Emerging multidrug-resistant organisms (MDROs) are alarming antimicrobial resistance threats to public health and have been targeted for extensive intervention to contain their spread. 1,2 These organisms cause infections that are difficult to treat and are typically transmitted in healthcare settings, spreading rapidly in a region through patient movement. Reference Pacilli, Kerins and Clegg3,Reference Won, Munoz-Price and Lolans4 Facilities that care for high-acuity patients with longer stays (eg, long-term acute care hospitals (LTACHs) and ventilator-capable skilled nursing facilities (vSNFs)) may have a disproportionate role in regional spread as they are often associated with higher MDRO prevalence, increased transmission, and connections to many facilities via patient sharing. Reference Won, Munoz-Price and Lolans4–Reference Lin, Froilan and Lolans7
Current strategies to prevent the spread of emerging MDROs rely on adherence to core infection prevention and control (IPC) practices and detection of infectious individuals to place under enhanced IPC practices (eg, isolation and contact precautions in acute-care settings). Patients infected with an MDRO are often identified through clinical cultures. However, colonized individuals are comparatively more numerous, represent an important pathogen reservoir, and can be identified with screening. Reference Southwick, Adams and Greenko8,Reference Freedberg, Zhou and Cohen9 Taken together, implementing colonization screening and improving core IPC practices at healthcare facilities following the initial identification of a patient with an MDRO infection are predicted to slow further spread. Reference Slayton, Toth and Lee10 These measures are recommended in the CDC Interim Guidance to Contain Novel or Targeted MDROs. 2 Emerging MDROs may also be widespread in a facility or region even before the first isolate is detected from a clinical culture, similar to the detection of Candida auris through enhanced surveillance in Orange County, California. Reference Karmarkar, O’Donnell and Prestel11 Therefore, additional strategies to proactively identify MDROs and place individuals with an MDRO under enhanced IPC practices prior to identification of the initial clinical infection within a facility may effectively limit spread.
Prevention-focused approaches to detect and contain MDROs across multiple healthcare facilities at different stages of emergence have previously been used to control MDROs in the Siouxland region of the United States, Israel, and Orange County, California. Reference Karmarkar, O’Donnell and Prestel11–Reference Schwaber, Lev and Israeli13 The interventions common across these efforts were public health collaboration with regional healthcare facilities: proactive screening to detect MDRO colonization, improved use of recommended IPC practices such as adherence to contact precautions and hand hygiene, and enhancing colonization status communication between facilities. We used mathematical modeling to predict the effects of implementing varying preemptive interventions, directed at a subset of facilities, on regional MDRO prevalence.
Methods
We extended a previously published compartmental model of CRE transmission within a US state Reference Paul, Slayton, Kallen, Walters and Jernigan14 to include skilled nursing facilities (SNFs). The facility types in the model included acute-care hospitals (ACHs), critical-access hospitals (CAHs), LTACHs, psychiatric inpatient facilities, inpatient rehabilitation facilities, SNFs, and vSNFs. In addition, infectious individuals were separated into those detected and placed under targeted enhanced IPC practices and those who were not.
Regional healthcare network
As previously described, Reference Paul, Slayton, Kallen, Walters and Jernigan14 patient flow in a regional healthcare network was characterized by interfacility transfer rates, admission rates, and mean length of stay in each healthcare facility. Patient flow at a facility was characterized by the diversity of originating and destination facilities among admitted and discharged patients and by the rate of dispersal through discharge and subsequent transfer of patients, quantified using Shannon entropy diversity index Reference Shannon18 as a mean number of facilities that patients are admitted from or discharged to. To represent regions with diverse healthcare facility compositions and for interregional comparisons, we used historical transfer data from the Centers for Medicare and Medicaid Services (CMS). We modeled 4 exemplar regional healthcare facility networks: Illinois, California (Los Angeles and Orange Counties), New Jersey, and New York. We parametrized the model with CMS patient-level fee-for-service claims data for CMS beneficiaries and SNF occupancy data from the CMS Nursing Home Minimum Data Set. The SNFs were further subcategorized as vSNFs based on facility characteristics (having beds dedicated for residents requiring a ventilator) from the CMS Provider of Services files. The Dartmouth Atlas of Health Care Hospital Referral Regions (HRRs) were retained as community components of the network. 15
Disease progression and transmission
Recovery (pathogen clearance) rate and setting-specific CRE transmissibility were incorporated into a susceptible-infectious-susceptible framework as previously described. Reference Paul, Slayton, Kallen, Walters and Jernigan14 Infectious individuals were further delineated as those under enhanced IPC and those not under these measures. Person-to-person transmissions in the model occurred within a facility or in the community at rates characterized by setting-specific transmissibility. Facility-level CRE transmissibility values were based on analyses of data from the National Healthcare Safety Network (NHSN) and SNF and vSNF outbreaks and are similar to other reported findings. Reference Toth, Khader and Slayton16,Reference Lee, Bartsch and Wong17 Short-stay facilities (ACHs and CAH) had the highest transmissibility in the model, followed by LTACH, vSNF, and SNF. Other facility types and the community were assigned low transmissibility. Regional outbreaks were initiated with one infectious patient at the largest ACH in the region. Regional spread occurred through transfer of patients among facilities and the community.
