Effects of Risk Adjustment for Groups of Variables: Sicker, Poorer, Readmitted to the Hospital

• Activity Number: 174 - Statistical Methods to Assess the Performance of Health Providers
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 10:30 AM to 12:20 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Alan M. Zaslavsky* and Eric T. Roberts and J. Michael McWilliams
• Companies: Harvard Medical School and University of Pittsburgh and Harvard University Medical School
• Keywords: risk adjustment; hospital quality; readmissions; health care; regression

Abstract

Performance measures for health care units such as hospitals may be adjusted to remove the effect of risk factors (patient attributes not caused by the unit, like comorbid diagnoses or SES). The mean square adjustment for a single covariate in a linear model is the product of its coefficient and the variation of the covariate means across units of assessment. When adjuster variables are entered in groups, defined by timely availability, theoretical basis, strength of association, and/or between-unit variation, we analyze shifts in adjusted performance measures for all the units as groups are added. The estimates are functions of covariance matrices of adjustment and outcome variables within and between units, facilitating analysis when adjustment data are accessed only within a secure data repository. Cholesky factorization orthogonally decomposes the sequence of incremental adjustments for groups of candidate adjuster variables, like factor analysis aiding interpretation. We illustrate analysis of adjustment of rates of mortality and/or early readmission for the Hospital Readmissions Reduction Program (HRRP), an important but controversial national quality-improvement initiative.