Among the 8,811 studies identified in the four databases, 29 studies met eligibility criteria for full-text analysis of modeling progress, trends, and gaps, including short-form questions regarding the model structure and study characteristics. The results are listed in Tables S1 - S9 in Appendix D. Figure 2 illustrates the scoping review process and the inclusion and exclusion of studies based on PRISMA-ScR guidelines. Of the 141 studies that met the criteria in the title and abstract screening, 111 studies were excluded based on the selection criteria. The most common reason was “irrelevant study design,” for which 100 studies were excluded. Among the excluded irrelevant study design studies, 16 studies lacked STACHs, 81 studies either lacked community or long-term health facilities in their models, and 45 studies only modeled admission and discharge as a simple fixed importation/exportation rate. Additionally, 11 studies were excluded because they were commentary, literature reviews, or conference abstracts.
Model Structure and Assumptions
The most common type of model was stochastic agent/individual-based (n = 14, 47%; Fig. 3A). There were eight differential equation-based models (27%), and all but one were deterministic (n = 7, 23%). In studies that reported their software implementation, most models were programmed using C++ (n = 7, 24%), followed by R (n = 5, 17%). Among the seven studies programmed in C++, six studies 18–23 utilized the same model.
As with most epidemiological models, individuals’ health/disease states were characterized as compartments (e.g., susceptible, infected, and colonized). Most studies included at least three disease states, including differing levels of susceptibility (e.g., high susceptibility due to antimicrobial exposure), infectiousness, or strains (antimicrobial resistant versus antimicrobial susceptible). In our analysis, 13 studies (45%) had varying levels of susceptibility, 11 (38%) had varying levels of infectiousness in the infected compartment state, and 4 (14%) studies had multiple pathogenic strains included in their model. We also found that 13 (45%) studies included the detection status of colonization.
Settings and Pathogens
Table 1 summarizes the studies that were included in our analysis. In the included studies, we found that 9 studies included both long-term care settings and other types of community settings (e.g., households and workplaces) in their model, while 8 (27%) only included long-term care settings and 13 studies (45%) only included other types of community settings. Figure 3B lists the HAI-causing pathogens that were included in models reviewed; the two most common pathogens modeled were Escherichia coli (E. coli) (n = 7, 24%) and Carbapenem-resistant Enterobacteriaceae (CRE) (n = 7, 24%). Figure 3C shows the distribution of transmission settings modeled each year. Table 1 shows the setting modeled based on publication date. Most studies utilized data from the United States (n = 13, 45%). No studies included data from low- and middle-income countries (LMIC).
Table 1
Brief summary of the eligible studies for full-text review.
Pathogens | Authors [Refs] | Year | Short-term Acute Care Hospital | Long Term Healthcare Facilities | Other community not including LTCF and N.H. | Country |
C.diff | Durham et al. 28 | 2016 | Yes | Yes | Yes | USA |
C.diff | McLure et al. 29 | 2019 | Yes | No | Yes | USA |
C.diff | McLure et al. 30 | 2019 | Yes | No | Yes | USA |
C.diff | Rhea et al. 51 | 2019 | Yes | Yes | Yes | USA |
C.diff | Rhea et al. 52 | 2020 | Yes | Yes | No | USA |
C.diff | Toth et al. 53 | 2020 | Yes | Yes | Yes | USA |
C.diff | Van Kleef et al. 54 | 2016 | Yes | Yes | Yes | UK |
Carbapenem-resistant Klebsiella pneumonia | Changruenngam et al. 27 | 2022 | Yes | No | Yes | Not specified |
CRE | Bartsch et al. 23 | 2020 | Yes | Yes | No | USA |
CRE | Lee et al. 22 | 2020 | Yes | Yes | Yes | USA |
CRE | Lee et al. 21 | 2021 | Yes | Yes | No | USA |
CRE | Lee et al. 20 | 2021 | Yes | Yes | No | USA |
CRE | Lee et al. 19 | 2016 | Yes | Yes | No | USA |
CRE | Lin et al. 55 | 2021 | Yes | Yes | Yes | USA |
CRE | Toth et al. 56 | 2017 | Yes | Yes | No | USA |
E.coli | Knight et al. 57 | 2018 | Yes | No | Yes | UK |
E.coli | MacFadden et al. 58 | 2019 | Yes | No | Yes | Sweden |
E.coli | Talaminos et al. 59 | 2016 | Yes | Yes | Yes | Spain |
ESBL-producing Enterobacteriaceae | Godijk et al. 60 | 2022 | Yes | No | Yes | Netherlands |
ESBL-producing Enterobacteriaceae | Haverkate et al. 24 | 2017 | Yes | No | Yes | Netherlands |
ESBLproducing Klebsiella pneumoniae | Salazar-Vizcaya et al. 31 | 2022 | Yes | No | Yes | Switzerland |
