We conducted a survey of symptoms and exposures among the cohort of workers at Hospital A, and a nested case-control study to assess risk factors for acquiring SARS-CoV infection after admission of the index case. For the latter study, only individuals who had worked at least one shift during the time the index patient was hospitalized were included (Figure 2).

During the outbreak, in the final two days that the hospital was open (March 17–18, 2003), investigators obtained a comprehensive list of hospital employees and held meetings with individual units (i.e., security, nursing, etc.) to recruit study participants. Upon providing written consent, participants were asked to complete a short, self-administered questionnaire and to provide a serum specimen for determination of antibodies to SARS-CoV. Questionnaires were translated into Vietnamese and responses were back-translated into English. Information was collected regarding the participant's contact with the index patient during his stay in the hospital from February 26^th (evening) to March 5th, as well as their symptoms from the time the index patient was admitted to the hospital until 10 days after he was transferred (~1.5 incubation periods [3-5]). Self-reported symptoms were checked against case investigation forms for 42 staff members. Participants were asked about whether they had ever engaged in a series of activities relating to exposure to the index patient. Questions were simplified in order to minimize translation errors, ensure that the activities would apply for multiple different staff job categories (e.g., nurse and security guard), and to facilitate self-administration.

All exposure variables were comprehensive for the period that the index patient was hospitalized at Hospital A. For example, did you enter the general ward, referred to whether or not the subject entered the general ward at least once at any time during the index patient's hospitalization. Variables were not all exclusive. For example subjects were asked whether they ever came within 1 meter of the index patient, and whether they ever did so without a mask (respirator). Further information pertaining to the number and duration of potential encounters was not requested due to time constraints and concerns about recall bias.

A physician who provided care for SARS patients at Hospital A throughout the outbreak (this physician resided temporarily at the hospital while the outbreak was ongoing) served as a proxy respondent for the staff who had died (n = 6) or were too ill (n = 1) to respond at the time of the survey. This physician had assisted WHO investigators in conducting the initial staff interviews in early March following recognition of respiratory illnesses among staff. Work shift schedules were available for 30.0% of the Hospital staff and were used to verify staff responses. The protocol was approved following expedited review by the hospital institutional review board and the Vietnamese Ministry of Health.

Laboratory confirmation of SARS-CoV infection was performed at the Centers for Disease Control and Prevention (Atlanta, Georgia, USA) and was based either on detection of RNA from SARS-CoV in clinical specimens (via reverse transcriptase polymerase chain reaction assays, RT-PCR), or by serology, as previously described [2]. Case definitions and laboratory assays employed during the outbreak in Vietnam are described in detail elsewhere [2].

To determine which work-related activities may have been associated with contracting SARS-CoV infection after admission of the index case, a case-control study was performed. Inclusion and exclusion criteria for study cases and controls are outlined in Figure 2.

Study cases were defined as persons having: (i) the presence of SARS-CoV antibody in at least one serum specimen collected prior to March 24^th; or (ii) for SARS cases confirmed by means other that serologic testing (i.e., viral culture or RT-PCR), illness onset on or before March 5^th, the last day the index patient received care at Hospital A. SARS Co-V infections resulting from exposure to the index patient were presumed to have resulted in antibody conversion on or before March 24^th (18 days after the index patient's discharge) based on a conservative estimated timeline that includes a 6 day incubation period plus a 12 day interval to seroconversion. The estimated 6 day incubation period is based on the observed median number of days from exposure to illness onset among persons in Vietnam who reported only a single potential exposure to a SARS patient (n = 28). The 12 day estimated interval to seroconversion is based on the observed median number of days from illness onset to SARS-CoV antibody seroconversion among cases in Vietnam for whom the date of seroconversion is known within 48 hours (n = 17). This time frame is expected to exclude individuals who contracted infection from someone other than the index, while capturing the majority of individuals who were infected during index case's hospitalization in Hanoi, including a proportion of those who could have been infected on the last day of exposure to the index patient. The timing of illness onset alone was insufficient as a case selection criterion for the case-control study as this information was unavailable for many study enrollees.

Controls were defined as individuals within the study cohort who (i) were never identified as SARS cases, (ii) worked at least one shift during the time the index patient was hospitalized, and (iii) did not exhibit evidence of SARS-CoV specific antibodies within at least 18 days after last exposure to the index case.

Exposure and demographic variables used for analyses were dichotomous categorical; unknown or missing data was rendered as a negative response. Non-parametric tests (Fisher's exact, Cochran's Chi-square) or odds ratios (OR) with 95% confidence intervals (CI) were used to assess differences between groups using negative responses as the referent. The threshold for statistical significance was established at a p-value < 0.05.

Advanced age (> 50 years) and underlying medical conditions were not significant correlates of case status in this outbreak [Uyeki, personal communication] and were therefore not included in the analysis. All statistical analyses were performed using SPSS version 11.0.1 (SPSS Inc., Chicago, IL).

A staff census prepared in February 2003, indicated that 193 individuals were employed at Hospital A at the time of the SARS outbreak. The distribution of personnel between clinical and non-clinical roles is shown in Table 1. Twenty-nine of the 36 SARS cases (81%) at Hospital A occurred among clinical personnel with direct patient care or ancillary clinical roles. Direct patient care activities included primary medical functions such as patient examination and diagnostic evaluation, performance of procedures, and conduct of ongoing care and monitoring. In addition, non-clinical personnel were affected as well, with 19% of cases (n = 7) occurring among housekeepers and other cleaning staff (n = 5), kitchen staff (n = 1), and receptionists (n = 1). Several of the affected housekeepers entered the index patient's room on the night he was admitted to clean vomitus and respiratory secretions from the floor and walls, and various kitchen staff entered his room to deliver meals or collect service trays. Figure 1 depicts the epidemiologic curve of SARS illness among Hospital A personnel.

