This study used data collected from the Employees Working from Home (EWFH) study conducted in Australia during the COVID-19 pandemic from October 2020 to November 2021. Sampling and recruitment and a full description of the study profile for the EWFH study have been described elsewhere (Oakman et al. 2022). Briefly, convenience sampling was used to recruit a sample of Australian adults aged 18 or more years who WFH 2 or more days per week during the COVID-19 pandemic. Recruitment occurred via Facebook’s paid service, professional and personal networks, the La Trobe University Facebook page, and LinkedIn.
Respondents were offered the opportunity to go into a prize draw to win a gift voucher, if they completed the questionnaire and provided contact details. In general, the surveys were open for approximately 4 weeks from the time of opening. Response numbers to the survey at baseline and for the subsequent two time points are outlined in Fig. 1.
Procedure
Data were collected by questionnaire at three time points via Qualtrics XM software (Qualtrics, Provo, UT). All respondents who consented to be recontacted after the first survey were invited to participate in the second and third surveys. Following the first survey, non-responders were provided with three reminders via email. Responses rates at Survey 2 and 3 were 67% and 53%, respectively. The study flow is outlined in Fig. 1.
Ethics approval was obtained through La Trobe University Human Ethics Research Committee, approval number HEC20388. All study participants were provided with written information about the study. All participants provided informed consent prior to participation.
Measures
Musculoskeletal pain
Musculoskeletal discomfort was recorded separately for five body regions (neck/shoulders, hands/fingers, arms, middle to lower back, and hips/bottom/legs and feet) using a measure with evidence of validity in a number of different industry sectors (Oakman et al. 2014). Question was, “in the past six months have you ever experienced discomfort or pain in part of your body, especially towards the end of your working day or night”. Response options for pain/discomfort frequency ranged from never (1) to almost always (5). For each body region, the score was dichotomised as no pain (0) or having pain (1). The sum score was then ranged from 0 (no pain sites) to 5 (pain in all 5 body regions).
Other variables
Questions on psychosocial factors were selected from the Copenhagen Psychosocial Questionnaire III (COPSOQ) (Burr et al. 2019). For the current study, constructs (number of items) included: Quantitative Demands (2), Quality of Leadership (2), Vertical Trust (3), Role Clarity (3), and Influence at Work (3). A sample item for quantitative demands was ‘I get behind in my work’. Each item was measured on a 5-point Likert-scale from 1 (Never/hardly ever) to 5 (Always) or 1 (To a very small extent) to 5 (To a very large extent), depending on the item. Mean rating scales for each construct were summed and divided by the number of items. Dimensions were treated as continuous variables in the current analyses, ranging from 1 to 5.
Work family conflict (WFC) included five questions from previously validated items (Netemeyer et al. 1996) with a seven-point scale from strongly disagree (1) to strongly agree (7). Average scores across the items were used to construct the final measure as a continuous variable ranging from 1 to 7.
Job satisfaction was measured from the item “How pleased are you with your job overall, everything taken into consideration?” with respondents selecting an option from 1 (very unsatisfied) to 5 (very satisfied) (Oakman et al. 2014).
Demographics Age was based on the question “What is your age group?” 18–25 years; 26–35 years; 36–45 years; 46–55 years; 56 years and over. The categories were then collapsed to 18–35 years; 36–55 years; 56 years and over. Gender was based on the question “Are you: Male, Female, Other”. Work hours were classified from the following question, “Currently what are your usual working hours (average per week)?”—with those answering ≥ 35 h per week classed as ‘full time’ and others as ‘part-time’.
Workstation location Based on the question, developed for this study, “When you are working at home, where do you usually work?”. Three response options were offered: Wherever—“I just find a place somewhere that’s free, such as on the kitchen table or other place”; Separate—“I have my own place in a separate room by myself”; and Interruptions—“I have my own place but in a room that can be busy with other people” (Oakman et al. 2022).
Workstation comfort Based on the question “How comfortable is your home workstation in comparison to your usual workstation?”, with four response options, very uncomfortable to very comfortable.
Domestic arrangements Questions included “Which of the following best describes your usual living arrangements?”, “Do you have caring responsibilities other than children”, and “When you are working at home are children usually at home with you?” A three-level classification was created: Single person household, Adults only, or Dependents.
Work sector Based on a question about the sector of employment at the time of the questionnaire.
Statistical analysis
To describe the course of MSP over the study period, Growth Mixture Modelling (GMM) analyses were used to identify latent classes with different growth trajectories of number of reported pain sites over the three time points. These models are less restrictive than a latent class analysis, as the GMM accounts for between-subject heterogeneity within the latent classes by including random effects. Respondents were required to have at least two survey responses to be included in the trajectory modelling. GMM models with one to five classes were examined, with each model being run 50 times with different starting values to ensure the optimal solution was found instead of local maxima. The optimal solutions for each class number were compared and the Bayesian information criterion (BIC) was used to select the best fit model (see Fig. 1; Table S1). Trajectory analyses were run with the ‘hlme’ function from the R package ‘lcmm’ (Proust-Lima et al. 2016).
Individuals were matched to a latent class using posterior probabilities, with each individual allocated to the group for which the probability was the highest (Berlin et al. 2014). Demographic differences between participants in each group were calculated using the chi-squared test of independence. Due to small numbers, the n = 3 respondents who identified their gender as ‘Other’ were excluded from further exploratory analysis. A multinomial logistic regression model was used to determine the associations between predictors at baseline and group membership based on the GMM. Multinomial regression analysis was used, because the response variable has several unordered categories. Odds ratios (OR) with 95% confidence intervals (CI) were determined, comparing membership in each group to the chosen reference category which was low stable.
All statistical analysis was performed in R version 4.1.1 “Kick Things” (R Core Team 2021). All tests of statistical significance were two-tailed, and p < 0.05 was considered significant.