External validation of predictive scores for mortality following Clostridium difficile infection

Background The burden of Clostridium difficile infection (CDI) has increased in the last decade, with more adverse outcomes and related mortality. Although many predictive scores were developed, few were validated and their performances were sub-optimal. We conducted an external validation study of predictive scores or models for mortality in CDI. Methods Published predictive tools were identified through a systematic review. We included those reporting at least an internal validation approach. A multicenter prospective cohort of 1380 adults with confirmed CDI enrolled in two Canadian provinces was used for external validation. Most cases were elderly (median age 71), had a healthcare facility-associated CDI (90%), and 52% were infected by NAP1/BI/027 strains. All-cause 30-day death occurred in 12% of patients. The performance of each scoring system was analyzed using individual primary outcomes. Results We identified two scores which performances (95% CI) are shown in the table. Both had low sensitivity and PPV, moderate specificity and NPV, and similar AUC/ROC (0.66 vs. 0.77 in the derivation cohort, and 0.69 vs. 0.75 respectively). One predictive model for 30 days all-cause mortality (Archbald-Pannone 2015, including Charlson score, WBC, BUN, diagnosis in ICU, and delirium*) was associated with only 5% increase in odds of death (crude OR = 1.05 (1.03–1.06)) with an AUC of 0.74 (0.7–0.8). Conclusion The predictive models of CDI mortality evaluated in our study have limitations in their methods and showed moderate performances in a validation cohort consisting of a majority of CDI caused by NAP1 strains. An accurate predictive tool is needed to guide clinicians in the management of CDI to prevent adverse outcomes. Disclosures J. Powis, Merck: Grant Investigator, Research grant; GSK: Grant Investigator, Research grant; Roche: Grant Investigator, Research grant; Synthetic Biologicals: Investigator, Research grant