Since 1978 when Clostridioides difficile (C. difficile) was first recognized as a cause of antibiotic-associated diarrhea and pseudomembranous colitis, C. difficile infection (CDI) has become the most common healthcare-associated infection (HAI) in the United States and is costly to the healthcare system [1]. Nonmodifiable CDI risk factors, like patient age and comorbidities (e.g., organ transplant, chronic renal disease, inflammatory bowel disease) have been described elsewhere [2]. However, a focus of prevention is to address established, modifiable CDI risk factors, specifically, exposure to antibiotics [2].

Exposure to antibiotics is associated with an increased risk of CDI [3]. Different antibiotics are associated with different levels of CDI risk. A meta-analysis quantified the risk of developing CDI in hospitalized patients receiving particular classes of antibiotics compared to those receiving no antibiotics [3]. Results from this meta-analysis support classification of clindamycin (associated CDI odds ratio (OR) = 16.8, 95% confidence interval (CI) 7.5 to 37.8), carbapenems and any cephalosporin (OR = 5.7, 95% CI 2.1 to 15.2), and fluroquinolones (OR = 5.5, 95% CI 4.3 to 7.1) as “high risk” CDI antibiotics, while penicillin combinations (OR = 2.7, 95% CI 1.8 to 4.2)  and trimethoprim/sulfonamides (OR = 1.8, 95% CI 1.3 to 2.4) can be classified as “moderate risk” CDI antibiotics [3]. Other antibiotics are considered lower risk for CDI [4]. The types of antibiotics prescribed and, therefore, the relative proportions of high-risk, moderate-risk, and low-risk antibiotics prescribed, can vary across healthcare settings (i.e., inpatient settings versus outpatient settings).

Improving antibiotic use is critical, not only to reduce an individual patient’s CDI risk, but also to mitigate emerging antibiotic resistance [5]. In U.S. acute care hospitals, 30%–50% of all prescribed antibiotics are considered inappropriate [6–11]. Inappropriate antibiotic prescribing can be defined as (1) unnecessary antibiotic prescribing (e.g., prescribing an antibiotic for a viral infection); (2) inappropriate antibiotic selection (e.g., prescribing an antibiotic associated with a higher risk of CDI when a lower risk antibiotic would be as or more effective); (3) inappropriate dose or duration of therapy; or (4) inappropriate antibiotic route of administration [12].

Efforts to reduce inappropriate antibiotic prescribing include antibiotic stewardship initiatives [13]. Antibiotic stewardship refers to a set of coordinated strategies designed to enhance patient health outcomes, reduce resistance to antibiotics, and decrease unnecessary costs [14]. Historically, antibiotic stewardship initiatives have focused on inpatient settings, especially within short-term acute care hospitals (STACHs). However, these efforts are expanding to long-term care facilities (LTCFs) (i.e., nursing homes and long-term acute care hospitals (LTACHs)) and outpatient settings in the community [13]. Antibiotic stewardship initiatives aimed at reducing inappropriate antibiotic selection, and restricting exposure to CDI high-risk antibiotics when lower risk antibiotics are appropriate, have proven effective in preventing CDI within STACHs [15]. Although emerging evidence documents the impact of antibiotic stewardship initiatives in LTCFs and in outpatient settings, the effects of specific interventions in these settings on CDI incidence are not well understood [16, 17].

The overall burden and transmission dynamics of C. difficile may be influenced by community sources and person movement among healthcare facilities [18–20]. Given this interconnectedness, a coordinated approach to antibiotic stewardship (i.e., implementation of interventions at multiple facilities, rather than at individual facilities) across a regional healthcare network may be most effective in reducing the overall burden of CDI in a region.

Agent-based models (ABMs) can capture the complexities of person movement among healthcare facilities and the communities they serve. Increasingly, ABMs are used to assess HAI prevention interventions, including antibiotic stewardship, which can inform decision making in public health and healthcare settings. Previous ABMs have demonstrated the reduction of CDI incidence associated with antibiotic stewardship initiatives in inpatient settings [21] and the benefits of a coordinated approach to interventions across inpatient settings in reducing incidence of CDI and other HAIs [22]. However, ABMs can also be used to estimate the impact of antibiotic stewardship efforts in LTCFs and outpatient settings and the effect of a coordinated antibiotic stewardship approach across regional healthcare networks.

