70. Evaluation of Existing Clinical Prediction Rules to Identify Patients at Risk of Recurrent Clostridium difficile Infection (rCDI) using Electronic Health Record (EHR) Data from the Veterans Affairs Health System
Session: Oral Abstract Session: Clostridium difficile Infections: Prophylaxis and Epidemiology
Thursday, October 8, 2015: 9:00 AM
Room: 5--AB

Background: Recurrent Clostridum difficile infection (rCDI) affects approximately 20% of patients with CDI. Accurate prediction of rCDI would allow for targeted prevention efforts. Several prediction rules for rCDI have been developed, but their generalizability across inpatient and outpatient settings or for implementation using EHR data is unknown. The objective of this study was to evaluate the ability of two published prediction rules to identify patients that go on to develop rCDI in the Veterans Affairs system

Methods: We conducted a retrospective cohort study of all patients with a new CDI within the US Department of Veterans Affairs (VA) health care system between January 1, 2006 and December 31, 2012. A diagnosis of CDI was based on a positive laboratory result by toxin or molecular assay. rCDI was defined as a second positive laboratory result between 2 and 8 weeks after the initial test. Only the first infection and the first recurrence were included in the analysis. Prediction rules developed by Hu and colleagues (2009) and Zilberberg and colleagues (2014) (see table) were adapted for use with VA EHR data. The risk of recurrence was modeled as a function of each prediction rule individually using logistic regression. Prediction rule performances were assessed using C-statistics and Hosmer-Lemeshow goodness-of-fit tests

Results: During the study period, 56,273 patients developed a new CDI, of whom 7,446 (14.8%) went on to develop rCDI. Approximately half (n=24,344, 48.4%) of patients were classified as high risk based on the adapted Hu score, and the model discriminated between patients with and without recurrence slightly more than half of the time (C-statistic 0.548). The adapted Zilberberg model achieved moderate discrimination (C-statistic 0.707). The strongest predictor of recurrence was ≥2 hospitalizations in the prior 60 days (OR 5.34, 95%CI 5.02 5.68). There was a significant lack of model fit (p<0.10) for both prediction rules.

Conclusion: Prediction rules for recurrence may help physicians choose the most appropriate treatment for incident CDI. The Zilberberg model achieved moderate discrimination to predict rCDI in a cohort of inpatients and outpatients using EHR data. Further work is needed to evaluate whether using prediction rules to guide treatment decisions can improve patient outcomes.

Vanessa Stevens, PhD1,2, Karim Khader, PhD2, Richard E. Nelson, PhD2, Makoto Jones, MD, MS3, Kevin Brown, PhD4, Michael Rubin, MD, PhD, FIDSA5 and Matthew Samore, MD, FSHEA6, (1)University of Utah College of Pharmacy, Salt Lake City, UT, (2)Ideas Center, VA Salt Lake City Health Care System, Salt Lake City, UT, (3)Internal Medicine, University of Utah School of Medicine Division of Epidemiology, Salt Lake City, UT, (4)VA Salt Lake City Health Care System, Salt Lake City, UT, (5)Internal Medicine, University of Utah School of Medicine, Salt Lake City, UT, (6)University of Utah School of Medicine, Division of Epidemiology, Salt Lake City, UT

Disclosures:

V. Stevens, None

K. Khader, None

R. E. Nelson, None

M. Jones, None

K. Brown, None

M. Rubin, None

M. Samore, None

Findings in the abstracts are embargoed until 12:01 a.m. PDT, Wednesday Oct. 7th with the exception of research findings presented at the IDWeek press conferences.