Some facility-level characteristics were examined to elucidate the role of individual facilities in regional spread. Recovery rate, setting-specific transmissibility, and mean length of stay were used to estimate the probability of onward transmission, defined as transmission to at least one other individual, before recovery, either at that facility or at any facility subsequently transferred to, and a closely related measure, the facility reproductive number (R H ). An R H below the critical threshold (R H = 1) will not support endemicity at a facility without steady importation of infectious individuals, whereas facilities above that critical threshold can support endemicity. Further disease progression and transmission details are in the Supplementary Materials (online).
Interventions
Interventions were classified as (1) detection and tracking of infectious individuals (ie, admission screening, periodic prevalence surveys (PPSs), and interfacility communication) and (2) prevention of onward transmission from detected or tracked infectious individuals through enhanced IPC practices. Admission screening was applied to selected facility types, where patients from specified originating facility types were subject to screening. PPSs were implemented at selected facility types at specified rate-based frequencies (eg, 1 per 90 days, 1 per 180 days, etc) for each individual within the facility on a given day. Facilities participated in interfacility communication with an effectiveness set equally for each facility, representing the completeness of the data registered by participating facilities. Detected infectious individuals and registered patients admitted to facilities using the interfacility communication registry were placed under enhanced IPC practices. Each intervention bundle was modeled as being in place prior to CRE introduction or was initiated at a set time into an outbreak.
We first evaluated the effect of interventions for individual facilities with an R H above the critical threshold. We predicted the effect of screening interval and enhanced IPC practices effectiveness combinations necessary to reverse or prevent endemicity in single facilities. We also evaluated the reduction of endemic prevalence from varied screening frequencies as another measure of single-facility intervention impact.
We then explored the impact of intervention bundles on CRE prevalence and daily transmission over a 10-year period. We focused most interventions at LTACHs and vSNFs because these facilities have a disproportionate impact on regional MDRO spread. Reference Won, Munoz-Price and Lolans4–Reference Lin, Froilan and Lolans7,Reference Paul, Slayton, Kallen, Walters and Jernigan14,Reference Toth, Khader and Slayton16,Reference Lee, Bartsch and Hayden19 The impact of interventions was assessed against a baseline scenario without screening or enhanced IPC practices. The intervention layers included PPS (every 90 days at vSNFs and LTACHs); admission screening (at ACHs on patients transferred from LTACHs or vSNFs, or at LTACHs or vSNFs on all patients); interfacility communication (100% compliance); and enhanced IPC practices (reducing transmissibility in LTACHs by 50%–70%, in vSNFs by 25%–50%, in SNFs by 25%, and in ACHs and CAHs by 70%). To evaluate onward interfacility transmission including the regional impact of interventions, we used the Illinois network as the exemplar region due to its varied composition of healthcare facilities and metropolitan areas.
Results
Regional healthcare network
Patient flow at the facility level varied considerably by facility type within the regions modeled but was consistent across the 4 regions (Fig. 1). ACHs had the greatest median diversity of originating healthcare facilities (ie, where patients are transferred from) and SNFs and CAHs had the lowest, although CAHs were not represented in each region. Both SNFs and ACHs had wide ranges of originating facility diversity, and this was observed across each of the 4 regions. The diversity of originating facility types for vSNFs had a wider range in the New York and California regions than the Illinois and New Jersey regions. Short-stay facilities (ACHs and CAHs) had higher daily dispersal rates compared to LTACHs, SNFs, and vSNFs. In all regions, ACHs had the highest combined dispersal rates and originating facility diversity, whereas SNFs had the lowest. Among LTACHs and vSNFs in the regions analyzed, most admissions were from ACHs and most discharges were to SNFs (Fig. 2).
