Data collection, setting and participants
We report findings from a longitudinal cohort study of people that were employed in a paid job or self-employed prior to the pandemic, residing in Australia and aged at least 18 years.
A total of 2603 participants enrolled in the study and completed a 20-minute baseline survey (either online or via a telephone survey) between 27 March and 12 June 2020, of which 2151 participants also consented to future follow-up surveys at 1, 3 and 6 months after baseline. Baseline survey measures for the cohort have been previously described [2], and only four respondents reported a positive COVID-19 test result during a 3-month follow-up survey [12]. The cohort includes a group of people experiencing work loss early in the pandemic and a control group of people who did not lose working hours.
Health outcomes
Three health outcomes were assessed at each of the four-survey time-points: (1) psychological distress, (2) mental health, and (3) physical health. Psychological distress was assessed using the total scores from the 6-item Kessler Psychological Distress scale [13] ranging from 6 to 30. Mental and physical health was assessed using the mental health component summary score and the physical health component summary score from the 12-item Short Form Health Survey [14], ranging from 0 to 100.
Work
Exposure to work was dichotomised into two groups at each survey timepoint, where individuals were described as either Working (W) or Not working (N). The state of Working describes people that worked more than zero hours during the prior week. The state of Not working describes people that had either lost their job, or that had been temporarily stood down from work.
Employment
An employment variable was defined to describe whether people were employed or not at each survey time-point.
Analytical approach
A total of 6859 observations were available for statistical analysis from 2603 participants across four surveys.
Firstly, summary statistics describe subgroups of people that were either working or not working at baseline. Groups were summarised according to demographics, pre-existing health conditions, residential location, and survey mode.
Secondly, linear mixed models were used to account for repeated measures. Four models labelled 1–4 were designed to evaluate health outcomes with different exposures describing work loss, and were designed to answer the four research questions. Model 1 focuses on baseline work status groups and their changes in health over time. Models 2–4 describe health outcomes and work status at any survey time-point. Four models were estimated for each of the three health outcomes.
Model 1: Does being out of work early in the pandemic affect health six months later?
Model 1 describes health outcomes across each of the four survey time-points, comparing individuals that were either working or not working at baseline. The exposure group for model 1 was working status at baseline (i.e. time-invariant work status) with an interaction term describing each survey time-point allowing us to estimate the differences in health outcomes for both groups at each survey time-point regardless of their future working status.
Model 2: Do health impacts differ for people not working if they are employed?
Model 2 comprises of two sub-models. Model 2.1 describes the health outcomes for individuals either working, or not working at each of the four respective survey-time points. The exposure for Model 2.1 was working status at the same survey time-point as the corresponding health score. Model 2.2 is the same as Model 2.1 but with those who were not currently working separated into those who were still employed and those who were unemployed.
Model 3: What are the health impacts of changes in exposure to work?
The exposure for Model 3 was defined as an interaction between the work status at any given survey time-point with the work status of the prior survey. Four outcomes are possible: (1) working at both time-points; (2) not working at both time points; (3) transition from working to not working; (4) transition from not working to working. Participants were assigned as working before baseline.
Model 4: How does the longitudinal context of changes in work effect health?
The exposure in Model 4 is a three-way interaction term for the work status of a particular survey with the work statuses for the prior two surveys. The reference group for Model 4 describes people that were currently working and that also were working at the two prior survey time-points. Participants were assumed to be working for two pseudo-survey time-points prior to baseline, supported by study participant eligibility criteria for being involved in paid work prior to the pandemic (and specifically during September to December 2019).
Fixed effects and random effects
Variables describing gender, age group, survey time-point, residential location, and pre-existing health conditions prior to baseline were included in models as fixed effects. Regression models for mental health and psychological distress included fixed effects for pre-existing anxiety and depression. Regression models for physical health included a variable describing the number of pre-existing medical condition categories as none, one, or two or more. An interaction term was included between survey time-point and whether participants resided in Victoria or the Rest of Australia, in addition to fixed effect for survey time-point itself due to known influences of an extended lockdown on health [15]. Previous analyses describe response categories [2] and also identified differences in health outcomes by survey mode, thus survey mode was also included in all models as a fixed effect. Reference groups for fixed effects were male, ages 35–44 years, no pre-existing medical conditions, residing outside of Victoria, and online survey mode. Intercept estimates correspond to these reference groups. Individuals whose state of residence or gender was unknown were excluded from regression analyses.
Across all linear mixed models, repeated measurements were incorporated by including random effects for survey time-point with a unique identification number for each participant.