Liu 2009.

Methods	The paper is a re‐analysis and publication in English of Ma 2004, a case‐control study carried out shortly after the SARS outbreak at the Armed Forces Hospital (AFH) in Beijing in which 16 HCW died. The data from Ma 2004 had been published in Chinese only. The paper assesses relationships between protective and risk factors in cases and controls using a 2‐step analysis procedure: univariate analysis and then multivariate analysis for those associations found significant up to the 10% level
Participants	Description of cases ‐ 51 HCW (age mean 29.5 years) who were admitted to AFH during 5 March to 17 May 2003 with clinical features fitting WHO’s SARS criteria. All enrolled analysed cases subsequently proved to be IgG SARS positive (1 case was excluded because he/she was negative). Probable cases of SARS are defined as: documented fever (temperature > 38°C), presence of cough, shortness of breath or breathing difficulty, and a significant history of exposure to a SARS patient not more than 10 days prior to onset of symptoms, plus radiographic evidence of infiltrates consistent with pneumonia or respiratory distress syndrome (RDS) on chest X‐ray (CXR) (World Health Organization criteria, 2003). The text mentions that cases were 76% (51 of the 67) “survived” staff in the AFH Description of controls; 426 HCW (age mean 31.4 years) working in AFH during the same period as cases with self‐reported exposure to SARS but had no symptoms (the text says “uninfected”). All enrolled analysed controls subsequently proved to be IgG SARS negative and their exposure within 1 month of a SARS case was confirmed. These are 90% of AFH employees exposed to SARS.
Interventions	Exposure and risk or protective factors were subsequently elicited by questionnaire and interviews in June to July 2003: gender, age, ethnic group, educational level, co‐morbidity, smoking status, alcohol intake, contact date, occupation, department, contacts with SARS and exposure time. None of these factors proved to be significant in a multivariate analysis. At univariate analysis 17 variables were significantly associated with SARS, 10 of which were protective (i.e. negative association): ‐ wearing a 12‐layer cotton surgical mask ‐ wearing 16‐layer cotton surgical mask (and wearing layers of mask) ‐ wearing glasses ‐ wearing gloves ‐ wearing goggles ‐ wearing multiple layers of protective clothing ‐ taking “prophylactic medicine” (such as “antivirals” and vitamin supplements), performing nose washes after contact and having  training prior to exposure N95 mask use was non‐significant probably because of the rarity of its use At multivariate analysis level, 12 and 16‐layer mask non‐use and not undergoing training, not taking medicine and not wearing multiple layers of masks were found to be associated with SARS onset
Outcomes	Laboratory: all clinically diagnosed hospital‐acquired SARS cases confirmed by + SARS‐CoV IgG ELISA and all controls confirmed by a ‐ SARS‐CoV IgG ELISA Effectiveness: univariate and multivariate analysis among the 28 variables elicited in questionnaires and by interview
Notes	The authors conclude that “this study identified exposure to high‐risk procedures (such as chest compression), and contact with respiratory secretions to be significant risk factors for SARS infection among HCWs in a hospital in Beijing. These results also provide confirmation that personal protective measures against droplet spread, such as wearing multiple layers of mask, are effective against the nosocomial spread of SARS” The main points to bear in mind when interpreting this study are: ‐ the possibility of selection bias in cases (only living cases were recruited whereas we know that 16 HCWs in AFH died) ‐ protective variables are not well‐defined (i.e. the make or type of masks used, whether fitted or not) ‐ information on the 10 protective interventions (variables) was elicited post hoc with a possibility of recall bias (mentioned by the authors in their Discussion) ‐ the lack of reporting of numerator and denominator data for cases and controls ‐ the apparent lack of mention of data assessment and analysis blinded to case or control status ‐ failure to attempt matching between cases and controls and the partly prospective nature of the study design
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	N/A
Allocation concealment (selection bias)	Unclear risk	N/A
Blinding (performance bias and detection bias) All outcomes	Unclear risk	N/A
Incomplete outcome data (attrition bias) All outcomes	Unclear risk	N/A
Selective reporting (reporting bias)	Unclear risk	N/A