Figure 1. Facility-level patient flow represented by diversity of originating facilities and dispersal rate to healthcare facilities for Illinois, California (Los Angeles and Orange County), New Jersey, and New York networks. Note. ACH, acute-care hospital; CAH, critical access hospital; LTACH, long-term acute care hospital; SNF, skilled nursing facility; vSNF, ventilator-capable skilled nursing facility.
Figure 2. Comparison of the proportion of admissions from, and discharges to, LTACHs and vSNFs for the Illinois, California (Los Angeles and Orange County), New Jersey, and New York networks. Note. ACH, acute-care hospital; SNF, skilled nursing facility.
Disease progression and transmission
The probability of any onward transmission—an infectious individual at a facility infecting another individual before recovery, either at that facility or at any of the facilities subsequently transferred to—in the Illinois network varied by facility type (Fig. 3). This probability of transmission is a function of the facility’s connectivity, length of stay, and transmissibility. Infectious individuals at vSNFs and LTACHs had the highest probability of onward transmission even though transmissibility is higher at short-stay facilities, underscoring their importance in regional spread. The next subset of facilities (ACHs, CAHs, and SNFs) all had similar probabilities of onward transmission, followed by other facility types and the community (health referral regions or HRR) with the lowest probabilities of onward transmission.
Figure 3. The probability that an infectious individual at a facility in the Illinois network will infect another individual before recovery, either at that facility or at any of the facilities subsequently transferred to. Note. ACH, acute-care hospital; CAH, critical access hospital; LTACH, long-term acute-care hospital; Psych, psychiatric inpatient facility; Rehab, rehabilitation inpatient facility; SNF, skilled nursing facility; vSNF, ventilator-capable skilled nursing facility.
Impact of interventions
Reducing LTACH and vSNF R H below the critical threshold (R H = 1) requires different combinations of PPS frequency and IPC effectiveness depending on the R H value (Fig. 4A). Every additional week between screenings requires a balancing increase in necessary IPC effectiveness; the increase is greater in facilities with larger R H values. The relative reduction in endemic prevalence (compared to unmitigated endemic prevalence) at single facilities through PPS was highest for quarterly PPS, even at higher levels of unmitigated prevalence, compared to longer screening intervals (Fig. 4B). Potential endemic prevalence reduction is decreased for longer PPS intervals.
Figure 4. The impact of interventions at individual facilities where the facility reproductive numbers (RH) is >1. (A) Comparison of IPC effectiveness versus screening frequency at individual LTACH and vSNF facilities to reduce the facility reproductive numbers (RH) below the critical threshold (RH = 1). Areas above each RH line represent intervention combinations that reduce the facility RH below the critical threshold and no longer support endemicity in the specific facility. Areas below each line represent intervention combinations that continue to support endemicity. (B) The impact of periodic screening frequency on unmitigated endemic prevalence. Note. PPS, periodic prevalence survey.
Considered at a horizon of 10 years after importation, the impact was greatest with intervention bundles that included PPS and enhanced IPC practices at vSNFs and LTACHs. Targeted admission screening in ACHs, LTACHs, and vSNFs and improved regional interfacility communication had modest impact. Delaying intervention implementation by 3 years decreased regional prevalence 10 years after MDRO introduction but less than when interventions were started before CRE introduction. Implementing interventions in all vSNF and LTACH facilities was more effective than implementing in a subset of those facilities (Fig. 5).
