Use of a Clostridium difficile clinical prediction rule to facilitate antimicrobial stewardship

Background

Antimicrobial stewardship programs (ASPs) are promoted as a strategy to optimise antimicrobial use, reduce the development of antimicrobial resistance and decrease Clostridium difficile infections (CDI) (Dellit et al, 2007; Goldstein et al, 2015). However, it can be challenging for health care institutions to implement ASPs that are robust enough to accomplish these goals (Bui et al, 2016; Davey et al, 2017). The most successful ASP tactics (e.g. antimicrobial preauthorisation, prospective audit and feedback and rounding-based “handshake” stewardship) are costly and labour-intensive (Barlam et al, 2016; Hurst et al, 2016). ASP approaches that are effective, but also time-saving and labour-sparing are needed.

Clinical prediction rules are tools that can be used to identify patients at risk for specific diseases or conditions (Dubberke et al, 2011; Press et al, 2016). In general, these tools rank the comparative contribution of various risk factors to a disease or condition, and then risk scores can be calculated for individual patients. Such systems enable the stratification of patients into risk groups and identification of those at highest risk. Tabek et al (2015) developed a clinical prediction rule that can identify patients at risk for CDI using data available at the time of hospital admission. Whether a tool that can identify patients at risk for CDI could also help ASPs rapidly identify patients at risk for antimicrobial overuse is not known. In this study, we assessed whether a CDI clinical prediction rule could be used to facilitate antimicrobial stewardship in an acute care hospital.

Methods

We retrospectively reviewed adult inpatients (>18 years old) diagnosed with CDI from October 2013 to December 2016 in our 617-bed acute care hospital. Demographic, pharmacy, laboratory and microbiology data were obtained from the electronic medical record and maintained in the REDcap® system (Harris et al, 2009). A prediction rule, adapted from Tabek et al (2015), and comprised of data accessible in our electronic medical record, was used to assign a score to each patient at the time of hospital admission. Points were assigned manually as follows: 2 points each for admission to the intensive care unit, serum creatinine >2.0 mg/dL, platelets ⩽150 or >420 × 10⁹/L, white blood cell count >11,000/mm³ and positive C. difficile toxin test within 3 days of admission; 3 points each for age >64 years, transfer from another hospital, hospital discharge within 30 days of admission and bands >32% on haematology panel; 4 points each for transfer from a skilled nursing facility, requiring mechanical ventilation on admission and albumin ⩽3 g/dL; and 5 points for prior diagnosis of CDI. Patient charts were reviewed, and data were collected for (1) all antimicrobials and proton pump inhibitors that each patient received on the day of hospital admission and at 72 h post-admission and (2) the numbers and types of potential ASP interventions that could have been enacted for each patient on the day of hospital admission and at 72 h post-admission. These interventions included stopping antimicrobials, de-escalating antimicrobials, escalating antimicrobials, changing antimicrobial dose and stopping proton pump inhibitors. Determinations regarding the appropriateness of antimicrobial and proton pump inhibitor therapy were based on national and institutional guidelines. Descriptive analysis was used to summarise patient demographics, medical comorbidities, pharmacy data and potential ASP interventions. Differences in categorical outcome variables were analysed with the Pearson Chi Square and Fisher’s Exact tests using SAS software version 9.3 (SAS Institute, Cary, NC). The Institutional Review Board of our institution approved this study.

Results

For the study period, a total of 316 patients were evaluated, and our annual National Healthcare Safety Network Standardised Infection Ratios for hospital-onset CDI ranged from 0.89 to 1.22 (California Department of Public Health, 2018; US Centers for Disease Control and Prevention, 2018). The mean patient age was 65.1 years (range: 18–102 years), 175 (55.4%) were men, and 132 (41.8%) had immunocompromising conditions (e.g. transplant recipients, HIV-infected patients and patients receiving immunosuppressant drugs). At hospital admission, 206 (65.2%) patients were started on antimicrobials. Patients >64 years old were started on antimicrobials more often than those <64 years old (72.0% vs 56.7%). Differences in the percentage of patients started on antimicrobials were not found between men and women (P = 0.91) or immunocompromised and non-immunocompromised patients (P = 0.90). At admission, a total of 73 potential ASP interventions were identified for 71 (22.4%) patients (each patient could have more than one intervention). Potential ASP interventions were identified for 20.6% of men, 21.3% of women, 22.9% of patients age >64 years, 8.6% of patients age <64 years, 22.7% of immunocompromised patients, and 19.6% of non-immunocompromised patients. The 73 interventions included 42 (57.6%) for stopping antimicrobials, 16 (21.9%) de-escalating antimicrobials and 15 (20.5%) stopping proton pump inhibitors.

