Performance of Published Clostridium difficile Severity Scoring Systems in an External Cohort

Background: Clostridium difficileinfection (CDI) can cause severe disease and death. Several objective scoring systems have been promoted to risk stratify individuals for development of severe disease, but have received little external validation. This study applies and compares published scoring systems within an external cohort to assess their performance.

Methods: A literature search on predictive scores for severe CDI revealed nine candidate models with objective criteria, amenable to retrospective external validation. These scores were applied to a pre-existing cohort of inpatients with CDI at the University of Michigan Healthcare System. The primary outcome of severe CDI was defined as any of the following attributable to CDI within 30 days of onset: death, colectomy, or intensive care unit admission. Metrics assessed included sensitivity, specificity, positive and negative predictive values (PPV & NPV), discriminant function (DF), and net reclassification index (NRI) compared to the IDSA (Cohen et al. 2010) severity score (a 1.5-fold rise in creatinine above baseline [AKI] OR a white blood cell count [WBC] >15,000 cells/mm³).

Results: In total, 1144 cases were included in the validation cohort, and the primary outcome occurred in 90 (7.9%). The PPV and NPV for the primary outcome of severe CDI for the nine severity scoring systems showed similar performance (range 0.14–0.27 and 0.93–0.95, respectively). More variability was observed for other metrics: sensitivity (0.24–0.59); specificity (0.68–0.98); and DF (0.56–0.66). No score performed significantly better than the IDSA score by NRI (−0.149 to +0.02). The score from Gujja, et al 2009 (WBC >30,000 cells/mm³and AKI) had the best performance overall (DF 0.67; NRI +0.01).

Conclusion: Of the nine CDI severity scoring systems analyzed in this validation cohort, none of the systems demonstrated a strong PPV or DF, nor did they significantly improve upon the IDSA criteria. The severity scoring system proposed by Gujja et al had the highest DF, but this still fell short of clinical usefulness. Other approaches, such as derivation of a score from a multicenter cohort and/or incorporation of novel biomarkers, may be necessary to improve the predictive modeling of severe CDI.