A Comparison of the Explanatory Power of Two Approaches to the Prediction of Post Acute Care Resources Use

Submitted by capri.sims@cgi… on Thu, 11/07/2019 - 02:05
Title
A Comparison of the Explanatory Power of Two Approaches to the Prediction of Post Acute Care Resources Use
Authors
Vertrees,James; Averill, Richard.; Eisenhandle, Jon; Quain, Anthony; Switalski, James
# of Pages
32

Abstract:

A risk model is used to adjust a patient centered episode (PCE)for the impact of chronic diseases. In this study, a patient centered episode is constructed with a hospitalization and the associated post acute care. A patient's chronic disease burden can be measured using pre-existing conditions at the time of hospitalization. Two risk adjustment approaches are available: 1) a categorical clinical model such as the Clinical Risk Groups (CRGs), or 2) a statistical regression-based model such as the Diagnostic Cost Group Hierarchical Clinical Conditions (HCCs). This study compares episode costs adjusted with the CRGs and the HCCs in terms of their ability to "explain" the variation in post acute care services following a hospitalization as measured by the standard R2 statistic. The important findings are: For Medicare charges CRGs have a substantially higher R2 across all windows For Medicare charges the R2 for both CRGs and HCCs increases as the length of the window increases but for payments the R2 is relatively flat as the length of the window increases For both CRGs and HCCs the R2 drops substantially when readmissions are added to the post acute care bundle For Medicare payments CRGs have substantially higher R2 for post acute care bundles composed of hospital outpatient, physician and other part B, DME, and home health. However when skilled nursing facility and hospice are added to the post acute care bundle HCCs have a slightly higher R2 The correlation coefficient for the predicted CRG and HCC values are 0.612-0.680 for charges and 0.715-0.769 for payments depending on the episode window.