A Multimodal Intervention to Reduce C. difficile Infections and Stool Testing

This project was conducted at a large quaternary-care children’s hospital. Testing options for C. difficile included a singleplex PCR assay (Xpert C. difficile, Cepheid, Sunnyvale, California) and GIP (FilmArray GIP, BioFire Diagnostics, Salt Lake City, Utah). The hospital, which uses Epic (Verona, Wisconsin), had limited experience with electronic medical record (EMR)-based testing restrictions.

Through process mapping, we built a key driver diagram identifying how children are diagnosed with C. difficile and drivers of stool testing overuse (Fig 1). We divided opportunities for improvement into 2 teams: (1) testing and treatment, and (2) C. difficile prevention. The testing and treatment team consisted of nurses, pharmacists, infection preventionists, a process improvement specialist, and physicians. The prevention team included physicians, nurses, a human factors engineer, and an environmental services manager.

Secondary drivers to improve testing and treatment included: (1) education and standardization of C. difficile treatment, (2) objective definition of diarrhea, (3) reducing C. difficile testing when pretest probability is low, and (4) reducing inappropriate testing. Secondary drivers to improve C. difficile prevention included: (a) improved hand hygiene, (b) adherence to universal gloving and use of bleach wipes, and (c) improved cleaning of affected rooms and shared spaces. The project was approved by the hospital’s Organizational Research Risk & Quality Improvement Review Panel and deemed exempt from institutional review board review.

To optimize C. difficile testing and treatment, we created a clinical pathway that included C. difficile risk factors, colonization rates, testing indications, what test to order, and treatment recommendations, which promoted increased vancomycin in alignment with national recommendations (Supplemental Fig 5).¹⁵ This pathway underwent extensive stakeholder engagement with widespread education before implementation in February 2019.

To address inappropriate testing related to inappropriate identification of diarrhea, we disseminated an objective definition of diarrhea using a pediatric version of the Bristol stool scale (BSS), “Choose your Poo,” from the Norgine Pharmaceuticals Web site. This scale was used for free with permission. The BSS describes stool consistency on a scale of 1 to 7, with 7 being the wateriest. Diarrhea was defined as ≥3 occurrences of Bristol 6 or 7 stool in 24 hours. A BSS picture was placed in all inpatient rooms and nursing stations. After individual unit training using train the trainer methodology, in February 2019, nurses began documenting BSS for each stool occurrence in the EMR. Units’ compliance with BSS documentation was monitored using run charts, data were shared with units monthly, and additional education was provided to low-performing units. Because the BSS is now embedded in standard documentation, new staff are trained during orientation. Through education on admission, parents were also encouraged to use the BSS to describe their child’s stool.

To address inappropriate testing for C. difficile related to the inability to test for more common causes of gastroenteritis without also testing for C. difficile, in April 2019, we developed a new test: GIP without C. difficile, where the laboratory ran the full panel but did not report C. difficile results. To help inform clinician pretest probability, C. difficile risk factors and scenarios with high colonization rates were included in the clinical pathway and educational sessions.

We developed a clinical decision support tool (CDS) in April 2019, adapted from work in adults, to restrict inappropriate stool testing.¹⁶ The tool pulled data from the EMR without requiring clinician input. There were 6 evidenced-based, hard-stop restrictions. Four restricted testing of any kind for children with: (1) no diarrhea defined using the BSS, (2) recent laxative use in previous 24 hours, (3) positive GIP in preceding 14 days, or (4) negative GIP in preceding 7 days. Two rules restricted certain types of testing: (a) no GIP testing for children hospitalized for >96 hours (recommend C. difficile PCR), and (b) no testing for C. difficile in infants <1 year old (recommend GIP without C. difficile).^5,17,18 The CDS looked for each restriction in a hierarchical fashion (order 1>5>3>4>2>6). If clinicians tried to order a restricted test, there was a hard stop and an alert informed them why it was restricted, and if applicable, suggested alternative tests (Supplemental Fig 6). There was an optional override if clinicians felt that testing was still indicated, which involved signing the order (which did not release it but sent it to a shared stool inbox) and calling the infectious diseases or gastroenterology fellow for approval. If approved, fellows would second sign the order from the shared inbox, which made it visible to nursing. We conducted several Plan-Do-Study-Act cycles to optimize the CDS. Lack of BSS-defined diarrhea was a consistent barrier because of limitations with real-time stool documentation; thus, we added alerts to remind nurses and liberalized the definition from 3 to 2 diarrhea stools in 24 hours.

Our prevention team focused on the oncology unit, the area with the highest CDI rate. We executed >40 prevention Plan-Do-Study-Act cycles (Supplemental Table 1); major interventions started in fall 2018. To improve hand hygiene, we revamped real-time audit-and-feedback processes and installed handwashing timers. To improve gloving and bleach cleaning compliance, we created new isolation signage. To improve compliance with UV light to clean affected rooms, we monitored and addressed barriers with environmental services. We also initiated regular UV cleaning of shared spaces. Finally, surface cleaning was monitored periodically with markers visible only with UV light, and feedback on persistent areas of concern (eg, intravenous poles) was shared with environmental services.

