Technical Appendix for Stop Spread of Antibiotic Resistance

This Technical Appendix is associated with the CDC Vital Signs report published in August 2015. To learn more about this report, please visit www.cdc.gov/vitalsigns/stop-spread/.

Technical Appendix:

Projected Incidence of Infections and Deaths Due to Antibiotic Resistant Healthcare-Associated Infections

Projections of infections and associated deaths in the US were built upon methods described previously  (1). For multidrug-resistant (MDR) P. aeruginosa and carbapenem-resistant Enterobacteriaceae (CRE) we utilized yearly percent of resistant infections reported to the National Healthcare Safety Network (NHSN) 2009-2013. While estimates of CRE, Clostridium difficile infections (CDI), and invasive methicillin-resistant Staphylococcus aureus (MRSA) included healthcare-associated infections (HAIs) occurring in a variety of healthcare settings, MDR P. aeruginosa projections were limited to HAIs identified during a hospitalization. CRE estimates also included an inflation factor derived from the Emerging Infections Program’s (EIP) Multi-Site Gram-Negative Bacilli Surveillance Initiative (MuGSI) to account for healthcare-associated community-onset infections. The ratio of healthcare-associated community-onset CRE to hospital-onset CRE was 3.57. Our estimate of invasive methicillin-resistant S. aureus (MRSA) utilized historic estimates of healthcare-associated infections from the EIP. CDI estimates included hospital-onset infections and community-onset healthcare-associated infections requiring hospitalization. Mortality was derived from one overall attributable mortality estimate from the published literature which included all infections except CDI (2). For CDI, we used a specific attributable mortality estimate for hospital-onset CDI and community-onset healthcare-associated infections requiring hospitalization in non-epidemic settings (3, 4). We estimated a 30% reduction in invasive MRSA and a 50% reduction in infections with the other pathogens over 5 years with equal annual reductions relative to the 2014 incidence. Projections of all infections included age-stratified U.S. Census Bureau population growth estimates (5).

Estimating Effect of a Coordinated Augmented Approach

The first model assessed the impact of the coordinated approach in a simulated network of 10 healthcare facilities, including 4 acute care hospitals (1 long-term acute care hospital (LTAC)) and 6 free-standing nursing homes serving adult patients. Patient “agents” were introduced into the network and individually tracked during inpatient facility stays and community stays between discharge and readmission. Assumptions characterizing the flow of patients among the facilities were made based on combined data from U.S. Department of Veterans Affairs (VA) facilities and non-VA facilities. Probabilities of discharges, readmissions, and transfers of patients to and from facilities, as well as distributions of lengths of stay in facilities and in the community between discharge and readmission, were calibrated based on actual patient data from the VA (6). VA-based probabilities of transfers to acute care hospitals were adjusted upward to account for the assumed presence of an LTAC hospital, which are rare in the VA, and LTAC patient discharge assumptions were based on data from nine non-VA hospitals in Arkansas (7). Based upon the literature and compared to CRE-negative patients, CRE-positive patients admitted from the community to acute care hospitals were assumed to be approximately 14 times more likely to be discharged to a long term care facility (8). All community-discharged CRE-positive patients were approximately one and a half times more likely to be readmitted if CRE positive compared to CRE negative, based on VA data for MRSA (9). The initial conditions of the model assumed that there were no CRE cases in facilities nor among recently discharged patients. At time zero, introductions of CRE-positive patients into the facility network from the general community began at a rate of approximately one every two weeks.

Within each facility, colonized patients were assumed to pose a constant per-time risk of transmission to all susceptible patients in the same facility at the same time. The values of these facility-type-specific transmission rate constants were calibrated to achieve expected 3-year facility prevalence within ranges observed in the literature for different facility types. It was assumed that no transmission occurred from CRE-positive patients while they were in the community. Colonized patients became decolonized after a random exponentially distributed time with mean 387 days (10). Progression from colonization to infection and symptoms and the clinical culture testing process were not modeled explicitly; the entire process from undetected colonization to positive clinical culture was represented by facility type-specific clinical detection rate parameters, which were calibrated to achieve target proportions of colonized inpatients clinically detected before discharge. Detected patients were assumed to be placed under contact precautions that decreased transmission rates from these patients by 50%. When detected patients were transferred or discharged and readmitted to the same or a different facility later, their detection status was not necessarily retained during the next stay, due to imperfect recordkeeping and/or communication. The probability was 75% for retaining positive detection status upon readmission to the same facility after a single community stay, 15% upon direct transfer to new facility, and 0% upon readmission to a different facility after a community stay. Selective changes to these assumptions were made and additional features added as part of augmented efforts, described next.

The augmented efforts implemented independently included one random, short-term acute care hospital beginning augmented activities after it identified 5 patients infected or colonized with CRE from clinical tests within 30 days. At this time, the facility began testing patients on admission and once weekly for CRE carriage (and placing them under contact precautions if positive), improving the effectiveness of contact precautions such that the transmission rate from detected patients decreased by 75% instead of 50%, and improving recordkeeping such that the probability of retaining positive detection information upon readmission to the same facility increased from 75% to 90%. Screening tests were assumed to have 80% sensitivity and 100% specificity. It was assumed that screening tests were not performed on patients already identified as positive.

