Factors Associated with SARS-CoV-2 Infection in Resident Physicians and Fellows in New York City During the First COVID-19 Wave

Risk factors for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection are not well-defined in resident physicians and fellows (trainees). We aimed to identify sociodemographic, occupational and community factors associated with SARS-CoV-2 infection among trainees during the first wave of the coronavirus disease 2019 (COVID-19) pandemic in New York City (NYC). In this retrospective cohort study, we administered an electronic survey between June 26 and August 31, 2020 to trainees at the Mount Sinai Health System in NYC to assess risk factors for SARS-CoV-2 infection between February 1 and June 30, 2020. We used Bayesian generalized linear mixed effect regression and structural equation models to examine associations. SAR-CoV-2 infection was determined by self-reported IgG antibody and reverse transcriptase-polymerase chain reaction results and confirmed with laboratory results. Among 2354 trainees invited to participate, 328 (14%) completed the survey and reported test results. The cumulative incidence of SARS-CoV2 infection was 20.1%. Assignment to medical-surgical units (odds ratio [OR], 2.51; 95% CI, 1.185.34), and training in emergency medicine, critical care and anesthesiology (OR, 2.93; 95% CI, 1.246.92) were independently associated with infection. Deployment to care for unfamiliar patient populations was protective against infection (OR, 0.16; 95% CI, 0.03-0.73). Community factors were not significantly associated with infection after adjustment for occupational factors. Our findings may inform tailored infection prevention strategies for trainees responding to the COVID-19 pandemic.


Introduction
New York City (NYC) was an early epicenter of coronavirus disease 2019  in the U.S. [1]. Following identification of the first case in NYC on March 1, 2020, incident daily cases rose to a peak of 8,593 cases on April 10, 2020 and gradually declined to a stable incidence of approximately 300 cases per day by June 2020 [2]. Healthcare workers (HCWs) experienced early unmitigated occupational exposure to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) until approximately mid-March 2020, prior to implementation of standardized infection prevention protocols including universal masking, patient symptom screening, and ubiquitous telehealth, and before risk factors for transmission in healthcare settings were identified [3][4][5][6]. Reported risk factors for SARS-CoV-2 infection in HCWs include hospital department, healthcare profession, personal  Did not complete the survey N=27 Figure 2. Risk factors hypothesized to be associated with SARS-CoV-2 infection in physician trainees.

Participant enrollment and survey response
2354 eligible trainees with valid contact information were invited to participate through email and text message links to the electronic consent and survey on June 26, 2020. Up to five reminder invitations were sent to non-responders through August 31, 2020. Of those initially contacted, 391 trainees (17%) responded to the invitation and 360 (15%) completed the survey (Figure 1).
To increase participation and to promote equitable representation of trainees from all affiliated hospitals, a subset of eligible trainees (n=281, 11%) selected using proportionate random sampling and stratified by hospital within the Mount Sinai Health System. Valid contact information was available for 267 selected trainees, of whom 72 (27%) consented to participate in the study. The randomly selected sample with a higher response rate was used to ascertain potential selection bias in the overall study sample.

Institutional process for employee COVID-19 testing
On March 6, 2020, Mount Sinai's Employee Health Services (EHS) established an online registry for employees to voluntarily report high-risk exposures and daily symptoms of COVID-19. Healthcare providers counseled registered employees on symptom monitoring and coordinated testing and clearance for return to work. RT-PCR swabs and IgG antibody testing were available to all symptomatic employees on April 7, 2020, and to asymptomatic employees by May 6, 2020. Sensitivity and specificity of the Mount Sinai Hospital Clinical Laboratory COVID-19 ELISA antibody test is 92.5% (95% CI: 80.1%-97.4%) and 100% (95% CI: 95.1-100%), respectively [16]. Sensitivity and specificity of the Roche Cobas RT-PCR test offered is 100% [17].