We included all hospitalized children 0 to 18 years of age who had a stool test performed over 30 months (January 2018–June 2020), excluding patients in the psychiatric and NICUs.

Our primary outcome was the rate of C. difficile HO-CDI, defined as a positive test in a child who had symptoms (eg, new-onset diarrhea) begin on or after day 3 of admission, which was adjudicated by real-time chart review per routine surveillance by infection preventionists. We evaluated the rate of C. difficile-positive results (ie, C. difficile detection) and as a process measure, testing capable of detecting C. difficile. All rates were per 10 000 PDs.

To monitor potential adverse events related to unreported positive C. difficile results, for 7 months after implementation, all charts of patients with positive, unreported C. difficile results were reviewed in real time by an infectious diseases physician. If there was concern for true CDI, results were released and treating clinicians were notified. As a balancing measure, we evaluated the proportion of patients with initially nonreported positive C. difficile results who ultimately had results released and were treated for CDI with vancomycin or metronidazole within 30 days. For children with positive C. difficile results that remained unreported and untreated, we collected demographics, clinical characteristics, and outcomes within 30 days of testing, including: (1) revisit for gastrointestinal symptoms (excluding scheduled visits with improved symptomatology), (2) repeat stool testing, (3) initiation of C. difficile antibiotics, and (4) C. difficile-related complications including toxic megacolon or death. We also estimated antibiotic days avoided annually on the basis of nonreported C. difficile results, assuming 100% treatment rate and 10-day course of vancomycin or metronidazole.⁵ 

We evaluated the rate of stool test utilization and negative stool results per 10 000 PDs, and percentage of stools documented with a BSS (process measure). We estimated savings on the basis of changes in test utilization and cost of testing reagents per test. As balancing measures, we evaluated the percentage of tests that were restricted but still ordered and sent for approval because clinicians felt they were clinically indicated (overall and by restriction) and the percentage of these tests that were ultimately approved (overall and by fellow specialty). We compared the percentage of actionable results (ie, bacteria and/or parasite identified) between initially restricted but ultimately approved tests and nonrestricted tests.

Measures were evaluated monthly using run and statistical process-control charts. Established rules for determining special cause variation were employed.¹⁹ Pearson’s χ² tests were used for bivariable analyses, and P < .05 was considered statistically significant.

There were 2001 tests performed for 1982 encounters (Supplemental Table 2).

There was special cause variation with a 55% decrease in hospitalwide HO-CDI rates (Fig 2), including a 48% decrease for oncology. We also found special cause variation with a 44% decrease in C. difficile detection (Supplemental Fig 7) and 48% decrease in testing for C. difficile (Fig 3).

During our monitoring period, 18% of GIPs without C. difficile were positive for C. difficile (84 of 477). Of these positive tests, 2.3% (2 of 84) were ultimately released and treatment was initiated. Thus, only 0.4% (2 of 477) of GIPs without C. difficile required intervention.

Of patients with positive C. difficile results that remained unreported and untreated (n = 82), 76% were <1 year old or had another organism detected (Supplemental Table 3). Four (6%) had a return visit, 3 of which were <1 year old or had another organism detected. Four patients had repeat stool testing, 2 were subsequently treated for C. difficile, and no patients had adverse events. On the basis of 82 unreported C. difficile results over 7 months, we estimate that 1406 antibiotic days were avoided annually.

We found special cause variation with a 29% decrease in stool testing (Fig 4) and 28% decrease in negative stool results (Supplemental Fig 8). Mean reliability of documentation with the BSS was 78% across the study (Supplemental Fig 9). The decrease in stool tests (29 tests per month) resulted in annual savings of >$55 649.

There were 23 restricted tests ordered per month that clinicians felt were clinically indicated and were sent to fellows. This represented 34% of all ordered tests and did not change over time. The most common reason for restrictions was a lack of documented diarrhea (Supplemental Table 4). A mean of 56% of restricted tests that were sent to fellows were accepted, without change over time or difference in acceptance rates between infectious diseases and gastroenterology fellows (P > .05). Restricted tests that were ultimately accepted had a lower rate of actionable results versus nonrestricted tests (25% vs 34%, P = .02).

The combination of a CDS tool, clinical pathway, and focused prevention efforts resulted in meaningful decreases in HO-CDI, C. difficile detection and testing, overall stool testing, and low-value testing. The decrease in CDI was likely driven by improved prevention, optimized treatment, and a reduction in misclassifying C. difficile colonization as CDI through reduced testing. Decreases in C. difficile testing and overall testing were likely because of EMR-based testing restrictions, optional C. difficile testing, and improved clarity regarding the definition of diarrhea. This quality improvement initiative resulted in substantial antibiotic reductions and cost savings.

