Category: Professional Posters
Purpose: Clostridioides difficile infection (CDI) is a significant challenge in healthcare. Patients receiving broad-spectrum antibiotics are at increased risk of acquiring CDI, but further elucidation of compounding risk factors may offer prevention targets. The primary objective was to determine if a CDI risk prediction model initially developed at Novant Health accurately identified patients at high-risk for healthcare facility-onset CDI at two hospitals within Emory Healthcare.
Methods: We conducted a retrospective case-control study including adult patients admitted between July 1st, 2016 and July 1st, 2018 at a large academic medical center and community teaching hospital. Patients diagnosed with healthcare facility-onset CDI who received systemic antibiotics prior to diagnosis were included as cases, and they were matched 1:1 with controls. We collected data on known risk factors for CDI and compared the rates between the cases and the controls using students’ t-test for continuous variables and chi-squared for categorical data. Only variables that were statistically significant (p-value < 0.05) in a univariate test and evaluable at hospital admission were included in the multivariate analysis. Multivariate logistic regression model was used to build a point-based tool with weighted risk factors. The weight of the risk factors is decided by dividing the adjusted odds ratio (OR) by half of the smallest OR and rounding it to the nearest integer. The performance of the model was assessed by calculating the positive predictive value, negative predictive value, etc and by calculating a ROC-AUC.
Results: The study included 362 subjects (161 controls and 161 cases). In the univariate analysis, cases were more likely to have been hospitalized in the last 90 days (44.7% v 18.6%, p<0.001), to have a hematologic or solid tumor malignancy (34.8% v 24.2%, p=0.038), to have received a proton pump inhibitor (62.7% v 44.7%, p=0.001) or histamine-2 receptor antagonist (48.4% v 35.4%, p=0.018) while inpatient, and to have received either glycopeptide (84.5% v 47.8%, p<0.001) or carbapenem (26.1% v 6.8%, p<0.001) antibiotics. In the multivariate analysis, hospitalization within 90 days (OR: 3.51, 95% CI: 2.12-5.83) and hematologic or solid tumor malignancy (OR: 1.65, 95% CI: 0.99-2.73) remained important variables of CDI. The Novant Health model including advanced age and prior hospitalization demonstrated poor utility when applied to patients at Emory Healthcare with a ROC-AUC of 0.62. When creating a separate Emory-specific model, with the removal of age and addition of malignancy in the model, the AUC improved to 0.65. The Emory-specific model includes 4 points for hospitalization within 90 days and 2 points for hematologic or solid tumor malignancy. A score of 6 would have a positive predictive value of 82% and a specificity of 96% for predicting development of CDI.
Conclusion: Hospitalizations within the past 90 days and a diagnosis of hematologic or solid tumor malignancy were associated with increased risk of CDI in patients receiving broad-spectrum antibiotics at two hospitals within Emory Healthcare. Healthcare exposures were important in both the Novant Health model and Emory Healthcare models for predicting CDI, but the presence of a significant oncology population at Emory precluded utilization of the same model at both institutions. Going forward, pharmacists plan to use the Emory Healthcare model to screen patients at admission in order to identify those patients at highest risk for CDI, and to intervene when possible.