Assessment of SARS-CoV-2 infection
We ascertained SARS-CoV-2 infection status by self-reported test results and categorized the results as positive (by IgG antibodies, RT-PCR, or both), negative (by IgG antibodies, RT-PCR, or both) or never tested. To reduce the likelihood of differential  [18] we excluded participants who denied testing at the time of survey completion, and for whom there was no record of an IgG antibody result through July 15, 2020 in the EHS COVID-19 registry (n=32). Among a subset of 199 participants who consented to review of test results, there was 100% agreement between self-reported and laboratory-confirmed results.

Assessment of potential risk factors for SARS-CoV-2 infection
The survey collected information regarding sociodemographic, occupational and community factors hypothesized to be associated with SARS-CoV-2 infection (Figure 2). The complete survey is available in Supplemental Table 1. Occupational factors included department of work during the study period, exposure to patients with confirmed or suspected (i.e., persons under investigation or PUI) SARS-CoV-2 infection, unprotected contact (without N95, eye shield, gown, or gloves) with confirmed cases or PUI, performing or attending AGPs, and factors related to deployment. Deployment was defined as a temporary assignment away from usual clinical duties to assist in the COVID-19 surge response, which could have required relocation to an affiliated but unfamiliar hospital within the health system, department, or change in usual patient population. For the analysis, we categorized trainees by specialty including 1) primarily non-procedural specialties, 2) high-risk, primarily procedural specialties and 3) surgical specialties (Supplemental Table 2).
Community factors assessed included primary residence (zip code), contact for more than 10 minutes with an individual with confirmed or suspected COVID-19 outside of work, number of adults and children in household, and primary mode of transportation to work and non-work locations.

Statistical analysis
Sociodemographic, occupational and community variables were compared between groups using Fisher's exact test for categorical variables and the Wilcoxon rank-sum test for continuous variables. Variables with a p-value < 0.30 in the bivariate analysis were included in Bayesian Generalized Linear Mixed Effect Regression (BGlmer) to estimate the adjusted odds of SARS-CoV-2 infection. We used BGlmer to stabilize estimates for predictors with zero or small numbers of observations in specified subgroups according to SARS-CoV-2 infection status. [19] We first tested associations in BGlmer models that were separately adjusted for sociodemographic factors (Model 1), occupational factors (Model 2) and community factors (Model 3). To control for over-adjustment, variables with a p-value < 0.30 after backward elimination in the BGlmer model were retained in the final adjusted models. Finally, the model was simultaneously adjusted for sociodemographic, occupational and community factors (Model 4) to test whether associations remained robust in a fully adjusted BGlmer model.
We used Structural Equation Models (SEMs) to evaluate associations of sociodemographic, occupational and community factors with SARS-CoV-2 infection. Three unobserved latent sociodemographic, occupational and community functions were estimated using variables associated with SARS-CoV-2 infection in the BGlmer analysis (Table 2) and regressed to SARS-CoV-2 test result (Table 3). All SEMs were fitted using diagonally weighted least squares and a probit link function. [20] The root mean square error of approximation (RMSEA) for the final SEMs was < 0.05. Sensitivity analyses included: 1) exclusion of trainees with RT-PCR test results but no IgG antibody results (n=314); 2) model adjustment for date of the SARS-CoV-2 test if available (n=186); and 3) comparison of main characteristics between the analysis population (n=328), the trainees from the randomly selected sample who reported SARS-CoV-2 test results (n=62), and all initially eligible trainees (n=2543). All statistical analyses were conducted using R version 3.6.1. Missing data for covariates (approximately 1%) were imputed using random forests with the Multivariate Imputation by Chained Equations R package [21]. The SEM analysis was conducted using the "lavaan" R package [22].
Assignment to work in an emergency department (ED) or intensive care unit (ICU), independent of deployment, was not statistically significantly associated with infection in the bivariate analysis. Similarly, SARS-CoV-2 infection was less frequent among trainees who worked in ambulatory clinics and on telemedicine compared to those who reported never working in these settings, whereas infection more likely among trainees who performed AGPs and who reported at least once instance of unprotected contact without N95, eye shield, gown, or gloves for over 10 minutes with a confirmed COVID-19 patient or PUI (Table 1). However, these associations were attenuated and not statistically significant after adjustment for other occupational factors (Table 2).

