASH and MAPCP Demonstration effect.

Similar to the SASH intervention group, the non-SASH/MAPCP Demonstration comparison group is drawn from HUD housing within Vermont. Like the SASH sites, these properties are distributed across the state and exhibit a similar range of housing units per property. The non-SASH/non-MAPCP Demonstration comparison group is drawn from properties in 16 counties in upstate New York. Located in the state's northeast quadrant, these counties form the closest non-MAPCP Demonstration area to Vermont.

C.3. Weights

All quantitative analysis in this report uses a beneficiary-level weight that is a function of the person's PS (for beneficiaries in the comparison group) and their quarterly Medicare eligibility during the demonstration period. The PS is the probability of participating in the SASH program, conditional on beneficiary and property characteristics. PSs are estimated from a logistic regression that uses the indicator for SASH program participation as the dependent variable and beneficiary and property characteristics as independent variables. For beneficiaries in the comparison group, the PS contributes a factor PS/(1-PS) to the final regression weight. The purpose of including this factor is to better align the comparison group with the intervention group in terms of beneficiary-level and property-level characteristics. As such, it reduces the confounding bias that can result from using a non-randomized comparison group. Covariates in the PS model include the following beneficiary-level and property-level characteristics:

  • Beneficiary-level: age (continuous), female, non-White, disabled, Medicaid dual-eligible, end-stage renal disease (ESRD), HCC risk score, Charlson score, household income, household size, length of occupancy, rent amount.

  • Property-level: subsidy type, number of units, percent of elderly residents.

A beneficiary's quarterly eligibility was measured as the fraction of days (out of 90) they met the following criteria: (1) they were a Medicare FFS beneficiary with Medicare as the primary payer; (2) they were attributed to a practice in the MAPCP Demonstration or comparison groups; and (3) they resided in Vermont or upstate New York. This quarterly eligibility fraction was then multiplied by PS weights to create the final analytic weight.

C.4. Regression Analysis

In the following notation, i is an index of the beneficiary and t the quarterly period. The outcome is denoted by Yit, Xit is set of beneficiary-level and property-level covariates included in the model as controls, and εit is an error term.

For the Medicare payment outcomes we use the following linear DID model to estimate the impact of the SASH program.

  Y subscript it = Alpha subscript 0 + Alpha subscript t + Beta subscript 1 I subscript i + Beta subscript 2 D subscript PILOT,it + Beta subscript 3 D subscript ATT,it + Beta subscript 4 D subscript SASH,it = Beta subscript 5 X subscript it + Epsilon subscript it     (B.1)  

The variables αt (t=1,2,…) are a set of time fixed effects for each quarter in the sampling period (the αt's are estimated by including as independent variables a set of indicator variables for each quarter). The variable Ii(=0,1) is an indicator for membership in the SASH intervention group; it equals 1 for beneficiaries who are SASH participants and 0 for beneficiaries in the comparison group. The variable DPILOT,it (=0,1) allows for a change in the average outcome starting in the quarter when a beneficiary's practice started participating in the Blueprint for Health program. The variable DATT,it (=0,1) is an indicator that equals 1 in the quarter when a beneficiary was first attributed to a practice participating in the MAPCP Demonstration, and all quarters thereafter. The variable DSASH,it (=0,1) is an indicator for SASH participation. For beneficiaries in the intervention group, it switches from 0 to 1 in the quarter when the property in which they reside started implementing the SASH program, and remains 1 thereafter. It is equal to zero in all quarters for beneficiaries in the comparison group. The coefficient for DSASH,it4) is the measure of the SASH program effect in terms of the change in the level of the outcome--relative to the comparison group--after implementation of the SASH program. It is the coefficient that is presented in Table 4.

The linear specification in Equation B.1 is less appropriate for the utilization outcomes which are count variables. For these outcomes we estimate a negative binomial model instead.2 The same parameters as on the right-hand side of Equation B.1 appear in this model, but the impact of the SASH program on utilization is calculated as follows.3

  t = exp(Alpha subscript 0 + Alpha subscript t + Beta subscript 1 + Beta subscript 2 + Beta subscript 3 + Beta subscript 5 X subscript ij)*[exp(Beta subscript 4) - 1]4     (B.2)  

The parameterτ measures the increase (β4>0) or decrease (β4<0) in utilization during the period of SASH program participation, among SASH participants relative to beneficiaries in the comparison group. We multiplyτ by 1,000 to express the SASH program effect in terms of a rate per 1,000 Medicare FFS beneficiaries.

View full report


"SASH1.pdf" (pdf, 432.46Kb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®