The coordinated augmented approach began with improved recordkeeping and communication of positive detection information across all facilities: from 75% to 90% upon readmission to the same facility, from 15% to 90% upon direct inter-facility transfer, and from 0% to 70% upon readmission to a new facility after an intervening stay in the community. These recordkeeping and communication improvements were implemented from the beginning of each coordinated simulation, to reflect the fact that they could be implemented by a regional authority. Additional coordinated activity began when one hospital reached its trigger, and interventions differed by whether the first facility to reach its trigger was an acute care hospital or an LTAC. When one acute care hospital reached its trigger of 3 positive clinical cultures in 30 days, the following actions were taken: 1) That acute care hospital began weekly testing of all patients staying longer than 1 week. 2) All acute care hospitals began admission testing of patients with a history of treatment in the triggering facility within 60 days. 3) All facilities improved implementation of contact precautions from 50% to 75%. When one LTAC reached its trigger of 4 positive clinical cultures in 30 days, the following actions were taken: 1) LTAC began weekly testing of all patients stating longer than one week. 2) LTAC began admission testing of all patients. 3) All acute care hospitals began admission testing of patients with a history of treatment in the triggering facility within 60 days. 4) All facilities improved implementation of contact precautions from 50% to 75%. When a second hospital reached its trigger (either the LTAC or an acute care hospital), all acute care hospitals began admission testing all patients with recent care in any facility in the network.

One thousand simulations covering a 5 year time span following introduction of CRE into the region were run for each of the three scenarios: baseline activity with no augmented intervention, augmented efforts implemented independently at individual subsets of facilities, and coordinated augmented approach across a health care network. The mean values for number of acquisitions, cumulative prevalence, and surveillance tests were calculated. Model input parameters and calibration targets are detailed in Table 1.

The second model assessed the impact of the coordinated approach in a larger region. We utilized the Regional Healthcare Ecosystem Analyst (RHEA) software platform and imported actual data (e.g., admissions, length-of-stay, and probabilities of inter-facility transfer) from Orange County, California (2011-2012 patient data) to generate a simulation model of all 28 acute care hospitals (5 LTACs) and 74 free-standing nursing homes serving adult patients in Orange County (11, 12). The model included detailed virtual representations of each facility including patient beds in wards and units with individual virtual patients moving among and within each facility. This model, previously used to simulate MRSA transmission and norovirus transmission (13-19), was re-parameterized to simulate the regional spread of CRE. The initial conditions of the model assumed that there were no CRE cases in facilities nor among recently discharged patients. Contact precautions were assumed to reduce CRE transmission by 50%. Contact precaution status transferred between hospitals on direct transfer of a patient; patients returning to a hospital where he/she was previously placed under contact precautions were automatically placed under contact precautions on admission. In nursing homes, contact precautions were used for 10 days for residents with a CRE infection. We assumed that 50% of known CRE in nursing homes involved an infection. Model input parameters and calibration targets are detailed in Table 2. A detailed comparison of the independent augmented efforts and coordinated approach for each model is found in Table 3.

We simulated three different approaches to CRE control. The first consisted of baseline activity in which CRE carriers could only be identified if CRE had been isolated from routine clinical cultures and were placed under contact precautions. In addition to detection via clinical isolates, two other experiments (independent efforts and the coordinated augmented approach) simulated an intervention consisting of active detection testing for CRE on any inter-facility transfer with use of contact precautions upon a positive test result. This intervention was implemented at specified trigger points depending on the scenario. The independent efforts were triggered once CRE had been identified in 10 patients in that hospital from routine clinical cultures, and each hospital had a 15% chance of implementing screening once this trigger was reached. The coordinated augmented activity was triggered once CRE had been isolated from routine clinical cultures from patients at any 10 hospitals in the network; when this network trigger was met, all hospitals implemented CRE screening of inter-facility transfers and use of contact precautions for carriers. Each simulation run consisted of running the model over a 15 year time period; each scenario was run 50,000 times to see how the results varied and mean values are presented.

 Top of Page

Table 1: Model input parameters and calibration targets, 10 facility region

^†Target prevalence at year 3 in the model with no augmented activity. Facility type-specific transmission rate constants were adjusted to hit the targets.

*Target proportions in model with no augmented activity. Rate constants for facility type-specific progression to clinical detection were adjusted to hit the targets.

 Top of Page

Table 2: Model input parameters and calibration targets, 102 facility region in Orange County, California

†Target prevalence at year 7 in the model from the time of CRE introduction into the model.

*Assumes a linear loss for the 70% of carriers that experience a loss rate

^Average increase in length of stay countywide; based on observed length of stay for patients with vancomycin resistant enterococcus (VRE)

1. And personal communication with Michael Lin and Mary Hayden (Chicago’s CRE experience) and Project MAPP data (OC’s CRE experience)
2. Based on unpublished data from Los Angeles County.

 Top of Page

Table 3: Detailed Comparison of Interventions Tested in the Models

Independent Augmented Efforts

Coordinated Approach

†Effectiveness of contact precautions was 50%.

*On readmission to the same facility and direct transfer to another facility information on CRE positive patients is perfectly retained.

 Top of Page

 Top of Page