At 72 h post-admission, 188/316 (59.4%) patients remained hospitalised. Of these, 96 (51.1%) were age >64 years, 103 (54.8%) were men, and 84 (44.7%) were immunocompromised. A total of 147 (78.2%) were receiving antimicrobials, including 77.7% of men, 78.8% of women, 82.3% of those age >64 years, 73.9% of those age <64 years, 84.5% of immunocompromised patients and 73.1% of non-immunocompromised patients. A total of 94 potential ASP interventions were identified for 88 patients. Potential ASP interventions were identified for 47.6% of men, 45.9% of women, 54.2% of patients >64 years old, 39.1% of patients <64 years old, 44.0% of immunocompromised patients and 49.0% of non-immunocompromised patients. The interventions included 60 (63.8%) for stopping antimicrobials, 18 (19.1%) de-escalating antimicrobials, 15 (15.6%) stopping proton pump inhibitors and 1 (1.1%) changing antimicrobial dose.

When the clinical prediction rule was applied, and scores calculated for the 316 patients, the mean score was 10.80 (SD 5.5; range: 0–24). The mean score was 10.95 (SD 5.7; range: 0–24) for men, 10.67 (SD 5.17; range: 0–24) for women, and 10.25 (SD 5.12; range: 0–21) for immunocompromised patients. Figure 1 shows the percent of patients with potential ASP interventions identified at the time of hospital admission (n = 71) and at 72 h post-admission (n = 88) for different score groups. When these patients were divided into low and high score groups, the high score group was more likely to have patients with potential ASP interventions identified on admission and at 72 h post-admission (Table 1).

OPEN IN VIEWER

Figure 1.

Percent of patients with potential for ASP interventions at hospital admission (n = 71) and at 72 h post-admission (n = 88) by patient score.

OPEN IN VIEWER

Table 1.

Comparison of patients with potential for ASP interventions at admission and 72 h post-admission by low vs high score group.

	Low score group (Scores = 0–9)	High score group (Scores = 10–24)	P value
Patients with potential for ASP intervention at hospital admission (n, %)	25 (17.1%)	46 (27.1%)	0.04
Patients with potential for ASP intervention at 72 h post-admission (n, %)	34 (35.8%)	54 (58.1%)	<.01

Discussion

In the present study, we examined whether a CDI clinical prediction rule could be used to facilitate antimicrobial stewardship. By applying a prediction rule, comprised of data available from our electronic medical record at the time of hospital admission, we were able to identify patients who were at risk of receiving antimicrobials unnecessarily and may benefit from ASP interventions. We found that, as CDI risk scores rose, the likelihood that patients received unnecessary antimicrobials at the time of admission, and at 72 h post-admission, increased. Also, as risk scores increased, there was a greater potential for ASP interventions, both at the time of admission and at 72 h post-admission. Rapidly identifying these at-risk patients via an automated prediction rule could expedite ASP involvement and, in turn, reduce adverse effects of unwarranted antimicrobial use.

Previous studies (Kuntz et al, 2016; Tabek et al, 2015) describe CDI prediction rules that can be used to determine, at the time of hospital admission, which patients have the greatest risk for CDI. The development of such prediction rules raised the prospect that, if at-risk patients could be identified in a timely manner, steps could be taken to mitigate their CDI risk. However, these studies did not specifically evaluate how prediction rules could be used to impact CDI risk factors, such as antimicrobial use – the greatest modifiable risk for CDI. We now demonstrate that CDI prediction rules have the potential to identify patients who are receiving unnecessary antimicrobials and who should be targeted for ASP interventions.

It is notable that we found substantial unnecessary antimicrobials use in the high score groups both at the time of hospital admission, and at 72 hours post-admission. The escalation of antimicrobial overuse in high score groups may reflect clinicians’ apprehension to withhold or discontinue antimicrobials for more ill patients or patients with non-specific symptoms. Also, in this time period, before microbiologic data is routinely available, clinicians may be inclined to over-prescribe broad-spectrum antimicrobials while awaiting test results. Since ASPs with limited manpower may not have the ability to review patients for potential interventions until after microbiologic data is established, employing a clinical prediction rule at the time of hospital admission could amplify an ASPs impact during this critical timeframe.

Limitations of the study are that this clinical prediction rule was evaluated retrospectively at a single academic medical centre. Prospective studies, with real time implementation of ASP interventions, are needed to determine whether CDI prediction rules can be used to decrease antimicrobial use and CDI rates. Nonetheless, this study highlights the potential value of leveraging a CDI clinical prediction rule, especially if automatically derived from electronic medical records, for antimicrobial stewardship.