Introduction: The ability to predict discharge destination early in the hospital course for patients admitted with acute ischemic stroke can improve efficiency of discharge planning, identify outliers for evaluation, and estimate future resource use.
Methods: A prediction model for discharge destination was developed using multinomial logistic regression with the response variable categories: Home, Inpatient Rehabilitation Facility (IRF), and either Skilled Nursing Facility or Long-Term Acute Care (SNF/LTAC). Data used to build the model was electronically extracted from the electronic health records of 1,364 patients admitted with acute ischemic stroke to a tertiary care hospital from Dec 6, 2008 to Sept 4, 2012. There were 34 candidate predictors identified based on the literature and clinical expertise. Only data available in the first 2 days of admission were used in the analyses. Multiple imputation was used to handle missing data. Variables retained in the model were selected using the nested bootstrap procedure, which resulted in 2,000 datasets that each underwent backward stepwise selection. Calibration of the final model was assessed graphically. The polytomous discrimination index (PDI) was computed to assess discriminative validity.
Results: The final model contained 7 predictors: living with family, requiring help prior to admission (both obtained from nursing flowsheets), history of heart failure, first systolic blood pressure, admission NIHSS, and antibiotic use. PDI was 0.71. Pairwise c-statistics were: Home vs IRF 0.83, Home vs SNF/LTAC 0.91, IRF vs. SNF/LTAC 0.79. (See also Figure)
Conclusion: Discharge destination of patients admitted with acute ischemic stroke can be predicted with good discrimination using 7 variables available within the electronic health record. Once validated in an external population, this prediction model can be used to help improve the efficiency of hospital stays.