Generic nosocomial bacteria | Bartsch et al. 18 | 2021 | Yes | Yes | No | USA |
Generic nosocomial bacteria | Belik et al. 61 | 2016 | Yes | No | Yes | Germany |
Generic nosocomial bacteria | Van Den Dool et al. 62 | 2016 | Yes | Yes | Yes | Netherlands |
Generic nosocomial bacteria | Van Kleef et al. 26 | 2017 | Yes | No | Yes | EU |
MRSA | Di Ruscio et al. 25 | 2019 | Yes | Yes | Yes | Norway |
MRSA | Gowler et al. 63 | 2022 | Yes | No | Yes | Not specified |
MRSA | Rocha et al. 64 | 2020 | Yes | Yes | No | Sweden |
Multidrug resistant Enterobacteriaceae | Piotrowska et al. 65 | 2020 | Yes | No | Yes | Germany |
Study Interventions
Studies that investigated improved surveillance and screening were more likely to include STACHs (n = 9, 31%) and LTCFs (n = 8, 28%) and less likely to include other community settings (n = 3, 10%). Similarly, only three of the eight studies that investigated the implementation of contact precautions and isolation included other community settings. The majority of interventions were evaluated using models of STACH and community interactions: Varying antimicrobial consumption and prescribing (n = 8, 28%), non-specified hospital transmission reduction (n = 6, 21%), non-specified community transmission reduction (n = 2, 7%), improved screening and surveillance (n = 9, 31%), improved HCW hygiene (n = 5, 17%), contact precautions and isolation (n = 8, 28%), environmental cleaning (n = 1, 3%), interfacility coordination (n = 3, 10%), decolonization treatment (n = 4, 14%), regional registry (n = 2, 7%), and vaccination (n = 2, 7%).
Population Characteristics
Besides disease states, most models assumed the population was homogenous and that behavior, susceptibility, and transmissibility were identical across all demographic segments of the population. However, six studies (21%) included age differences in the population, three studies included gender in their analysis, two studies (7%) included race, and one study (3%) included ethnicity (immigrant versus non-immigrant). No study investigated health disparities between populations.
Movement and Transmission Characteristics
All models included admissions and discharges, 12 studies (41%) included transfers between STACHs and other healthcare facilities, and 15 studies (52%) included readmission explicitly. Most studies (n = 22, 76%) assumed direct transmission and did not distinguish between patients and healthcare workers. We found three studies (10%) that considered HCW-mediated transmission in their analysis, where HCWs explicitly acted as vectors between patients. Haverkate et al. 24 included hospital-visitor interactions in their model. In three studies (10%) with healthcare workers, Di Ruscio et al.25 and van Kleef et al. 26 modeled HCWs identically to the general population (homogenous mixing), while Changruenngam et al.27 modeled transmission as only possible through HCW-mediated contact networks. Two studies (7%) with community transmission included zoonotic or foodborne transmission.29,30
In terms of incorporating movement, we found that two studies (7%) utilized data to investigate the impacts of movement on colonization in the community setting between high and low-prevalence geographical regions 24,31. Nine studies (31%) incorporated transfers between STACHs; among those studies, three of them modeled additional movement between STACHs and communities. Two studies (7%) simulated international travel outside the country of interest. Seven studies with spatially defined locations included geospatial features and considerations, including hospital or community settings (n = 5, 17%) and movement assumptions based on near proximity to hospitals (n = 2, 7%).
Role of Data and Parameterization
Among the 29 included studies, all used data to inform their parameters in some manner. There were 14 studies (48%) that incorporated contact or movement networks, for example transfers between healthcare facilities, contact rate matrix between populations. Parameters in most studies (n = 27, 93%) were informed by primary source data, such as a survey or electronic health records (EHR) data (e.g., admission rates, average length-of-stay), while 16 studies (55%) were parameterized by fitting their models to data, such as observed cases. In addition, sensitivity analyses (Latin-hypercube parameter sampling) were performed in 22 (76%) studies. Finally, for parameters that could not be directly informed, most studies utilized parameter values from other literature (n = 28, 97%) or expert opinion (n = 23, 79%); however, only two studies (7%) quantified the uncertainty of less-known parameters.