We conducted a retrospective cohort study of patients admitted to a medical intensive care unit (MICU) and a comprehensive transplant unit (CTU) from July 1, 2016 to June 30, 2017. Although the CTU is not considered an intensive care unit (ICU), it deliveres ICU-level care and, along with the MICU, has private patient rooms and uses contact precautions (gown and gloves) for those with a history of multidrug-resistant organisms or recent (<6 months) international hospitalization.²³ All patients admitted to the MICU and CTU had perirectal ESwabs^™ (Copan Diagnostics, Inc.) collected at unit admission and weekly thereafter as part of a longstanding vancomycin-resistant Enterococci surveillance program. As part of a research protocol (previously described ²⁴), residual media from this surveillance program, in addition to those cultures resulting from clinical care, were analyzed for the presence of CRO: Enterobacterales resistant to ertapenem, meropenem, and/or imipenem, which were classified as CRE, as well as glucose non-fermenting (NF) Gram-negative bacilli resistant to meropenem and/or imipenem, which were classified as NFCRO. CRO were further identified as carbapenemase-producing-(CP) versus non-CP- CRO by the phenotypic modified carbapenem inactivation method.^25,26

The main outcome under study was incident acquisition of CRO, which was defined as 1) a patient that had a negative clinical or surveillance culture at unit admission and 2) a clinical or surveillance culture that was obtained more than two days after unit admission and grew a CRO. Each positive culture was assigned to either potentially transmission-mediated or not potentially transmission mediated.¹⁸ Potentially transmission-mediated CRO acquisition was assumed when a patient grew a CRO of the same species as another patient in the same unit that had overlapping days of care. Not potentially trnamission mediated was assumed when no other patient with the same CRO species was in the unit. Patients could be incident cases more than once if they acquired a CRO of different species and met the above criteria. Once a patient became incidentally colonized or infected for a given CRO species, he (or she) was no longer included in the analysis of incident acquisition risk, but he (or she) remained in the study as contributing to the risk of transmission to others.

Trained research staff rounded on each unit during weekdays to determine whether patients were on contact precautions for any indications (e.g., methicillin-resistant S. aureus, Clostridioides difficile, influenza virus). Staff from the Department of Hospital Epidemiology and Infection Control measured HCW hand hygiene and unit environmental cleaning compliance. The monitoring method to assess unit cleaning practice was based on fluorescent markers.²⁷

Patient encounter data were retrospectively collected using bulk extraction methods from the hospital’s electronic health records (EHR) system. Patient data included demographic information, laboratory test results, medication administration, and room assignments.

Time-stamped in-room visits by HCWs were extracted from the electronic health records system to estimate a patientś social network. That is, HCW-mediated connections with other patients in the hospital ward. We considered two patients, say patients A and B, were epidemiologically linked if the same HCW visited their rooms within a 60 minutes period. HCW interactions with patients were estimated through timestamps of in-room medication administrations, laboratory specimen collections, assessments, and other in-room routine care tasks based on the methodology developed by our team and presented in ¹⁶.

The primary statistical analysis involved the specification of a probabilistic model describing the risk of a susceptible patient becoming colonized or infected as a function of measured attributes of the individual, surrounding patients, and the unit environmental cleaning levels. A Bayesian hierarchical logistic regression model was fitted with the Markov Chain Monte Carlo method (see Parameter Estimation in the Supplement). Outputs from the model were adjusted odds ratios (adj. OR) and 95% credible intervals (CrI). The OR and CrI estimates were reported and used to estimate the absolute risk reduction in susceptible individuals. All statistical analyses were performed in R version 3.5.1, using the freely distributed statistical package BayesianTools version 0.1.6.²⁸

The probabilistic model had two levels, with information about transmission-mediated sources at level-1 and other sources at level-2 (see Model Formulation in the Supplement). To control for varying length-of-stay among patients, we modeled the risk of CRO acquisition per day. All model levels were assessed simultaneously to disentangle spatiotemporal patterns of each outcome, manifesting across each acquisition mechanism with respect to the individuals’ characteristics, their HCW-mediated social network, and the broader unit characteristics.

Since carbapenemase-producing (CP)-CROs were considered the highest threat due to their resistance to multiple antbiotic classes and potential for plasmid transmission,^24,29 the acquisition dynamics of CP- versus non-CP-CRO were analyzed independently. The clinical and surveillance data did not allow us to ascertain the exact moment when a patient acquires a new CRO (unobservable clinical event). To evaluate the possibility that unobserved events might explain associations, we also examined if randomization of the actual acquisition date, between the dates of the last known CRO-negative culture and the CRO-positive culture, substantially impacted the direction and size of the estimated effects (see Extension of Main Results in the Supplement).