Figure 5. All interventions: Periodic prevalence survey (PPS), infection prevention and control (IPC) intervention, admission screening (Adm Sc), and interfacility communication (IFC). IPC Intv: Targeted enhanced IPC practices intervention that reduces transmission by 70% in LTACHs and 50% in vSNFs. No IPC Intv: Estimated baseline level of targeted enhanced IPC practices (Transmission reduction: LTACH 50% and vSNF 25%). Note. Shaded areas represent the range of all possible outcomes from the intervention bundles. CRE prevalence shown as a percentage of endemic prevalence over a ten-year period in the Illinois network. (A) The effect of varying combinations of interventions started prior to the seeding of 1 infectious individual in the largest ACH. (B) The same trajectories as Figure 6A with the addition of interventions starting 3 years after initial infection. (C) The effect of removing varying percentages of the largest vSNFs and LTACHs prior to the seeding of one infectious individual. The intervention layers included PPS (every 90 days at vSNFs and LTACHs), admission screening (at ACHs on patients transferred from LTACHs or vSNFs, or at LTACHs or vSNFs on all patients), interfacility communication (100% compliance), and enhanced IPC practices (reducing transmissibility in LTACHs by 50%–70%, in vSNFs by 25%–50%, in SNFs by 25%, and in ACHs and CAHs by 70%).
Interventions focused mainly at LTACHs and vSNFs reduced daily transmissions across all facility types; ACHs had the largest decrease in daily transmissions following the implementation of interventions and under this scenario, vSNF were the facility type with the most transmissions. (Fig. 6). The reductions were uniform across time, except in the initial transient phase. Further results are available in the Supplementary Materials (online).
Figure 6. Left: Daily transmissions by facility type over a ten-year period following initial importation with no interventions in place. Right: Daily transmissions by facility type over a 10-year period following initial importation with all interventions in place. The intervention layers included PPS (every 90 days at vSNFs and LTACHs), admission screening (at ACHs on patients transferred from LTACHs or vSNFs, or at LTACHs or vSNFs on all patients), interfacility communication (100% compliance), and enhanced IPC practices (reducing transmissibility in LTACHs by 50%–70%, in vSNFs by 25%–50%, in SNFs by 25%, and in ACHs and CAHs by 70%). Note. ACH, acute-care hospital; CAH, critical-access hospital; HRR, community; LTACH, long-term acute-care hospital; Psych, psychiatric inpatient facility; Rehab, rehabilitation inpatient facility; SNF, skilled nursing facility; vSNF, ventilator-capable skilled nursing facility.
Discussion
Our analysis of regional healthcare networks parametrized with CRE progression and transmission data highlight strong associations—observed consistently among regions analyzed—between facility-level properties and a facility’s predicted role in the regional spread of MDROs. These properties can be leveraged to target MDRO prevention interventions to a relatively small number of facilities for regionwide benefit. Longer-stay facilities that provide high-acuity care influence regional MDRO prevalence disproportionate to their size. These influential facilities have the potential to drive regional transmission dynamics. Strategic implementation of interventions can leverage the outsized role of influential facilities in their respective regions to reduce regional prevalence and has been predicted to be effective in an alternate model. Reference Lee, Bartsch and Hayden19 In the United States, traits of influential facilities are commonly, though not exclusively, associated with LTACHs and vSNFs. Another key set of facilities comprises those that frequently receive individuals from the influential facilities. These highly connected facilities, typically ACHs and SNFs, are the next most likely contributors to onward transmission.
MDRO spread, as modeled, occurs by patient transfer between facilities, and through person-to-person transmission within facilities. Influential facilities contribute to MDRO spread principally through intrafacility transmissions. Therefore, interventions in influential facilities that combine expedited detection of incident infectious individuals through frequent PPS with enhanced IPC practices are likely to be effective at reducing overall transmission in the region. In contrast, enhancing detection (PPS) at the cost of prevention (IPC), or vice versa, may be ineffective. For highly connected facilities, identifying importation through interfacility transfers from influential facilities, which tend to have higher prevalence, has the greatest impact on regional spread. For ACHs, this corresponds to admission screening and interfacility communication, especially for individuals arriving from influential facilities. In our model, SNFs only received improved interfacility communication as a direct intervention but realized large indirect benefits from the collective regional prevention interventions, which prevent infectious individuals from reaching them.
In our analysis, strategic interventions primarily targeted at influential facilities reduce incidence in all facility types regionwide. This is consistent with previous modeling work suggesting that a centrally coordinated prevention approach has the potential to more comprehensively address the emergence and spread of MDROs than independent facility–based efforts and that focusing these efforts in LTACH and vSNF has the greatest regional benefit relative to other selection strategies. Reference Slayton, Toth and Lee10,Reference Lee, Bartsch and Hayden19,Reference Friedman, Carmeli, Walton and Schwaber20 Furthermore, although prevention strategies are predicted to have the greatest impact when interventions are bundled and implemented at all targeted facilities before an MDRO is identified in a region, substantial reduction in overall prevalence is still possible if implementation occurs after an MDRO has started spreading regionally.
