SAN DIEGO — Prediction models can be incorporated into electronic health record (EHR) systems to improve clinicians’ ability to identify patients who are more likely to experience recurrence of Clostridium difficile infection (CDI), according to study findings presented at ID Week.1
“One of the ways we can reduce C. difficile infection is through secondary prevention of recurrent cases,” said lead investigator Vanessa Stevens, PhD, a research assistant professor at the University of Utah College of Pharmacy in Salt Lake City and an investigator at the VA Salt Lake City Health Care System. “Now that we have more treatment options that can be used specifically to reduce the risk of recurrence, it is more important to be able to identify which patients might benefit from these treatments before they actually recur.”
Dr Stevens and her colleagues examined how two prediction models modified for use in the Veterans Affairs health system EHR database predicted recurrent CDI from Jan. 1, 2006 to Dec. 2012. The researchers calculated prediction scores manually.
One model was developed by Mary Y. Hu, MD, and colleagues and published in Gastroenterology; 2 the other was developed by Marya D. Zilberberg, MD, and colleagues and published in the Journal of Hospital Medicine.3
During the study period, a new CDI developed in 56,273 patients. Of these, 7,446 experienced a recurrent CDI. Based on the adapted Hu score, 24,344 patients were classified as high risk for recurrent CDI. The Hu and Zilberberg models discriminated between patients with and without recurrence 55% and 71% of the time, respectively.
The strongest predictor of recurrence was two or more hospitalizations in the previous 60 days, which increased the risk of recurrence more than five-fold.
“There are a number of predictive models in the literature, but the use of these models to guide treatment decisions is fairly limited,” Dr Stevens told Infectious Disease Advisor. “One reason is that clinicians may not have time to calculate yet another risk score. An alternative approach would be to have the score automatically calculated inside the electronic health record.”
Dr Stevens pointed out that the majority of predictive models in the literature were developed using real-time clinical data, such as diarrhea and other symptoms. “A score embedded in the electronic record would have to rely on structured data such as medication use and ICD-9 codes alone, and we don’t know how well the scores perform without those clinical data elements.”
Another issue is that most have not been tested outside of the studies in which they were developed, she said. “C. difficile is a remarkably genetically diverse organism, resulting in a fair amount of geographic variability in the type and severity of infections,” Dr Stevens said. “Predictive models can sometimes be so specific to the study population that they don’t translate well to other patients.”
Clinicians probably could do a reasonable job predicting CDI recurrence using readily available electronic health data, she noted, but it is not known whether using predictive models to guide intervention decisions can improve patient outcomes.
1. Stevens V, Khader K, Nelson RE, et al. Oral Abstract 26. 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. Presented at: ID Week 2015. Oct. 7-11, 2015. San Diego.
2. Hu MY, Katchar K, Maroo S et al. Prospective derivation and validation of a clinical prediction rule for recurrent Clostridium difficile infection. Gastroenterology. 2009;136:1206-1214.
3. Zilberberg MD, Riske K, Olsen M, et al. Development and validation of a recurrent Clostridium difficile risk-prediction model. J Hosp Med. 2014;9:418-423.