Discussion

Using a large and current dataset, we found an independent impact of hospital-level high-risk antibiotic use on HA CDI even after adjusting for confounding factors such as community CDI pressure, proportion of patients aged 65 years or older, average length of stay, and hospital teaching status. This investigation demonstrates a robust relationship between hospital-level high-risk antibiotic use and CDI risk, building upon a recent multicenter study using claims data.^{Reference Kazakova, Baggs and McDonald16} Our recent data sheds additional light on high-risk antimicrobial use since the CMS and the Joint Commission support for antimicrobial stewardship programs (ASPs), which increased the proportion of hospitals reporting incorporation of all 7 elements of CDC core elements in their facility ASPs (from 41% in 2014 to 76% in 2017).^24,25

Our finding that facility-level high-risk antibiotic use was independently associated with HA CDI differs from a VA study in which facility-level antibiotic use was independently correlated with facility-onset CDI only in long-term care facilities but not in the acute-care facilities.^{Reference Brown, Daneman and Jones5} This difference may be explained, in part, by our finding of a wide range of high-risk antibiotic use (a 110% increase from 10th to the 90th percentile), whereas the VA study found a narrow range of antibiotic use across 131 VA acute-care facilities (only a 14% increase from the 10th to 90th percentile). Such relative homogeneity in antibiotic use across the 131 VA acute care facilities may partially explain the lack of correlation between the acute-care facility-level antibiotic use and the facility-onset CDI rate.^{Reference Brown, Daneman and Jones5}

Our findings, along with the recent multicenter study,^{Reference Kazakova, Baggs and McDonald16} support the clinical relevance of hospital-level tracking of high-risk antimicrobials for CDI. High-risk antibiotics accounted for approximately half of all-risk antibiotics analyzed and may represent a more practical focus for CDI reduction programs in antimicrobial stewardship that may be short on resources.^{Reference McDonald, Gerding and Johnson1,Reference Yu, Rho and Morcos26} Targeted stewardship has been successfully implemented: the United Kingdom reported a reduction in CDI after implementation of cephalosporin- and fluoroquinolone-sparing programs.^{Reference Dingle, Didelot and Quan27}

Regarding specific antibiotic use at the facility level, Kazakova et al^{Reference Kazakova, Baggs and McDonald16} reported when assessed individually, cephalosporins (third and fourth generations) and carbapenems were significantly associated with HO CDI. Our study largely confirms these findings, although we included second generation in our evaluation of cephalosporins and we did not find carbapenem use alone to be associated with CDI. This finding may be due to less frequent carbapenem use (13% of overall high-risk antibiotic use; median 25.7 per 1,000 DP) in this more recent analysis, which was conducted >2 years after ASP guidelines were introduced.²⁴

The lack of association of fluoroquinolones and CDI in our study may reflect changes in antibiotic use in a post-ASP environment.²⁸ With ASP in place for multiple years and 2016 additions to black box warnings for quinolone use,²⁹ quinolones’ current contribution to CDI seems reduced, perhaps partially due to its reduced use. Meanwhile, cephalosporin use appeared in the highest volume and is corroborated here as a significant CDI risk. The “preferred” class of high-risk CDI antibiotic use may continue to change in the future. Thus, tracking high-risk antibiotic use as a group with several components is likely needed in time- and volume-based assessments, to better inform stewardship in the mitigation of CDI risk. This continuous monitoring is now feasible because the NHSN’s standardized antimicrobial administration ratio (SAAR) metric^{Reference van Santen, Edwards and Webb30} enables tracking clinical subclasses of drugs at the facility level. The SAAR provides an infrastructure to support more timely updating of ASP practices and CDI risk based on current epidemiology and practice patterns.

From a statistical perspective, a lack of statistical significance for the association of a specific class of antibiotics to CDI does not necessarily mean “no risk.” Volume of use, the range of variations, sample size, study design, and other factors may affect statistical test results. For this reason, aggregating subclasses of antibiotics based on prior documented risk for CDI may be a practical and more robust way to detect signals and to identify drivers of CDI.

Although we corroborate some findings in the study by Kazakova et al,^{Reference Kazakova, Baggs and McDonald16} there are methodologic differences. First, the Kazakova study used International Classification of Disease, Ninth Revision (ICD-9) secondary diagnosis of CDI in conjunction with oral anti-CDI antibiotics to identify cases, whereas we used positive nonduplicate microbiological test results to define cases, consistent with the current CDC NHSN AU module.²³ Second, the Kazakova study used pharmacy billing records to calculate the antibiotic use (DOT),^{Reference Kazakova, Baggs and McDonald31} whereas we used hospital pharmacy prescription start and stop orders to calculate the DOT. Third, our CDI categories are broader, including both HO CDI and same-hospital–associated, CO CDI, which potentially enabled assessment of a more representative burden of HA CDI. Because of the inclusion of the latter, we used admissions as the denominator to calculate the HA CDI rate rather than patient days to calculate the HO CDI rates. Despite these methodological and data-source differences, both studies found convergent independent association of acute-care facility-level antibiotic use and increased risk for CDI.

