BackgroundVaccine efficacy trials against C. difficile infection (CDI) require a study population with an observed event rate that supports feasibility and timely study completion. The goal of this study was to develop a predictive algorithm to identify patients at high risk of developing CDI over the next 2–12 months for enrollment in a vaccine clinical trial.MethodsWe conducted a two-stage retrospective study of patients within the US Department of Veterans Affairs Health system between January 1, 2009 and December 31, 2013. Included patients were age 50 years or older, had at least one visit in each of the 2 years (2007 and 2008) prior to the study, and had no CDI in the 12 months prior to and the first month of the study. In stage 1, we used a case-cohort design to identify factors (patient-month level) for predicting the risk of CDI in months 2–12. The stage 1 algorithm was then applied to the VA population and a calculated risk score generated at each patient visit (inpatient or outpatient). Patients were included in the stage 2 cohort at the first time their stage 1 risk score crossed the designated risk threshold. The stage 2 algorithm incorporated additional factors including medications and laboratory variables. Multivariable logistic regression with elastic net regularization was used to select final predictors in both stage 1 and stage 2 algorithms. The final algorithm consisted of sequential application of predictive algorithms in stage 1 and 2. Performance was measured using the positive predictive value (PPV) and the area under the curve (AUC).ResultsRisk factors in the final algorithm included baseline comorbidity score, acute renal failure, sepsis, UTI or pneumonia in the current month, UTI in the previous month, having an albumin or hematocrit test performed, high-risk antibiotics in the previous 3 months, hemodialysis in the last month, race, and the number and duration of acute and long-term care visits in the last month. The final algorithm resulted in an AUC of 0.69 and a PPV of 3.4%, largely enriching for the risk of CDI in months 2–12.ConclusionWe developed a predictive algorithm to identify a patient population with increased risk of CDI over the next 2–12 months. Predictive models have the potential to improve enrollment and feasibility of clinical trials.Disclosures V. Stevens, Pfizer, Inc.: Grant Investigator, Research grant. E. Russo, Pfizer, Inc.: Grant Investigator, Research grant. Y. Young-Xu, Pfizer, Inc.: Grant Investigator, Research grant. Y. Zhang, Pfizer, Inc.: Investigator, Research grant. C. Zhang, Pfizer, Inc.: Investigator, Research grant. H. Yu, Pfizer, Inc.: Employee and Shareholder, Salary. B. Cai, Pfizer, Inc.: Employee, Salary. E. Gonzalez, Pfizer, Inc.: Employee, Salary. D. N. Gerding, Merck: Scientific Advisor, Consulting fee. Actelion: Scientific Advisor, Consulting fee. DaVolterra: Scientific Advisor, Consulting fee. Summit: Scientific Advisor, Consulting fee. Rebiotix: Medical Officer and Scientific Advisor, Consulting fee. Pfizer: Consultant, Consulting fee. MGB Pharma: Consultant, Consulting fee. sanofi pasteur: Consultant, Consulting fee. Seres: Investigator, Research grant. CDC: Investigator, Research grant. US Dept VA: Investigator, Research grant. Treatment/Prevention of C. difficile: Patent Holder, no license or royalties. J. Lawrence, Pfizer, Inc.: Employee, Salary. M. Samore, Pfizer, Inc.: Grant Investigator, Research grant.