Our use of a deterministic (ie, ordinary differential equation based) compartmental model parametrized with CRE data and average historical overall patient flow patterns allowed us to predict the effects of different intervention bundles on regional MDRO prevalence. However, this approach had several limitations. A deterministic model does not capture random fluctuations that may be important when outbreaks are still small, for example, extinction early in an outbreak. We parametrized the model with CRE data as a proxy for all emerging MDROs based on availability of robust data, but there may be pathogen-specific differences. Also, the patient flow network captures neither seasonal variations nor differences associated with the disease state of patients. However, there is evidence of CRE moving through ACHs, LTACHs, and long-term care facilities in a similar fashion to our network. Reference Tadese, Fujimoto, DeSantis, Mgbere and Darkoh21 In addition to the effect of interventions in the full Illinois network, we also analyzed interventions in other networks and found similar results to what we present here, suggesting that the findings from one network parametrized with CRE and the interventions described are reasonably generalizable to other regional networks.
Frequent screening produces the best results in the model but presents considerable resource and logistical burdens. Although screening capacity at public health laboratories has been steadily expanded, decentralized models that move screening closer to the point of care, and development of point-of-care tests for carbapenemase-producing organism colonization, could improve feasibility and uptake, particularly for admission screening. Current and future work to improve testing efficiency, such as by pooling surveillance cultures, or detecting initial facility introductions using pooled community samples such as wastewater, could optimize testing strategies and reduce resource burden. Identifying MDRO cases also creates the need for increased facility-level resources to prevent further transmission through the application of indefinite transmission-based precautions, highlighting a need for innovative approaches including decolonization strategies to substantially reduce pathogen burden and the likelihood of clinical infection and transmission to other patients. 22 Most interventions and resource burden described in this model are in LTACHs and vSNFs, but the entire region benefits through decreased CRE prevalence. These observations are important for informing implementation and demonstrate that positive regional outcomes may occur even if comparable benefits are not achieved in every facility type.
The CDC provided funding in 2016 to begin building the US public health infrastructure to support public health responses to contain the spread of antimicrobial resistant organisms. Through the Strengthening HAI/AR Program Capacity supplement of the American Rescue Plan of 2021, CDC outlined additional critical resources for state, local, and territorial health departments. The resources support a broad range of healthcare IPC activities and epidemiologic surveillance related activities to detect, monitor, mitigate, and prevent the spread of severe acute respiratory coronavirus virus 2 (SARS-CoV-2), healthcare-associated infections, and antimicrobial-resistant organisms in healthcare settings. These additional funds enable jurisdictions to continue to increase capacity and expertise to directly supply IPC, screening, and communication resources to facilities to address the threat of antimicrobial resistant organisms through strategies outlined here.
Our results and those from other models suggest the potential for important decreases in MDRO prevalence following implementation of regional prevention interventions focused on influential facilities. However, whether the will and capacity exist to widely implement these interventions remain to be seen. Success of the interventions highlighted in our model will require recognition of MDROs as a regional concern and commitment to action from across the healthcare spectrum; similar lessons are evident from previously described campaigns in Siouxland, Israel, and Orange County. Reference Karmarkar, O’Donnell and Prestel11–Reference Schwaber, Lev and Israeli13 This model helps quantify the expected effects of regional prevention programs and adds to discussions of whether their level of impact justifies the effort and resources needed for implementation. Without support from a wide range of stakeholders for the intensive detection and IPC measures included in our models, they are unlikely to be successful.
In conclusion, we modeled the impact of implementing intervention bundles to prevent MDRO spread across healthcare facilities within a jurisdiction. Focusing efforts in the subset of facilities that have greatest influence on regional prevalence led to substantial reductions in MDRO prevalence across the facility network. This reduction in prevalence is predicted to slow spread but not eliminate MDROs entirely, and additional interventions will be needed.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/ice.2023.278
Acknowledgments
The authors thank Elizabeth Soda, Alex Kallen, and John Jernigan. The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.
Financial support
No financial support was provided relevant to this article.
Competing interests
All authors report no conflicts of interest relevant to this article.