Community factors and SARS-CoV-2 infection
After multivariable adjustment for community factors (Table 2, Model 3), contact for more than 10 minutes with an individual with confirmed or suspected COVID-19 outside of work (OR, 2.38; 95% CI, 1.14-4.98), and use of public transit (subway or bus) as the primary mode of transportation to non-work locations (OR, 2.25; 95% CI, 1.01-5.01) were associated with increased odds for infection. Primary residence in boroughs of NYC outside of Manhattan was associated with decreased odds of infection in the bivariate analysis, however, associations of community factors with SARS-CoV-2 infection were attenuated and not statistically significant after adjustment for occupational factors (Table 2, Model 4).

Structural equational model
The SEM analysis (Table 3) produced concordant results with the multivariable adjusted regression ( Table 2). The likelihood of SARS-CoV-2 infection was statistically significantly increased with an overall increased in the latent function of occupational factors. This association remained after adjustment for sociodemographic and community latent functions (adjusted SEM estimate 0.35; 95% CI, 0.15-0.54). The magnitude of the associations of sociodemographic and community factors with SARS-CoV-2 infection was attenuated and not statistically significant compared with occupational factors. Table 3. Adjusted effect estimates for associations of sociodemographic, occupational and community latent functions with SARS-CoV-2 infection.

Sensitivity analysis
Associations in the multivariable adjusted models remained statistically significant after excluding trainees with RT-PCR results but who did not report IgG antibody results, and after adjustment for the date of the SARS-CoV-2 test (Supplemental Table 4). Trainees based at Mount Sinai Hospital, the largest of all affiliated sites, were overrepresented in the analysis sample (64% vs. 55% among all initially eligible trainees). We did not observe additional statistically significant differences between the final analysis sample compared with eligible trainees, or with the randomly selected sample (Supplemental Table 5).

