499. Determining Risk of Clostridium difficile Using Electronic Health Record (EHR) Data

BACKGROUND: Hospitals may now be penalized for Clostridium difficile infection diagnosed after hospital day 3, which are classified as “hospital-onset” (HO) regardless of existence of true disease. Highly sensitive PCR-based testing has made this additionally problematic. As part of a C. difficile testing stewardship initiative, we sought to validate a C. difficile risk scoring tool using existing EHR data. METHODS: We conducted this study in a 2-hospital, >1,100-bed community-based academic healthcare system in northern Delaware. After piloting a paper-based Cdiff risk scoring tool, intended for use after hospital day 3 to discourage testing in low-risk patients, we created a C. difficile-specific analytic application using the Health Catalyst clinical analytics platform over the existing data warehouse (Cerner). The scoring tool was modified from those in the literature and included patient age, body mass index and albumin (if available); prior hospitalization or long-term care facility stay (within 90 days); and receipt of any fluoroquinolone, cephalosporin or piperacillin/tazobactam (within 30 days). Only antibiotics received within our system were included. Using data from September 2015–April 2018, we calculated a receiver operating characteristic (ROC) curve for the risk score’s ability to predict a positive HO Cdiff PCR. To increase specificity, we defined “true positive” C. difficile as +PCR tests occurring in patients with ≥3 diarrheal episodes and no laxative use during the 48h prior to testing, and either WBC >12 or temperature >38C 24 hours before or after the +PCR. RESULTS: During the study period the health system had 150,554 inpatient encounters, of which 411 had positive PCR tests for HO C. difficile and 138 (33% of all PCR+) met our definition of “true positive.” The C. difficile risk stratification tool demonstrated an area under the ROC (AUC) of 0.77 (95% CI 0.75–0.79) to predict a +PCR test (Figure 1), with very similar results (AUC 0.76, 95% CI 0.73–0.80) if the outcome was “true positive” C. difficile (Figure 2). CONCLUSION: Using readily available EHR data, we developed a C. difficile risk stratification tool that was able to predict C. difficile positivity with reasonable distinction, but did not differentiate colonization from true illness. The next step is to further refine the tool to better predict true C. difficile illness. DISCLOSURES: All authors: No reported disclosures.