Clinical prediction rule uses EHR to estimate risk of recurrent C. difficile infection

SAN DIEGO — An adapted Zilberberg model clinical prediction rule was able to identify patients at risk for recurrent Clostridium difficile infection with moderate discrimination using electronic health record data from the Veterans Affairs Health System, according to data presented at IDWeek 2015.

“The objectives of this study were in a real world setting to predict recurrent CDI in a Veterans Affairs system using two published prediction rules, to evaluate the prediction rules across inpatient and outpatient settings, and to evaluate the prediction rules using electronic health record data,” Karim Khader, PhD, from the VA Salt Lake City Health Care System, said during his presentation. “We were interested in investigating the possibility of a robust prediction rule across a large geographically diverse population that can be implemented with minimal effort on the part of the clinician.”

Khader and colleagues performed a retrospective cohort study of 56,273 patients with a new Clostridium difficile infection within the U.S. Department of VA health care system from January 2006 through December 2012. They adapted prediction rules developed by Hu and colleagues in 2009 and Zilberberg and colleagues in 2014 for use with VA EHR data, and evaluated the performance of each in predicting risk for first recurrence between 2 and 8 weeks after the initial infection.

During the study period, 14.8% of patients developed a recurrent CDI. The Hu score categorized 48.4% of patients as high risk and discriminated between patients with and without recurrent CDI slightly more than half of the time (C-statistic = 0.548), while the Zilberberg model did so with moderate discrimination (C-statistic = 0.707). Having two or more hospitalizations in the prior 60 days was the strongest predictor of recurrent CDI with a more than fivefold increase in risk (OR = 5.34; 95% CI, 5.02-5.68), and there was a significant lack of model fit for both prediction rules (P < 0.1).

“In conclusion, the Zilberberg model achieved moderate discrimination to predict recurrent CDI in a cohort of inpatients and outpatients using EHR data,” Khader said. “Further work is needed to evaluate whether using prediction rules to target secondary prevention efforts can improve outcomes and reduce recurrent infections.” – by Adam Leitenberger

Reference:

Stevens V, et al. Abstract 70. Presented at: IDWeek; Oct. 7-11, 2015; San Diego.

Disclosures: This study was supported by an investigator-initiated research grant from Merck Pharmaceuticals. The researchers report that the views expressed in this project are those of the authors and do not necessarily reflect the position policy of the Department of Veterans Affairs or the U.S. Government.