LAUSR

244 Background: Up to 50% of cancer patients experience an unplanned hospitalization. This represents opportunities for both improvements in quality of care and utilization of low value resources. Hospital readmission scores, to best target post-discharge navigation and follow-up, are needed yet not widely available. Methods: We conducted a retrospective cohort study among oncology patients admitted to Duke University Hospitals in 2015. The readmissions risk test model was built using multivariate analysis to identify and ‘weight’ four key readmission predictors. The model was subsequently analyzed in a validation data set using 2016 & 2017 admission data. Results: Of the 4987 admissions in 2015, 55% were male, 73% were Caucasian, and mean age was 61 (SD 14.1 years). Common cancers were GI (31%), Thoracic (27%) and GU (24%). Factors used to build the readmission predictor model based on the relative Odds Ratios were race (1), length of stay (1), discharge disposition (2), and previous admission with < 90 days (2) or 91-180 days (1). The patient cohort used to test the model had 1926 admissions and each were assigned a point value according to the model specifications, see Figure 1. We found significant differences in risk for readmission among differences in scores (p < 0.05). We have subsequently launched a post-discharge navigation program, with frequency of contacts dictated by readmission risk from the model. Conclusions: Predictor models in the oncology setting can identify combinations of factors that are associated with readmission. Programs that integrate such models may identify cancer patients at high risk of readmission and thus tailor a personalized approach to preventing a subsequent admission.[Table: see text]