M. Graham1, P. Parikh1,2, S. Hirpara2, M. McCarthy1, P. P. Parikh1 1Wright State University,Department Of Surgery,Dayton, OH, USA 2Wright State University,Department Of Biomedical, Industrial, And Human Factors Engineering,Dayton, OH, USA
Introduction: Delay in discharge planning could result in extended length of stay leading to increased hospital costs, ineffective utilization of resources, and delays in rehabilitation treatment in trauma patients. Limited work has been done in developing models predicting discharge disposition in trauma patients. These models are developed using a single institution data and have not be demonstrated to be generalizable. The objective of this study is to develop a predictive model using the National Trauma Data Bank (NTDB) and evaluate its generalizability on data from a Level I trauma center.
Methods: NTDB data from 2015 were used to build and validate a binary logistic regression model using derivation-validation (i.e., train-test) approach to predict patient disposition location (home vs nonhome) upon admission. Patient demographics and clinical variables available at the time of admission were considered in the analysis. A Mann-Whiney U-test was used to compare patient parameters. The regression model was then converted into a 20-point score using an optimization-based approach. An appropriate threshold was selected to achieve a score with a sensitivity of >0.80 and specificity of >0.50. The generalizability of this score was then evaluated on the trauma registry data at our Level I trauma center in Midwest US.
Results:A total of 558,599 cases in the NTDB were included in the study, out of which, 178,666 (31.98%) went to a nonhome location and 379,933 (68.02%) patients went home. The average age of patients with a nonhome disposition compared to home disposition was significantly higher (68.11 ± 20.69 years vs. 43.23 ± 23.09 years; p<0.001) and had more severe injuries measured using the ISS (11.26 ± 8.25 vs 8.04 ± 6.28; p<0.001). Increased age, female sex, higher ISS, and the comorbidities of cancer, cardiovascular, coagulopathy, hepatic, neurological, psychiatric, renal, substance abuse, and diabetes were independent predictors of nonhome discharge. The logistic regression model’s AUC was 0.83; the score achieved a correlation of 0.94 with the predicted probabilities from the regression model. A threshold value of 4 or higher indicated higher likelihood of nonhome discharge; this threshold resulted in a sensitivity of 0.86 and specificity of 0.62 on NTDB validation data (n=167,580). The score generalized well on the insitutional data (n=3,384) obtained from trauma registry of our Level I Trauma Center; sensitivity of 0.85 and specificity of 0.60.
Conclusion:A model and a score developed using NTDB could be implemented at a Level I trauma center to predict upon admission a trauma patient’s discharge disposition location, home or nonhome. This score can aid in early hospital preparation for patients predicted to be discharged to a nonhome location yielding a smoother transition, increased satisfaction, effective utilization of hospital resources, and potentially decrease total operating costs.