Study design and population
We conducted a cross-sectional survey from April 20–May 22, 2020, in Central (Mulago National Referral Hospital, Entebbe Regional Referral Hospital), Eastern (Jinja Regional Referral Hospital), Western (Kabale Regional Referral Hospital), and Northern (Arua Regional Referral Hospital) regions of Uganda. At the time of this study, these hospitals were the only hospitals managing active COVID-19 case-patients. By the time the study began, the hospitals had managed 212 cases [18].
We designed a self-administered, structured questionnaire based on previous studies in outbreaks of respiratory infectious diseases, including COVID-19 in China [1, 25,26,27]. We chose a convenience sample of HCWs (doctors, clinical officers, nurses, midwives, radiographers, cleaners, drivers, administrators, laboratory personnel, and support staff) present on day-shift duties who consented to participate in the survey. The number of questionnaires distributed was based on the number of HCWs on duty in the respective hospitals (total of 335). Recruitment took one day in each referral hospital.
The principal investigator explained the study purpose and procedures to the HCWs in the respective departments and obtained written informed consent from all the participants prior to participation in the study. The participants indicated their consent by checking an appropriate box for consent before filling the questionnaire. The number of questionnaires distributed was based on the number of HCWs on duty, as determined by the respective heads of departments, in the respective hospitals. The questionnaires were returned by the heads of departments after 24 h. HCWs were categorized as ‘direct contact group’ if their jobs involved direct contact with patients and ‘indirect contact group’ if they were in contact with patient-related items (e.g., biological specimens, equipment), as defined previously [27].
Study variables and data collection instruments
We captured data on HCWs’ socio-demographic and occupational characteristics, concerns and attitudes regarding COVID-19, and their immediate psychological status. Data collected included age, sex, professional cadre, level of education, years of professional experience, number of hours worked per week, number of children, persons with whom the HCW resided, and whether the HCW had ever provided care to a suspected or confirmed COVID-19 patient.
We assessed risk perception towards COVID-19 using 27 concern statements related to fear of contracting COVID-19, fear of spreading COVID-19, workplace-related conditions, and stigma. Each concern statement had four response options: ‘strongly agree,’ ‘agree,’ ‘disagree,’ or ‘strongly disagree’. We applied a scoring system using a four-point Likert scale from zero points (‘strongly disagree’) to three points (‘strongly agree’). Concern statements were negatively-worded (e.g., “there is no adequate personal protective equipment (PPE) at my workplace”), so that a higher score signified a higher degree of risk perception.
We used the 12-item General Health Questionnaire (GHQ-12) developed by Goldberg to assess the psychological state of HCWs [28]. The tool is multi-dimensional and has questions that assess social dysfunction, anxiety, and depression. The GHQ-12 has been widely used in assessing psychological state in outbreaks of infectious respiratory diseases (e.g., SARS and COVID-19) and found to have high reliability and validity [1, 5, 29,30,31]. The instrument includes 12 items (six negatively-worded and six positively-worded). The scoring method (from 0 to 36) is described elsewhere [32]. Briefly, we adopted the four-point Likert scale, with each item score ranging from ‘0’ to ‘3’. For negatively-worded items, ‘0’ indicated ‘Not at all’, ‘1’ indicated ‘No more than usual’, ‘2’ indicated ‘Rather more than usual’ and ‘3’ indicated ‘Much more than usual’. Positively-worded items were scored as follows: ‘0’ indicated ‘More so than usual’, ‘1’ indicated ‘Same as usual’, ‘2’ indicated ‘Less so than usual’, and ‘3’ indicated ‘Much less than usual’. All items were added to obtain the total score, ranging from 0 to 36 (with a higher score signifying worse mental health status). We classified respondents with scores greater than the cut-off point of 12 as having psychological distress, as previously described [33].
Data management and statistical analysis
We entered data into EpiData 3.1 (EpiData, Odense, Denmark) and exported it to STATA version 13 (Statacorp, College Station, Texas) for analysis. Categorical data were summarized by frequencies and percentages; continuous normally-distributed data (risk perception score and GHQ-12 score) were presented as means with standard deviations (SDs), and continuous non-normally-distributed data (hours worked, number of children) as medians with interquartile ranges.
We dichotomized responses to concern statements into non-concern (strongly disagree and disagree) and concern (strongly agree and agree). Respondents were categorized into three groups: low risk perception (at or below the first quartile of concern scores); moderate risk perception (in the second quartile); and high-risk perception (third and fourth quartiles), as used previously [27]. The prevalence of psychological distress was determined as the percentage of respondents with GHQ-12 score greater than 12.
Finally, we performed univariable and multivariable analyses with psychological distress as a binary outcome to identify factors associated with psychological distress among HCWs. We considered risk perception among HCWs as our main exposure variable of interest and adjusted for other variables, including duration of professional experience, contact with confirmed COVID-19 case, and sex as potential confounders.
We reported prevalence ratios (PRs) with corresponding 95% confidence intervals (CIs) as measures of association between psychological distress and associated factors. We obtained PRs via modified Poisson regression, using a generalized linear model with Poisson as family and a log link without an offset but including robust standard errors. We did not use odds ratios as the measures of association because they could potentially overestimate the effect given the high prevalence of our outcome variable.