S. O. Dennis1, J. K. Canner2, D. T. Efron2, E. R. Haut2, J. V. Sakran2, K. A. Stevens2, C. Jones2 1East Carolina University Brody School Of Medicine,Greenville, NC, USA 2Johns Hopkins University School Of Medicine,Department Of Surgery,Baltimore, MD, USA
Introduction: Trauma readmission rates are used to assess quality of care, hence identifying risk factors for readmission has become a priority. Prior studies have had disparate results and result in few predictors of readmission. We sought to examine a large data set to determine risk factors increasing odds of readmission after trauma.
Methods: We used Maryland’s Health Services Cost Review Commission (HSCRC) Inpatient Data Set to identify injured patients admitted to acute care hospitals from 2013-2015; the HSCRC includes unique identifiers to track patient admissions statewide across institutions. We compared patients readmitted within 30 days of discharge from an initial trauma admission to those not readmitted. We included variables previously identified as potentially affecting readmission (Table). After univariable comparison, we included potentially statistically significant (p < 0.1) factors not collinear with others in a multiple logistic regression analysis to identify those independently associated with readmission (p < 0.05).
Results: We identified 300,925 index trauma admissions. 50,309 (17%) were followed by a readmission; 14,724 (29%) of these were admissions to a different hospital. All variables evaluated except injury mechanism were statistically significant on univariable and multivariable analyses, each independently associated with readmission risk (Table). For this complex model, the area under the receiver operating characteristic curve is only 0.61, suggesting even the inclusion of all variables is inadequate for predicting readmission.
Conclusion: These data demonstrate a small number of variables will not adequately predict readmissions; rather, a broad swath of variables is needed to quantify readmission risk. Future formulations should use a wider range of available data and may need to be combined with advanced techniques to determine a patient’s individual risk of readmission.
