Identifying and Accommodating Statistical Outliers When Setting Prospective Payment Rates for Inpatient Rehabilitation Facilities


  • Susan M. Paddock,

    Search for more papers by this author
    • Address correspondence to Susan M. Paddock, RAND, 1700 Main St., Santa Monica, CA 90407-2138. Barbara O. Wynn and Melinda Beeuwkes Buntin are with RAND, Arlington, VA. Grace M. Carter is with RAND, Santa Monica, CA.

  • Barbara O. Wynn,

  • Grace M. Carter,

  • Melinda Beeuwkes Buntin

  • This work was funded by the Centers for Medicare and Medicaid Services (Contract #500-95-0056).


Objective. To demonstrate how a Bayesian outlier accommodation model identifies and accommodates statistical outlier hospitals when developing facility payment adjustments for Medicare's prospective payment system for inpatient rehabilitation care.

Data Sources/Study Setting. Administrative data on costs and facility characteristics of inpatient rehabilitation facilities (IRFs) for calendar years 1998 and 1999.

Study Design. Compare standard linear regression and the Bayesian outlier accommodation model for developing facility payment adjustors for a prospective payment system.

Data Collection. Variables describing facility average cost per case and facility characteristics were derived from several administrative data sources.

Principal Findings. Evidence was found of non-normality of regression errors in the data used to develop facility payment adjustments for the inpatient rehabilitation facilities prospective payment system (IRF PPS). The Bayesian outlier accommodation model is shown to be appropriate for these data, but the model is largely consistent with the standard linear regression used in the development of the IRF PPS payment adjustors.

Conclusions. The Bayesian outlier accommodation model is more robust to statistical outlier IRFs than standard linear regression for developing facility payment adjustments. It also allows for easy interpretation of model parameters, making it a viable policy alternative to standard regression in setting payment rates.