Our definition of HA CDI was previously used in a study that included both index HO and delayed outpatient-onset CDI as the outcome.^{Reference Tartof, Rieg, Wei, Tseng, Jacobsen and Yu7} The linked assessment of index HO (ie, HO) and delayed postdischarge (ie, CO-HA) CDI may better capture the full impact of hospital-level antibiotic use. If a patient had a prior, same-hospital stay within a short period, it may be reasonable to assume a potential delayed effect of hospital-level antibiotic use in the development of CDI. Our secondary analysis on HO- and CO-HA separately demonstrated that in-hospital antibiotic use increase CDI risk not only during hospital stay but also after hospital discharge. Nationally, the standardized infection ratio, a risk-adjusted metric of HO-CDI used by the NHSN, appears to be slowly decreasing whereas community-associated CDI (a subset of community-onset CDI), as reported by the CDC Emerging Infection Program, appears to be increasing or unchanged.³² With more HA CDI manifesting in the outpatient setting, spread of CDI spores may become an increasingly important issue.^{Reference Zilberberg, Tabak and Sievert21} CDI prevention initiatives focused on reducing C. difficile transmission and unnecessary antimicrobial use in both inpatient and outpatient settings may therefore be necessary to address the changing epidemiology of CDI.

The finding of moderate correlation between the hospital-level high-risk antibiotic use and HA CDI rates reflects considerable interhospital variation and may be partially due to unmeasured factors such as facility-level infection prevention and stewardship efforts that may modify the impact of antibiotic use on CDI. Another unmeasured variable is potential testing practice change, which may affect observed CDI rate. This aspect may be better addressed with longitudinal data. Furthermore, community differences in prescribing antimicrobials in the outpatient and other non-acute healthcare facilities could also be contributing factors. Although we included the CO-NHA CDI rate as a covariate in the model, it is not a direct assessment of variation of the outpatient antibiotic use.

The observation that teaching hospitals had higher HA CDI risk might be a result of detecting admissions of more severely ill patients. Prior studies have shown higher risk for CDI in the elderly, different ethnicities, and multiple antibiotic class exposure.^{Reference Brown, Daneman and Jones5,Reference Tartof, Rieg, Wei, Tseng, Jacobsen and Yu7,Reference Zilberberg, Tabak and Sievert21} Teaching hospitals may use multiple antimicrobial classes due to sicker patients, or as a byproduct of a hierarchical training environment.

CO-NHA CDI appears to contribute to the HO-CDI risk in admitting hospitals; hence, adjusting HO rates by a community-onset CDI variable appears to be appropriate for hospital performance reporting.^33,34 The role of CDI community pressure on hospital CDI risk is likely multifactorial, including skin contamination and environmental shedding of C. difficile spores, which is particularly high among patients recently treated for CDI.^{Reference Bobulsky, Al-Nassir, Riggs, Sethi and Donskey11,Reference Sethi, Al-Nassir, Nerandzic, Bobulsky and Donskey12} The overall wide range of CO-NHA CDI rates observed in this study suggests a large variation in community CDI pressure, and as such, awareness of local CO-NHA CDI pressure can help inform stewardship and infection prevention strategies.

Our study focused on high-risk antibiotics at a facility-level in the new era of ASP. It may be helpful to further delineate the relationship of facility thresholds of total and tiered (high/medium/low CDI risk) antimicrobial use to help inform targets for antimicrobial stewardship programs. In this paradigm, the concept of lower-risk antibiotics becomes of clinical value because a replacement agent will often be needed to treat the infecting organism.^{Reference Talpaert, Gopal Rao, Cooper and Wade35} New in 2019, a specific SAAR has been developed to measure use of antibiotics that pose a high-risk for CDI; this metric includes most of the same antibiotic classes included in our study.²³ From a stewardship perspective, a comprehensive CDI reduction program would therefore ideally involve assessment of specific patient demographics, risk profiles, and community pressure in addition to high-risk antimicrobials.

This study has several limitations. First, despite the study including a large number of community hospitals of diverse characteristics, it does not necessarily represent the characteristics of all US hospitals. Second, as an ecologic study, it depicts associations and not causal relations. Nevertheless, ecological studies are important for estimating likely degree of impact of a population-level intervention, especially through convergent insights involving diverse data sources, patient population, and methodologies. Although the complex physiologic interactions between the intestinal microbiome and antibiotics are not completely understood, prescribing practices comprise a risk factor that is malleable and can change with antimicrobial stewardship and benchmarking.^{Reference McDonald, Gerding and Johnson1,Reference Leffler and Lamont4–Reference Tartof, Rieg, Wei, Tseng, Jacobsen and Yu7,Reference Kazakova, Baggs and McDonald16,34} Future studies would ideally combine both facility- and patient-level measures of high-risk and total antibiotic use to further inform CDI mitigation strategies.