CCDS for Appropriate C. difficile Testing

At a 619-bed tertiary-care hospital, we performed a quasi-experimental retrospective cohort study analyzing rates of inpatient C. difficile test ordering and National Healthcare Safety Network (NHSN)-defined hospital-onset (ie, occurring on hospital day >3) C. difficile infection (HO-CDI) laboratory-identified (LabID) events⁶ before and after the introduction of a CCDS tool with nurse and provider education along with a financial incentive for graduate medical education (GME) trainees. The CCDS tool was developed after internal auditing by antimicrobial stewardship identified that 10 of 15 HO-CDI events (67%) during a 1-month period potentially lacked an indication for testing.
The 2-part CCDS tool first displayed a duplicate-order information screen listing C. difficile test results within 28 days. Second, a series of questions designed to guide appropriate testing was presented to the ordering provider. The algorithm (Figure 1) was designed to highlight duplicate C. difficile tests that may be low yield⁷ and practice guidelines recommending testing only of symptomatic patients, while considering risk factors including antibiotic use, intra-abdominal surgery, and advanced age.⁸ A step-wise algorithm was chosen based on limitations of screen size and ease of reading (see Supplementary Material for software demonstration). A test could be ordered regardless of provider responses. According to the existing laboratory protocol, solid stool specimens were rejected for NAAT testing.FIGURE 1

Two-part clinical decision support algorithm. NOTE. NAAT, nucleic acid amplification test for Clostridium difficile; C. diff, C. difficile; PPV, positive predictive value; WBC, white blood cell count.

The CCDS tool was preceded by a series of educational efforts presented to all providers and nurses, including email, flyers, and a brief video (see Supplementary Material). These efforts explained the rationale for the CCDS, provided guidelines on appropriate C. difficile specimens, and demonstrated the tool. A representative body of GME trainees performed in-person training with house staff in each inpatient department. Education occurred over a 2-week period prior to activation of the CCDS tool with a reminder message on the day of implementation. In addition, GME trainees were provided a 0.8% bonus (jointly funded by the UVA Office of Graduate Medical Education and UVA Health System) at the end of the academic year (June 2017) if testing by GME providers fell by ≥25% compared to the preintervention period.
The CCDS tool was developed in response to a broad, multidisciplinary commitment to HO-CDI reduction endorsed by hospital leadership. Monitoring was conducted daily using an electronic C. difficile dashboard reflecting all daily tests, new positive tests, duplicate tests, and test attempts “prevented” by the CCDS, provided to hospital staff and administration with unit and service attributions. The antimicrobial stewardship team performed chart reviews of patients with positive tests, evaluating appropriate testing and other opportunities to reduce HO-CDI. In addition to CCDS implementation, peroxyacetic acid/hydrogen peroxide-based cleaner was adopted hospital-wide in May 2017 to replace quaternary ammonium and bleach for daily and terminal hospital room disinfection. In addition, a policy change on April 2017 restricted antibiotics for neurosurgical drain prophylaxis. No other major new C. difficile-related infection control interventions were implemented during the study period.
An 18-month preintervention control period (June 2014 to November 2016) was compared to a 10-month postintervention period (December 2016 to September 2017) following CCDS implementation on December 5, 2016. In this analysis, HO-CDI and test count data were normalized to monthly patient days. An order was considered prevented if providers initiated but did not complete a C. difficile NAAT order. Canceled test orders and samples not submitted to the laboratory were excluded from the analysis.
Testing rates and proportions of positive tests were compared between the intervention groups using independent sample t tests and the χ² test, respectively. Due to fewer total HO-CDI events, a quasi-Poisson model was used to assess changes in HO-CDI counts between pre- and post intervention periods, using patient days as an offset. Analyses were performed using statistical R version 3.4.1 software (R Core Team, Vienna, Austria). The University of Virginia Internal Review Board approved this study (no. 20082).