We found a meaningful decrease in HO-CDI; although some of our success was likely because of focused prevention efforts in oncology and optimized treatment of CDI, a substantial decrease in hospitalwide CDI occurred with implementation of a clinical guideline supported and reinforced by CDS, suggesting that some decrease was likely because of decreased testing and detection of colonization. This is supported by previous work which found reductions in testing led to reductions in CDI, likely because of inappropriate classification of colonization as infection.²⁰ Because the gold standard HO-CDI definition involves confirmation of symptoms (eg, diarrhea) in the EMR, the gold standard is only as good as the information is in the EMR. With increasing use of highly sensitive diagnostic panels, our results highlight the ability of diagnostic stewardship to reduce CDI misdiagnosis and promote antibiotic stewardship.²¹ 

We successfully used a CDS tool to decease C. difficile testing and overall stool testing. CDS tools have been used to improve guideline implementation by presenting clinicians with information instantaneously, improving accuracy, and optimizing order efficiency.¹⁶ Previous work on this topic also found that EMR alerts could reduce C. difficile testing, but 1 study required manual data entry and another was done only in children <3 years old.^22,23 Our study builds upon this work by evaluating test restrictions across a range of ages without requiring manual input. We found that tests performed after initial restrictions had less frequent actionable results compared with nonrestricted tests, suggesting that our CDS tool functioned appropriately. The percentage of ordered tests that were restricted remained constant over time, suggesting that despite real-time feedback through CDS, clinicians’ ability to accurately predict low-yield testing and/or their desire to order low-yield tests remained unchanged; thus, the CDS tool will be required for sustainability.

A strength of our tool was the optional human override with specialist approval, which allows for individual patient variation and improves cultural acceptance. More than half of initially restricted tests were approved, which may question the process efficiency. However, because we could not quantify tests that were abandoned after restriction alerts, the proportion only considers orders that clinicians felt strongly enough about to try bypassing the restrictions. When reviewing a random sample of approved tests for appropriateness, often there was no consensus among experts, suggesting that it was reasonable to approve them. While we provided some continued education to fellows, we did not extensively focus on decreasing approved overrides for the above-mentioned reasons and to maintain cultural acceptance of this new process. The override process resulted in <1 call daily, which we believe is feasible for many other hospitals, including those without fellows.

In addition to testing restrictions, implementation of the GIP without C. difficile test that suppressed positive C. difficile results was an effective strategy for reducing C. difficile testing. Among patients with nonreported positive C. difficile results, <3% had results released and antibiotics started because of concern for infection. Additionally, a high proportion were <1 year of age or had another organism detected, suggesting a high likelihood of colonization. This further underscores the utility of not reporting C. difficile when CDI pretest probability is low. Few patients with suppressed results had medical reutilization, subsequent antibiotic initiation, or adverse events. Although poor outcomes are very rare in pediatrics, our work suggests that suppression of these results is likely safe, but future studies are needed.²⁴ 

Improving shared definitions of stool quality likely also contributed to our success. Initially, everyone had different definitions of diarrhea. The BSS, which was reinforced both in nursing documentation and parent engagement, allowed everyone to speak the same language. The CDS reinforced our shared definition of diarrhea with real-time feedback. BSS documentation was high, indicating it was not overly burdensome. To our knowledge, this is the first pediatric study to evaluate the ability of this scale, along with other interventions, to reduce low-value stool testing.²⁰ 

Although not the primary objective, our project saved >$55 000 in testing reagents alone and >1000 antibiotic days annually. Given the estimated cost of pediatric CDI ($1917–$8317 per case) and decrease in technician time associated with fewer tests, our cost savings were likely even more substantial.⁶ 

Our study is subject to limitations. First, we were unable to quantify orders abandoned after clinicians saw restriction alerts; thus, we likely underestimated the impact of our interventions. Second, the CDS tool looked for restrictions in order of a hierarchy, and if 1 restriction was met, it did not look for others. Thus, we cannot determine the true contribution of each restriction. Third, restrictions may have resulted in missed cases of CDI. However, the optional override process likely minimized this. Fourth, even though the proportion of ordered GIPs without C. difficile that required intervention because of concern for suspected CDI during our monitoring period was low (0.4%), it is possible that we missed CDI cases after intensive monitoring ended. However, this risk was mitigated by allowing providers to override restricted C. difficile orders if approved by fellows and the observation that most patients received testing for C. difficile if symptoms persisted. Fifth, the generalizability of our findings may be limited in hospitals without GIPs. We suspect that hospitals with individual organism testing can still use some of our interventions to decrease test utilization; however, the ability to impact C. difficile may be more limited. Finally, because of coronavirus disease 2019, we were unable to obtain more recent data to evaluate sustainability beyond our 17-month postintervention phase.

In conclusion, we found that a CDS tool coupled with education, standardized definition of diarrhea, and targeted prevention efforts led to significant and sustained reductions in C. difficile detection, infection, and stool test utilization.

We thank Alicia Swaim and Chris Poppy for building the clinical decision support tool; Ethan Lew for extracting the data; and Dan Hyman, Lalit Bajaj, and David Brumbaugh for their institutional support and assistance with contract negotiation. The Pediatric BSS used in this project was provided by Norgine Phamaceuticals.

BSS

Bristol stool scale

C. difficile

Clostridioides difficile

CDI

C. difficile infection

CDS

clinical decision support

EMR

electronic medical record

GIP

gastrointestinal panel

HO-CDI

hospital-onset C. difficile infection

PCR

polymerase chain reaction

PD

patient-day