We developed a geospatially explicit ABM of a regional healthcare network in North Carolina (NC) capable of examining the relative effectiveness of antibiotic stewardship interventions on CDI incidence [23]. We simulated reductions in unnecessary antibiotic prescribing and inappropriate antibiotic selection with intervention scenarios at individual and network healthcare facilities, including STACHs, LTCFs, and outpatient settings. Antibiotic exposure assigned to agents located in the community was conceptualized as the agent being “prescribed” the antibiotic at an outpatient healthcare facility. We estimated the effect of these interventions on hospital (healthcare) onset (HO)- and community acquired (CA)-CDI incidence trends across multiple simulations and discuss how this information could be used to inform antibiotic stewardship efforts across a regional healthcare network.

This study was approved by the Institutional Review Boards at the University of North Carolina at Chapel Hill and RTI International.

We implemented antibiotic stewardship interventions in a geospatially explicit ABM of a regional healthcare network and examined the relative effectiveness of these interventions on CDI incidence compared to baseline. Our ABM demonstrated relative changes in CDI incidence based on simulated antibiotic stewardship initiatives to (1) reduce unnecessary antibiotics at individual facility types compared to a coordinated approach across network locations; and (2) reduce inappropriate antibiotic selection in STACHs and in nursing homes. Although modest, the general trends in CDI incidence reduction and the relationships between implementation at individual facility types and other network locations are apparent.

We observed a greater impact on network HO-CDI by simulating a reduction in antibiotic prescribing in nursing homes alone (14% HO-CDI reduction) compared to the same intervention in STACHs alone (2% HO-CDI reduction). This finding is notable considering that the absolute number of antibiotic courses prescribed in our ABM varies by healthcare facility type. During a typical run of the ABM, the number of antibiotic courses initiated in STACHs is approximately seven-times greater than the number of antibiotic courses initiated in nursing homes. Therefore, a 30% reduction in total antibiotics administered in network STACHs is a larger reduction in the absolute number of antibiotic doses than a 30% reduction in total antibiotics administered in nursing homes. However, the nursing home LOS (mean: 95 days) is longer than the STACH LOS (range of means: 4.3–6.6 days) in the ABM (S1 File) [23]. Agents in nursing homes are, therefore, at increased CDI risk for a longer time period compared to agents at STACHs. Additionally, these agents who are prescribed antibiotics will likely remain in a nursing home for most their antibiotic exposure time, increasing their CDI risk because of facility-based transmission rates (S1 File) compared to agents discharged to their homes. Finally, agents in nursing homes are inherently at increased risk of CDI due to their age (≥65 years of age).

By simulating a reduction in inappropriate antibiotic selection, we estimated a relative reduction in CDI high-risk antibiotics prescribed, in favor of moderate-risk antibiotics, in STACHs and nursing homes. In these simulations, the interventions focused at nursing homes alone resulted in a larger impact on CDI incidence, compared to similar interventions at network STACHs alone. Again, this finding can largely be explained by the longer LOS for agents in nursing homes, compared to STACHs, in the ABM and their inherent increased risk of CDI due to age.

With each of the two antibiotic stewardship interventions modeled, we observed occasional simulation results of increased CDI incidence. For the intervention scenario of reducing total antibiotics prescribed, these observations were inconsistent across percent reductions (Fig 2A, 2B and 2C). For example, 10%, 20%, and 30% reductions in total antibiotics prescribed in nursing homes resulted in a 2% increase, a 2% decrease, and a 2% increase, respectively, in STACH HO-CDI incidence (Fig 2B). Similarly, the greater the reduction in high-risk antibiotics prescribed in nursing homes, the smaller the increase in STACH HO-CDI incidence (Fig 3B). This may be related to random variation during simulations. In the future, we will consider conducting additional simulation runs, beyond the 40 simulation runs presented here, to further investigate this variation.