Discussion
In this study of physician trainees in a large NYC-based healthcare system, assignment to inpatient medical-surgical units and training in high-risk procedural specialties, including EM, anesthesiology, and critical care, were statistically significantly associated with SARS-CoV-2 infection. Assignment to unfamiliar hospital sites or clinical responsibilities was not associated with SARS-CoV-2 infection, and assignment to unfamiliar patient populations was associated with decreased risk of infection, suggesting that deployment of trainees was a safe mitigation strategy during the first wave of COVID-19 in NYC. Associations of community factors and SARS-CoV-2 infection were not statistically significant after adjustment for occupational factors, indicating that infection in trainees was largely attributable to occupational exposures.
The cumulative incidence of SARS-CoV-2 infection by June 30, 2020 was 20.1%, similar to reported seroprevalences in other HCW subgroups and the general population of NYC during this period [23,24]. The NYC Department of Health and Mental Hygiene reported a 22.7% seroprevalence among 5101 grocery store customers tested between April 19-28, 2020, suggesting that the prevalence of SARS-CoV-2 infection among trainees did not exceed the frequency of infections in the general population of NYC during the initial COVID-19 wave [25].
In this study, assignment to inpatient medical-surgical units was a statistically significant risk factor for SARS-CoV-2 infection, contrary to prior studies of HCWs, in which the minority (less than 10%) were physicians [23,24]. Medical-surgical units may have been less familiar to trainees who, prior to the COVID-19 pandemic, spent a greater proportion of duty hours in ambulatory care sites, or in operating rooms or procedural environments. Among trainees from surgical and primary care specialties, decreased familiarity with routine infection prevention protocols specific to medical-surgical units may further explain our findings. Additionally, caring for PUI in medical-surgical units may have diminished the urgency of adherence to optimal infection prevention protocols, compared with caring for confirmed COVID-19 patients. Finally, working in an ED or ICU was not associated with SARS-CoV-2 infection in this study, consistent with prior reports of HCWs in the greater New York area [23,24].
Trainees in high-risk procedural specialties were at increased risk for SARS-CoV-2 infection in this study, consistent with prior studies. Breazzano et al. reported a higher frequency of SARS-CoV-2 infections among trainees in EM and anesthesiology compared with other specialties [10]. Trainees in EM, anesthesiology and critical care routinely perform endotracheal intubation, and likely had unmitigated exposure to aerosolized virus from undiagnosed COVID-19 patients early in the study period, prior to implementation of routine infection prevention protocols [26][27][28].
Deployment to unfamiliar hospital sites and clinical responsibilities was not a statistically significant risk factor for SARS-CoV-2 infection, despite limited time for patient surge planning [14,15]. Moreover, we found that deployment to care for unfamiliar patient populations was associated with decreased adjusted odds of infection. Among survey respondents, pediatrics trainees most frequently reported a patient population change, most commonly to care for adult patients in ED or ICU environments. Deployment strategies differed according to department in the Mount Sinai Health System, and the Department of Pediatrics and Mount Sinai Hospital deployed trainees on a voluntary basis. It is plausible that trainees who cared for unfamiliar patient populations may have performed more administrative tasks and had fewer instances of direct patient care, thus reducing direct exposures and SARS-CoV-2 transmission risk.
Use of public transportation, particularly use of the subway or bus, was associated with increased risk of SARS-CoV-2 infection, similar to prior reports of HCWs [10,29].
Our findings suggest that community exposure, defined by area of residence, use of transportation and direct contact with individual with suspected or confirmed COVID-19 outside of work, may contribute to infection risk among NYC-based trainees, albeit less significantly than occupational exposures.
Strengths of our study include collection of robust data directly from physician trainees pertaining to both occupational and community exposures in NYC, an early epicenter of COVID-19 in the U.S. Associations of occupational factors and SARS-CoV-2 infection are strengthened by our ability to verify self-reported test results with laboratory-confirmed data. Additionally, results of a sensitivity analysis indicate similar sociodemographic characteristics among all eligible trainees, the randomly selected subset, and study participants in the analysis sample, suggesting reduced likelihood of selection bias. The limitations include the retrospective design and the potential for measurement error pertaining to trainees who reported having been tested for SARS-CoV-2 without results available for review in the EHS registry, indicating that testing may have been performed outside of the Mount Sinai Health System. The sensitivities and specificities of external tests were unknown and may differ from tests performed at our institution. Finally, underestimation of the cumulative incidence of SARS-CoV-2 infection is possible if participants were infected but asymptomatic and never tested. Finally, results may not be generalizable to trainees outside of NYC, as hospital infection prevention protocols and community transmission vary by geographic location.

Conclusions
Among physician trainees at a large healthcare system situated in an early U.S. epicenter of COVID-19, assignment to medical-surgical units and training in high-risk procedural specialties were most robustly associated with SARS-CoV-2 infection, out of a comprehensive list of occupational, community and sociodemographic factors assessed. Assignment to unfamiliar or non-routine hospital sites or clinical responsibilities was not associated with SARS-CoV-2 infection, and assignment to unfamiliar patient populations was noted to be protective, suggesting that deployment was a safe mitigation strategy during the initial wave of COVID-19. Our findings can inform more tailored infection prevention strategies for trainees during the COVID-19 pandemic.
Supplementary Materials: The following are available online at www.mdpi.com/xxx/s1, Table S1: Survey instrument to assess risk factors hypothesized to be associated with SARS-CoV-2 infection in trainees from February 1 through June 30, 2020; Table 2: Categorization of trainees who responded to the online survey between June 26, 2020 and August 31, 2020; Table 3: Self-reported SARS-CoV-2 test results; Table 4: Comparison of main characteristics between study participants (n=328), the randomly selected sample (n=62) and all initially eligible trainees (n=2543); Table 5