Predictability of a 30-Day Readmission Risk Model in Kidney Transplant Recipients. Abstract# 637

Background: 30-day readmissions (30DRA) are a highly scrutinized surrogate measure of healthcare quality, with Medicare penalizing hospitals for high 30DRAs for certain admission types. Thus, understanding important factors associated with 30DRA may lead to improved risk assessment, resource allocation and reduced events. Methods: This was a large-scale retrospective analysis with the primary aim of developing a predictive risk model for 30DRA. Adult kidney transplant (KTX) recipients from 2005-12 were included; baseline donor and recipient data was collected, along with detailed in-hospital clinical and cost data. Risk models were developed using backward conditional binary logistic regression and tested for predictability using ROC curves. Results: 1,176 KTX patients were included, of which 130 (11%) had 30DRA. Univariate risk factors for 30DRA included recipient comorbidities, donor factors, transplant factors and index hospitalization data. The initial risk model included 27-variables and demonstrated significant predictability (C-statistic 0.77, NPV 71%, PPV 67%, Figure 1). Through backwards elimination, this model was reduced to a 9-variable risk model (Table 1), which performed as well as the 27-variable model (C-statistic 0.75, NPV 66%, PPV 67%, Figure 2). The most significant predictors of 30DRA included recipient sociodemographics, induction therapy, DGF and a number of index hospitalization data (blood pressures and costs), while donor factors were not significantly associated with 30DRA. Conclusion: 30-Day readmissions are common after KTX. A risk model that includes recipient sociodemographics, induction therapy, DGF and index hospitalization data is significantly predictive of readmission. This risk model, if validated, may be utilized to focus resources on high-risk patients and reduce these events.