Among our interventions for reduced antibiotic prescribing, we found a maximum impact on CDI incidence when modeling the intervention across all network inpatient and outpatient locations, with a 17% reduction in network HO-CDI incidence and a 7% reduction in CA-CDI incidence. These results support a coordinated approach to antibiotic stewardship across multiple facilities and facility types as an effective strategy to reduce the burden of CDI in regional healthcare networks [55, 56].

This modeling study is unique in considering outpatient prescribing, conceptualized as agents located in the community node being “prescribed” the antibiotic at an outpatient location. Healthcare facilities within a network are interconnected with each other and their catchment areas through person movement. By including the community in our ABM, this study brings an enhanced understanding of the potential impact of antibiotic stewardship interventions on CDI incidence in the community. This not only provides additional nuance to our interpretations but also makes the ABM more realistic, which could aid in translation of findings to public health and healthcare stakeholders, especially those who are unfamiliar with ABMs.

We calculated daily antibiotic prescribing rates at network STACHs using patient-level data from UNC Health Care, which varied slightly across facility intensive care unit bed categories. Although the impact of this variability is not established, future work could include evaluating the sensitivity of the model output to variable daily prescribing rates across different healthcare facility types (e.g., tertiary care versus critical access STACH). Further, these daily prescribing rates could be informed by data from other healthcare systems in other regions.

There are several limitations to the study. Our results suggest modest relative effects that make some scenario comparisons challenging to interpret. For example, a reduction in antibiotics prescribed at STACHs alone resulted in a 14% reduction in STACH HO-CDI incidence, while a reduction in antibiotics prescribed across all network locations produced only an 11% reduction in STACH HO-CDI incidence. This counterintuitive finding can be explained by the negligible difference between the actual HO-CDI incidence values, which are quite similar over the 40 model simulations (6.0 cases per 10,000 patient-days and 6.2 cases per 10,000 patient-days, respectively). This may also be related to random variation during simulations or to the antibiotic assignment rules that we employed (e.g., an agent could only be assigned a new course of antibiotics once it completed an existing course). In the future, we will consider conducting additional simulation runs, beyond the 40 simulation runs presented here, to further investigate this variation. Although community administration of antibiotics did not increase the likelihood that an agent was transferred to a healthcare facility, this update could be implemented in subsequent iterations of the ABM. Finally, other types of models (e.g., deterministic differential equation model) could arguably be used to evaluate antibiotic stewardship interventions as presented in this manuscript. However, we view these interventions and our results as initial steps in our effort to use this ABM to study a variety of simulated interventions as follow-on studies.

We employed the antibiotic stewardship interventions uniformly across all 10 network STACHs, but this does not impart the maximum possible variation that the ABM can provide across these 10 sites. Similarly, outpatient prescribing was treated uniformly across the community location node. However, variability in outpatient prescribing varies by geography and facility type [46]. We assumed that the interventions were implemented with perfect specificity, only reducing inappropriate antibiotic prescribing and not appropriate antibiotic prescribing.

We evaluated the two antibiotic stewardship interventions independently. Realistically, these interventions would be applied in coordination and to varying degrees across individual facilities and multiple facilities, as stewardship efforts should be directed toward the most relevant factors applicable to that setting [57]. Future simulations will address these limitations by accounting for variability in prescribing practices across locations and by exploring implementation of antibiotic stewardship interventions at single STACHs of different types (e.g., tertiary care, critical access) on incidence of C. difficile and other HAI pathogens. We conceptualized reduction of inappropriate antibiotic use as shifts from higher-risk to moderate-risk CDI antibiotics. The impact of successful antibiotic stewardship efforts at different facility types will lead to greater variation in proportions of low, moderate, and high-risk CDI antibiotics than was modeled in this study.

All antibiotic prescribers and healthcare facilities are responsible for antibiotic stewardship efforts. This responsibility includes a determination if antibiotics are necessary and, if so, the optimal antibiotic selection, dose, duration, and route [12]. Antibiotic stewardship efforts may be maximized if implemented across multiple healthcare facilities serving a population. Our findings build on the established modeling literature that supports the effectiveness of a coordinated approach for antibiotic stewardship across multiple facilities. This study also uniquely demonstrates the added impact of antibiotic stewardship interventions in LTCF and outpatient locations across a regional healthcare network for a reduction in the burden of HAIs.