The effects of predetermined payment rates for Medicare home healthcare

Health Services Research, Oct, 1997 by Randall Brown, Barbara Phillips, Christine Bishop, Craig Thornton, Grant Ritter, Amy Klein, Peter Schochet, Kathleen Skwara

The patient-level analyses relied on large samples of patients (about 30,000 for each demonstration year), for whom Medicare claims data were available. Beneficiaries' number of home health visits during their episode was regressed on a binary variable indicating whether the beneficiary was being treated by a treatment or control group agency, plus a large number of control variables. Key control variables denoted the beneficiary's diagnoses, comorbidities, medical conditions, functional limitations, and severity of illness - all measured at the time of their admission to the home health agency - plus their use of Medicare-covered services in the year prior to admission. Other control variables in the model included agency and area characteristics used in the agency-level regressions, plus indicators of the availability of informal care. Inferences about whether the program affected the dependent variables examined were based on tests of the coefficients on the treatment-control indicator. Similar models were used to estimate program effects on dependent variables not reported on here, including beneficiaries' use and costs of various Medicare-covered services, use of home and community-based services not covered by Medicare, and quality of care indicators.

Throughout the analyses of patient-level data, observations were weighted so that each agency had equal representation. This weighting reflected the fact that the intervention was at the agency level and ensured that results were not dominated by a large agency. (A single agency accounted for 40 percent of all episodes delivered by treatment group agencies.) All t-statistics displayed have been adjusted to account for the underestimation of standard errors that arises in the patient-level analysis due to failure of most statistical packages to account for weighting and clustering of the sample. The adjustment, based on results obtained from the software package SUDAAN (Shah, Barnwell, and Bieler 1996) for a representative set of outcome measures, resulted in a tripling of the standard errors relative to those from a simple random sample of the same size.

DATA SOURCES FOR IMPACT ANALYSIS

The evaluation required data on outcomes at both the agency and patient level. Agencies' annual Medicare cost reports were the key source of data on agency-level outcomes. These reports provided information on the number and average cost of visits, by type of visit, plus agency revenues from Medicare and non-Medicare sources. Information on patient outcomes was drawn from a variety of sources, including demonstration claims and Medicare claims files for all patients in participating agencies, as well as patient surveys conducted for a sample of patients at discharge and six months after discharge. The claims data were used to construct episodes of care (defined as the period between the start of care and end of care dates on the demonstration home health claims) and services used during and after those episodes. The discharge and follow-up surveys provided information on patient satisfaction with care, functioning, other measures of patients' well-being, and use of non-Medicare home care services, both during and after the episode. A separate survey conducted at three weeks after admission provided information on visit length. Data collected by the New England Research Institute to monitor the quality of care provided measures of quality for our analyses as well, including identified quality assurance problems and functioning measures.