Two peaks of illness onset among staff were evident, on March 4^th, and 9 days later on the 13^th (Figure 1). All but one of the non-clinical staff cases became ill during the first apparent wave of the outbreak, which encompassed the period the index patient was an inpatient at Hospital A. Housekeepers and certain members of the kitchen staff had access to the index patient's room(s) during his stay, for purposes of cleaning and meal delivery. Overall, observed attack rates were highest among personnel with direct patient care responsibilities. Notably, however, nursing staff at Hospital A who were assigned to the operating Room (Op.Rm.)/intensive care unit (ICU) had significantly lower attack rates (7.1, p < 0.006) than did nurses in other staffing categories. Though mid-wives might not be anticipated under normal circumstances to have had opportunities for direct contact with the index patient, several women who had had complicated deliveries were hospitalized in the inpatient ward or ICU during the index patient's stay; two briefly shared a room with him.

Overall, 79% of Hospital A staff completed the exposure and symptom questionnaire, and 64% contributed at least one serum specimen. The lowest rates of participation in the exposure and symptom survey occurred among physicians and ICU nurses staffing categories (58.6%, 57.1%, respectively), but in both instances participation increased to over 64% for the serosurvey (Table 1). Staff work schedules were used, where possible, to independently verify a worker's presence at Hospital A during the index patient's hospitalization. A comparison between staff work schedules and worker responses is shown in Table 2. In general, there were fewer non-clinical staff than clinical staff participants in the serosurvey, but overall, serosurvey participation met or exceeded 50% in all but five staffing categories, these being, 'kitchen', 'dental', 'laboratory', 'administration', and 'reception'.

The questionnaire was completed by proxy for seven individuals who had died or were too ill to complete the questions at the time of survey administration. Removal of these respondents from the study pool did not substantively affect study findings (shown below), therefore these responses were included analyses as appropriate.

Results of the survey regarding symptoms of illness experienced by study participants are summarized in Figure 3.

Consistent with findings from a clinical study of SARS patients conducted in Vietnam [6], fever, fatigue, myalgia, chills, anorexia, headache, and cough were the most frequently reported symptoms experienced by SARS infected staff at Hospital A. These symptoms occurred in over half of SARS cases, and all occurred at levels significantly above those reported by employees without SARS-CoV infection. Dizziness, shortness of breath and vomiting were also reported at significantly higher frequencies among SARS cases than uninfected staff, but nevertheless were reported by = 40% of SARS cases. Only diarrhea, sore throat, and rash were not reported at higher frequencies among SARS cases than uninfected staff. The background occurrence of rash was low among all participants, but both sore throat and diarrhea occurred at appreciable background rates (19% and 20%, respectively) among staff not infected with SARS-CoV.

Of the 124 Hospital A staff who participated in the serosurvey, 36 (29%) had at least one serum specimen that tested positive for the presence of antibody to SARS-CoV antigen (Table 1). Four Hospital A staff members with confirmed SARS-CoV infection did not complete the questionnaire or participate in the serosurvey. All 32 confirmed SARS cases that did participate in the survey tested positive for the presence of antibody to SARS-CoV antigen. Four additional, previously unrecognized, seropositive individuals were identified as a consequence of the serosurvey. Thus, 4 of the 36 initially identified SARS cases during the outbreak did not have serological confirmation, and they were replaced by the 4 additional, previously unrecognized, seropositive individuals identified as a consequence of the serosurvey (n = 36).

These 4 newly identified seropositive individuals (2 general ward nurses, 1 laundry worker, and 1 receptionist) had mild illness and were not identified during the outbreak despite active surveillance conducted among staff at Hospital A. Among this group, the 2 General Ward nurses reported having had at least three symptoms associated with SARS illness in this outbreak (the first reported fatigue, headache, and shortness of breath; the second also reported headache along with myalgia, chills, dizziness, anorexia, and vomiting), but neither reported having experienced fever or cough, which were inclusion criteria in the WHO SARS case definition used at the time. Both nurses also reported experiencing diarrhea during the study period. However, diarrhea was common among all Hospital A workers (23% of all reporting workers) regardless of SARS case status. Neither the laundry worker nor the receptionist reported having experienced symptoms significantly associated with SARS illness in Vietnam, with the exception that the latter reported vomiting.

Twenty-two study cases and 45 controls were identified (Figure 2). (Confirmed SARS cases not included in the study were excluded on the basis of illness onset date, date of seroconversion, or insufficient information to classify.)

Nearly all activities associated with physical proximity to the index patient or to his hospital rooms were significantly associated with SARS-CoV infection by univariate analysis (Table 3). However, touching the index patient (with or without personal protective equipment in the form of gloves, gown, face mask) or speaking to him in his room, were not significantly associated with SARS-CoV infection. Having had a job that involved direct patient care or sanitation/kitchen duties was also not associated with SARS-CoV infection. Having a non-clinical staff position was nominally protective (O.R. = 0.2, p = 0.011) against SARS-CoV infection, as was having had an upper-respiratory infection ('head cold', O.R. = 0.2, p = 0.039) within the prior 6 months.

Multivariate analyses were not performed due to the limited size of the study population.