Factors associated with coronary artery bypass grafting excess readmission ratios.

Abstract

Background: The Hospital Readmissions Reduction Program determines Medicare readmission penalties through risk-adjusted excess readmissions ratios. This study uses interpretable machine learning to identify associations with coronary artery bypass grafting excess readmissions ratios.

Methods: Florida Agency for Healthcare Administration database (2017-2019) was queried for patients with Medicare who underwent coronary artery bypass grafting and linked to Hospital Readmissions Reduction Program and RAND Hospital data. Univariable analysis compared hospitals above compared with below the median excess readmissions ratios. LightGBM and XGBoost modeled excess readmissions ratios using patient data. SHapley Additive exPlanations were used for interpretation.

Results: The cohort had 7,776 patients from 64 hospitals. Hospitals above the median excess readmissions ratios had more emergent admissions (35.4 vs 29.8%, P < .01), greater mortality (2.1 vs 1.3%, P < .01), and more expensive postoperative care charges (P < .01). Models had mean absolute errors <0.06 and R² > 0.79. SHapley Additive exPlanations revealed charges and length of stay as most influential. Greater postoperative care charges and shorter length of stay were associated with greater excess readmissions ratios.

Conclusion: Hospitals with greater excess readmissions ratios charge more for postoperative care. Whether this association is caused by hospital differences alone or also by unadjusted differences in patient severity and functional status is unknown. The association of shorter length of stay with greater excess readmissions ratios could suggest that hospitals with greater excess readmissions ratios discharge patients too early. Future research should investigate the causes of these associations and their impact on excess readmissions ratios to ensure that hospitals are not unfairly penalized.