As background, Clostridium difficile infection is a gastrointestinal infection associated with significant morbidity in the United States. C diff is traditionally considered to be a healthcare associated infection although it is becoming more common in ambulatory populations. The primary risk factors for CDI are advanced age, underlying comorbidity, and antimicrobial use. First, you are likely wondering who will benefit from being able to accurately predict a patient’s risk for CDI.1) Vaccine developers – to determine clinical trial eligibility2) In turn, if or when a C. difficile vaccine is developed, healthcare providers could determine which patients would benefit from vaccination. 3) Finally, physicians might use the risk score to identify patients for more judicious utilization of antimicrobials or risk management for CDI.
To develop the risk score, we conducted a retrospective cohort study among KPNW patients who had an index outpatient visit for any reason between July 2005 and September 2008. We then followed these patients for one year--until the end of follow-up or the first occurrence of CDI. C. difficile was identified through toxin tests in combination with treatment for CDI or through ICD-9 codes. We then identified potential predictors for CDI based on the current scientific literature and input from clinicians. As we were designing the risk score to be pragmatic and useful in clinical practice, we also tried to select predictors which can be easily obtained during patient care, through electronic health records. Predictors included: healthcare utilization—hospitalization, stay in a nursing home; medication use, specifically antimicrobials and gastric acid suppressants; and history of immunosuppression, renal dialysis, and chemotherapeutic procedures or therapies. These were measured during the 60-day baseline period before the index outpatient visit. The presence or absence of comorbid conditions were measured in the one year prior to the index visit.We used Cox regression to evaluate baseline patient characteristics that might predict CDI in the one year following an outpatient visit. The risk score results from the translation of the coefficients from the Cox regression into a points-based system, in which a higher number of points indicates a higher risk of CDI. First, the linear predictor in the Cox model was mapped to the corresponding one-year risk for CDI. Following this, the components of the linear predictor were rescaled to an arbitrary axis in which a score of zero points was assigned to the lowest-risk category for each variable, with increasing points counted for proportionate increases in the linear predictor. These risk score points approximate the exact hazard ratio for CDI. We calculated the observed risk of CDI for each decile of patients’ predicted risks of CDI to measure discrimination and calibration. The observed and predicted risks were then plotted using failure curves.Following development of the risk score at KPNW, we validated and recalibrated the risk score using a retrospective cohort of KPCO members with an index outpatient visit during the same time period. We used the same patient characteristics for the development and validation risk scores.