2.1 Study population
Data of this study were obtained from a time-lagged panel survey with three waves of measurement from February 2020 to May 2020. Informed consent was obtained from all the participants. The survey was conducted at an infectious diseases hospital of Shenzhen (Hospital T). This hospital accommodated all the 471 confirmed COVID-19 cases of Shenzhen during the outbreak and epidemic of COVID-19. The study sample was recruited through a cluster sampling strategy, targeting healthcare professionals who were directly involved in providing diagnosis, treatment and nursing care for COVID-19 patients in this hospital. All frontline healthcare professionals working with COVID-19 were asked to participate in this survey. The Wave 1 data was collected in February 2020, with a total of 492 healthcare professionals completing the survey. Among these, 241 participants were excluded from the data given high proportion of missing values, leaving 251 eligible participants for wave 2 and wave 3 survey. There were 241 and 117 participants excluded for nonresponse at wave 2 and wave 3, respectively. Therefore, in the final study sample, 134 respondents were followed for all three waves of surveys, with a 27.24% follow-up response rate.
2.2 Measurement
2.2.1 Depressive symptoms
Depressive symptoms were assessed at the 3rd wave survey, using the 12-items General Health Questionnaire (GHQ-12). Participating healthcare professionals were asked to rate on a 5-point Likert scale ranging from 1 (never) to 5 (always) on items like how often she/he felt “constantly under strain”, “unhappy and depressed”, etc. GHQ-12 has been reported as a valid tool for detecting depression in both general population and hospital staff. In the study, items 2, 5, 6, 9, 10 and 11 were scored directly, while items 1, 3, 4, 7, 8 and 12 were reverse coded to ensure the direction of each item score was consistent. The average value of all the items was calculated to obtain an overall scale score. The Cronbach’s α of the scale in this study was 0.88, 0.85, and 0.88.
2.2.2 Trauma exposure
At the first wave of survey, we investigated the trauma exposure using a six-item questionnaire, the Explosion Exposure Questionnaire [44], which represented two dimensions of trauma exposure: direct exposure and damage. The direct exposure dimension consisted of three items: a) is there any relative confirmed to be a case or died during the COVID-19 pandemic? (0, no; 1, infected; 2, dead); b) is there any acquaintance confirmed to be a case or died during the COVID-19 pandemic? (0, no; 1, infected; 2, dead); c) Whether you have witnessed someone infected or dead during the COVID-19 pandemic? (0, no; 1, yes).The damage dimension was measured by the following three items: d) What was the impact of COVID-19 pandemic on your health status? (0, none; 1, mild; 2, medium; 3, severe); e) What was the impact of COVID-19 pandemic on your property? (0, none; 1, mild; 2, medium; 3, severe); f) What was the impact of COVID-19 pandemic on yourself? (0, none; 1, mild; 2, medium; 3, severe). The average score of the six items was calculated to form the overall scale score. The Cronbach’s α of the Explosion Exposure Questionnaire in this study was 0.60.
2.2.3 Intrusive rumination
At the second wave of survey, we assessed the intrusive rumination and organizational silence. Intrusive rumination was measured by three items from the simplified Chinese Version of Event Related Rumination Inventory (C-ERRI) [45]. C-ERRI was translated into Chinese by Dong and Liu from its English version developed by Cann [28]. As Dong and Cann reported, C-ERRI had high reliability and validity for measuring intrusive rumination [28, 45]. The original inventory used ten items to assess intrusive rumination. In this study, to save responding time of frontline hospital staff, three key items out of the ten were selected from C-ERRI to form a short version: a) thoughts about the event distracted me or kept me from being able to concentrate; b) other things kept leading me to think about my experience; c) I tried not to think about the event, but could not keep the thoughts from my mind. Respondents were asked to rate on a 5-point Likert scale ranging from 1 (Strongly disagree) to 5 (Strongly agree). Mean value of the three items was considered as score of intrusive rumination. The Cronbach’s α of the short version scale in the study was 0.89.
2.2.4 Organizational silence
Organizational silence was assessed by the 6-item Employee Silence Scale (ESS) [46]. It consists of three construct dimensions: acquiescent silence, defensive silence and disregardful silence. Acquiescent silence means that employees keep silent and passively obey orders when they perceive they are not able to change the current situation. Two items measured acquiescent silence: a) The leaders have made decisions, and my opinion would not take any effect; b) The leaders would not change decisions, and there is no significance even I express my opinion. Defensive silence refers to employees not expressing their opinions in order to avoid conflicts and estrangement. It was measured by two items: c) Remain silence and restraint so as not to be the targets; d) It is not necessary to offend the leaders and colleagues to express my opinions. Disregardful silence refers to employees not giving any views or opinions for they have low organizational commitment in the present job or institution. Disregardful silence was assessed by two items: e) I am not concerned of the hospital affairs; f) I choose the “middle way”, and will not say anything and bear any more responsibility. Participants were asked to rate six items on a 5-point Likert scale ranging from 1 (never) to 5 (always). Mean value of the six items was used as score of organizational silence. The Cronbach’s α of this scale was 0.90.
2.2.5 Control variables
Two factors were included as control variables in data analysis given their potential effects on depressive symptoms. First, gender was controlled as a dichotomous variable (0, male; 1, female). As reported, there was gender differences in the prevalence of depression among frontline healthcare professionals [5]. Second, it was also documented that years of working was associated with depressive symptoms and organizational silence among doctors and nurses [47, 48]. Therefore, we included working years in the analysis model, which was categorized into four groups: <=5 years, 6—15 years, 16—25 years, and 26 years +.
2.3 Procedure
The survey was conducted online during the pandemic of COVID-19. The ethical approval of the survey was obtained from the Institutional Review Board of the Department of Psychology, Tsinghua University. Healthcare Professionals All healthcare professionals from that department were invited to complete the online survey. Once they agreed to participate, a WeChat link of Questionnaire Star was sent to them, with informed consents being attached to the survey Upon receipt of the completed questionnaires,, data could be obtained and downloaded from the sever of Questionnaire Star. The first wave of survey was conducted in February 2020, followed by the second and third wave of survey in March 2020 and May 2020, respectively.
2.4 Data analysis
The hypothesized second stage moderation model was tested using PROCESS v2.15 for SPSS provided by Hayes [49]. All continuous variables were mean centered prior to analysis, whereas gender was binary variable (0, male; 1, female) and working years was categorized into 4 groups (1, <=5 years; 2, 6–15 years; 3, 16–25 years and 4, 26 years +). Two steps were taken to test the hypothesized model: First, an ordinary least squares regression was performed to test the effects of trauma exposure on depressive symptoms; second, a moderated mediation model was tested with intrusive rumination as the mediator and organizational silence as the moderator. To test the significance of direct and indirect effect, 5000 bootstrap samples were used to calculate the bounds of 95% confidence intervals. Last, we plot the interactions at three values of organizational silence (mean and ±1 SD) to visualize the conditional effects of the moderator.