A predictive model to identify Clostridium difficile toxin in hospitalized patients with diarrhea.

Abstract

Although Clostridium difficile is a common pathogen in hospitalized patients with diarrhea, no study has attempted to develop a predictive model to estimate the likelihood of C. difficile positivity. We conducted a cross-sectional study at a single hospital of 271 patients with diarrhea for whom stool was tested for C. difficile toxin. The sample was randomly divided into a subset to derive the model (n = 180) and another to validate it (n =91), and independent predictors of toxin positivity were identified using logistical regression analysis. C. difficile toxin was present in 81 patients and absent in 190. In the derivation set, a positive toxin was independently predicted (p < 0.0005) by readmission within 2 wk of prior hospitalization, by a white blood cell count > or = 10,000/mm3, and by presence of fecal leukocytes. In the validation set, C. difficile toxin was present in 24, 29, and 77% of patients with 0, 1, and > or = 2 risk factors, respectively. If validated prospectively and/or other centers, the model could identify patients who should be considered for empirical management while awaiting results of toxin assays.