Keywords: performance measurement, casemix adjustment, readmissions
The importance of adjusting health care performance measures for patient-level clinical factors has long been recognized, but recent years have seen a surge of interest in adjusting for social risk factors as well. This new emphasis calls for statistical methods that can reveal patterns in adjustments with many predictor variables, both to aid interpretation of the adjustments and to discern their contributions to equity and accuracy. Our statistical approach proceeds by examining the incremental effects of adjustments for additional variables on the n-dimensional vector of adjusted unit scores. Effects with a large component in the direction that separates high- and low-quality units may shrink or expand the estimated spread of adjusted quality. Conversely adjustment effects that are largely orthogonal to that direction may reorder the scores of units without affecting the spread of the scores; it then may be informative to identify directions in the orthogonal space that correspond to interpretable subsets of the predictor variables. This work is motivated by research on adjustment of hospital readmissions measures inspired by the IMPACT legislation.