Abstract W P281: Improving Prediction of Readmission After Ischemic Stroke: A Nationwide Retrospective Cohort Analysis
Background: Previous models for predicting readmission after ischemic stroke rely solely on administrative claims. The Minimum Data Set (MDS) is a nationwide clinical database required for all U.S. nursing home residents. We hypothesized MDS variables would improve prediction of 30-day acute hospital readmission after ischemic stroke.
Methods: Medicare inpatient claims for the year 2008 were used to identify ischemic stroke cases over age 65 as defined by ICD-9 principal diagnosis codes 433.x, 434.x, and 436.x. Unique individuals were linked to full MDS records. Logistic regression was used to construct risk-adjusted models of unplanned readmission 30 days post acute care hospitalization. Covariates were derived from Medicare Part A claims and clinically relevant MDS variables. Model performance was assessed through construction of receiver operator characteristic (ROC) curves.
Results: Among 252,569 ischemic stroke admissions, there were 87,094 (36%) individuals admitted to nursing homes. The analytical sample was composed of 39,178 (14%) patients directly admitted to nursing homes. At 30-days post acute hospital discharge there were 11,648 (21%) readmissions and 6,394 (12%) deaths. Independent predictors of readmission included feeding tube (OR 1.20, CI 1.06-1.35), congestive heart failure (OR 1.18, CI 1.08-1.29), chronic obstructive pulmonary disease (OR 1.26, CI 1.15-1.39), renal disease (OR 1.26, CI 1.12-1.43), and pressure ulcers (OR 1.34, CI 1.20-1.51). Clostridium difficile infection was most strongly associated with readmission (OR 1.44, CI 1.05-1.97). Hemiparesis (34%) and aphasia (16%) were not significant predictors of 30-day readmission. Area under the ROC curve for predicting readmission at 30 days was 0.65.
Conclusions: Nursing home admission is required for a significant proportion of stroke survivors over age 65. The addition of multiple individual clinical variables derived from the MDS resulted in a modest improvement in area under the ROC curve (.65) compared to previous models utilizing only Medicare Part A data (.60). The prediction model presented here provides novel targets for readmission reduction programs among stroke patients cared for in nursing homes.
Author Disclosures: C.R. Fehnel: Research Grant; Significant; Surdna Foundation Fellowship Award - Brown University Center for Gerontology and Health Care Research. Y. Lee: None. V. Mor: Research Grant; Significant; National Institute on Aging grant PO1AG027296, Commonwealth Fund. Honoraria; Significant; Alliance for Nursing Home Quality, Yale University, Rutgers University, Congressional Budget Office. Expert Witness; Significant; U.S. Senate Health and Aging Committee on End of Life Care. Ownership Interest; Modest; PointRight Inc., NaviHealth Inc.. Consultant/Advisory Board; Significant; Tufts Health Plan Foundation, Home and Hospice Care of Rhode Island, Jewish Alliance of Rhode Island, PointRight Inc., NaviHealth Inc., hcr-Manorcare, Academy Health, Alliance for Nursing Home Quality, Econometric Inc, Research Triangle Inc, AARP Public Policy Center.
- © 2015 by American Heart Association, Inc.