Background: Prevention of Clostridium difficile infection (CDI) remains a significant healthcare challenge. Risk prediction tools can potentially identify high-risk patients and allow for early prophylactic interventions. Various tools have been studied but none have been widely adopted. Our objective was to develop a simple risk prediction tool to identify medicine inpatients at high risk for developing primary CDI.
Methods: We conducted a retrospective, single-centre case-control study including patients admitted to the internal medicine service at our institution with a positive C. difficile polymerase chain reaction assay in loose stool. Controls were randomly selected from the same population. Risk factors for CDI were identified using univariate and multivariate logistic regression analyses. A model was created using variables that minimized Akaike Information Criterion and yielded higher area under the curve values.
Results: A total of 314 patients were included (157 with CDI and 157 controls). Variables included in the final 5-point, 3-variable risk prediction tool were age, modified Horns index and antibiotic use within 3 months. The tool demonstrated good discrimination with a C statistic of 0.79 and model optimism of 0.04 based on a bootstrap sample of 2000 replicates.
Conclusion: Our simple 3-variable risk prediction tool based on age, disease severity and recent antibiotic use facilitates rapid bedside assessment by clinicians to identify medicine patients at high risk of CDI on admission. Further research is needed to determine whether this tool can reduce primary CDI incidence and healthcare costs.
Table 1. 5-Point CDI Clinical Risk Tool and Predicted Risk.
Figure 1. Receiver Operating Characteristic (ROC) Curve for CDI Risk Prediction Model
Figure 2. CDI Risk Prediction Model Calibration Plot. Clostridium difficile infection risk prediction model calibration plot showing agreement between observed and predicted risks. Dashed line shows perfect agreement; lines above the dashed line indicate the predicted risks are lower than the actual risk.
V. Leung, None
V. Su, None
J. Puyat, None
S. Shalansky, None