We evaluated our results’ robustness by varying the time window of 60 minutes that established a patient’s HCW-mediated social network from 15 minutes to 12 hours, the latter was included because it is consistent with the maximum length of most HCWs’ shifts. We also studied acquisition dynamics at each hospital ward in separate models (see Robustness of Main Results in the Supplement).

The study cohort included 2,193 unit admissions (1,715 unique patients) to the MICU and the CTU (Table 1). Patients were predominantly adults between 45 and 59 years old (724/2,193, 35%), male (1,131/2,193, 52%), black (1,085/2,193, 50%), non-Latino (2,098/2,193, 96%), and Maryland residents (1,893/2,193, 86%) who were typically admitted as an emergency or urgent case (1,951/2,193, 89%). Patients in the cohort were connected through 216,069 distinct HCW-mediated connections, representing a daily average of 591 HCW-mediated contacts between patients. Most patients in the cohort were on contact precautions for CRO or some other antibiotic-resistant organism (87.4% and 88.9% amongst non-incident and incident CRO admissions, respectively). Susceptible patients had on average 4.8 daily connections through HCWs with CRO-positive patients on contact precautions and 1.9 with CRO-positive patients not on contact precautions.

A total of 126 out of 2,193 (5.8%) unit visits had a negative swab on admission and at least one positive swab more than 2 days after admission and were classified as incident for CRO acquisition (Table 1).³⁰ Amongst the 126 visits, 120 (93.0%) were linked with potentially transmission-mediated CRO acquisition events, because the patient grew a CRO of the same species as another patient in the hospital ward, while all of the 126 visits (100%) were detected to have at least one CRO acquisition event with no evidence of transmission. We recall the reader that patients could be incident cases more than once if they acquired a CRO of different species an met the criteria out outcome definition criteria. Non-incident and incident CRO patients had similar baseline demographics at unit entry. However, incident CRO individuals were more likely to have HCW-mediated contacts with CRO-positive individuals who were not on contact precautions (median (IQR), 1 (1–2) vs. 2 (1–2)). Compared to non-incident CRO patients, incident CRO patients were more likely to receive carbapenems (10.0% vs. 47.6%).

The use of contact precautions for CRO-positive patients was associated with reduced rate and odds of CRO acquisition among susceptible individuals (7.4 vs. 93.5 per 1,000 patient-days at risk; adj. OR 0.03, 95% CrI 0.01–0.17). The estimated absolute risk reduction of contact precautions for CRO-positive patients, compared with CRO-positive individuals without contact precautions, was 9.0% (95% CrI 7.6–9.2%), corresponding to 3 events prevented per 1,000 patient-days at risk. For susceptible individuals, recent exposure to carbapenems (last seven days) was the primary driver of CRO acquisition (adj. OR 2.38, 95% CrI 1.70–3.29) (Table 2).

The mechanism of resistance appeared to be important. When the cohort was restricted to the acquisition of CP-CRO (versus non-CP-CRO), recent carbapenem exposure in susceptible patients had no statistically meaningful (i.e., Bayesian significance) relationship with incident CP-CRO acquisition (adj. OR 2.19, 95% CrI 0.96–4.56), whereas carbapenem exposure was significantly asscociated with non-CP-CRO acquisitions (adj. OR 2.01, 95% CrI 1.33–2.95). The use of contact precautions for CRO-positive patients, however, was linked with reduced rate and odds of CP- (48.0 vs. 417.6 per 1,000 patient-days at risk; adj. OR 0.04, 95% CrI 0.01–0.19) and non-CP-CRO acquisition (12.1 vs. 157.8 per 1,000 patient-days at risk; adj. OR 0.03, 95% CrI 0.01–0.16) among susceptible patients (Table S1).

We evaluated the robustness of our results to unobservable clinical events, i.e., our inability to precisely determine the actual CRO acquisition date during the patient stay (Table S2). We randomized the CRO acquisition date in our models within a reasonable time interval, from the last known CRO-negative culture and the CRO-positive culture. As presented in Table S2, the direction and magnitude of effects shown in Table 2 remained similar.

We also investigated the robustness of our results to our definition of a patient’s social network: HCW-mediated contacts with other patients in the ward. When modifying the time-lapse of 60 minutes from 15 minutes to 12 hours, we found that the magnitude and direction of the contact precautions’ effect were virtually unchanged (adj. OR ranges between 0.03 and 0.04). However, we found much uncertainty (95% CrI 0.01–0.17) within the posterior coefficient estimation (Table S4), signaling potential workflow differences between wards. To determine if workflow differences explained some of the variability in the effect of contact precautions, we fitted separate models for each ward. We found no significant differences in the effect of contact precautions between the MICU (adj. OR 0.03, 95% CrI 0.01–0.13) and the CTU (adj. OR 0.05, 95% CrI 0.01–0.28).