The impact of COVID-19 lockdown on physical activity and sedentary behaviour in secondary school teachers: a … – BMC Public Health
Participants
A non-probability cluster sampling strategy was used to recruit Flemish secondary school teachers. In August and September 2019, all secondary schools in Flanders (Belgium) were contacted through e-mail and telephone. To increase the response and participation rate, the Flemish Department of Education (Vlaams Departement Onderwijs) as well as all education networks (i.e., Flemish community schools, subsidised public schools, subsidised free schools) were involved in the recruitment and were asked to promote the study among all school principals. To stimulate school involvement, a convenient selection of schools in Flanders were visited to promote our study face-to-face. Schools that were willing to participate in the study were asked to send an e-mail with a link to an online questionnaire to their entire teaching staff. Furthermore, the same link was spread through social media (e.g., Facebook, Twitter) and by posting advertisements via the Flemish Department of Education. Teachers being in sick leave or having a distorted physical activity/dietary pattern (by e.g., injuries, diseases, following a diet) were excluded from the final sample. As this study was part of a larger longitudinal follow-up study, in which we questioned distorted physical activity/dietary patterns by one question, we could not differentiate between the two. As a result, participants who reported being distorted in either category were excluded.
This prospective cohort study is part of a larger longitudinal study, including six measurements throughout the 20192020 school year, i.e., Sep/Oct, Nov/Dec, Jan/Feb, Mar/Apr, May/Jun, and Jul/Aug. For the purpose of the present study (i.e., measuring the impact of COVID-19 lockdown on secondary school teachers PA and SB), the Jan/Feb measurement (Jan 27 Feb 11, 2020) will serve as baseline (T0). The measurement performed in Mar/Apr (Mar 23 Apr 7, 2020), which is five days after the installation of the lockdown measures, will serve as measurement under lockdown-exposure (i.e., primary endpoint (T1)). The measurements prior to T0 (i.e., Sep/Oct (T-2) and Nov/Dec (T-1)) will serve as pre control measurements, whereas the measurement after T1 (i.e., May/Jun (T2)) will serve as post control measurement. The Jul/Aug measurement was omitted due to anticipated summer holiday bias. The timeline of the measurements is displayed in Fig.1.
Timing of the measurements
At each time point participants were asked to complete an online questionnaire, including sample characteristics and primary outcome measures. Sample characteristics include socio-demographics, work-related information, and other health-related variables. Primary outcomes in the present study are PA and SB. During each measurement period of two weeks, three reminders were sent to the non-responders, each on the fourth, eighth and eleventh day after activation of the online questionnaire.
Socio-demographics include sex, age, highest diploma (i.e., secondary school degree, post-secondary school degree or certificate, Bachelors degree, Masters degree, PhD degree), having an extra job (yes/no), marital status (i.e., single, married, unmarried, living together with partner, divorced, widowed), having children (yes/no) and ethnicity (i.e., White European, White other, North-African, Afro-American, Indian, Middle-Eastern, South-Asian, Southeast-Asian, other). Work-related factors include education network (i.e., Flemish community schools, subsidised free schools, subsidised public schools) and total working hours per week. Health-related variables include self-reported height and weight (from which body mass index (BMI; kg/m) was calculated) and smoking status (yes/no).
The validated International Physical Activity Questionnaire (IPAQ Dutch long version) was used to estimate PA domains and intensities during the last seven days [23]. This self-report questionnaire includes 31 items and assesses four contextual PA domains: [1] work-related [2], transport-related [3], domestic and garden, and [4] leisure-time PA. The participants were asked to fill in the number of days and the amount of time (hours and minutes) spent in three different PA intensity levels within each domain, namely [1] walking [2], moderate-intensity PA, such as carrying light loads, washing windows, cycling or swimming at a regular pace, and [3] vigorous-intensity PA, such as heavy lifting, aerobics, running and fast cycling or fast swimming (as specified by the IPAQ). The outcome measures are domain- and intensity-specific PA as well as total PA expressed in min/week. Multiple criteria from the IPAQ scoring protocol were applied [24]: [1] only values of ten or more minutes of activity were retained; [2] non-relevant observations were excluded (e.g., answering in step counts instead of minutes); [3] PA levels higher than 960min/day (i.e., 16h/day) were excluded, as this would be unrealistic. Total scores per domain were calculated by multiplying the frequency of each PA per week by its duration expressed in minutes. Next, the domains were combined into total walking, moderate-intensity PA, and vigorous-intensity PA. Lastly, total PA was calculated by summing all items. It should be mentioned that total light-intensity PA, in which walking is just one component, is not questioned in the IPAQ. Therefore, total PA in this study only represents walking and moderate-to-vigorous-intensity PA. Note that the IPAQ scoring protocol includes a section Truncation of Data Rules, which is not applied in the current study. The protocol states that this rule attempts to normalize the distribution of levels of activity which are usually skewed in national or large population data sets [24]. Instead of truncating and forcing data into a normal distribution, we opted to tailor the statistical analyses to the non-normal data distributions (see Statistical analysis section). The IPAQ has fair to good psychometric properties (reliability: =0.80 and validity: r=0.30) [25].
SB was assessed by using the Dutch version of the validated context-specific sedentary behaviour questionnaire for adults developed by Busschaert and colleagues [26]. This self-report questionnaire assesses SB in three domains: [1] work-related [2], transport-related, and [3] leisure-time SB. Participants were asked to specify how much time they spent sitting/lying down during the last seven days (weekdays and weekend days separately) within each domain. The outcome measures are domain-specific SB as well as total SB expressed in min/week (i.e., sum score of minutes during the week and weekend). Participants were asked to fill in the number of days and the amount of time spent sitting/lying for several items/activities (e.g., TV watching, computer use, reading) within each of the three domains. For each item, a specific time interval could be chosen; e.g., 1 to 15min, 15 to 30min, 30 to 60min, 1 to 2h, etc. Midpoint values (e.g., 7.5min, 22.5min, 45min, 90min, etc.) of each test item interval were calculated. As it was not mentioned in the protocol how the upper limit time intervals more than seven hours a day and more than eight hours a day had to be interpreted, it was decided to consider these time intervals as 450min and 510min, respectively. Total sedentary time for an average day was estimated by summing all midpoint values of the specific SB contexts (weekdays and weekend days separately) and was estimated as follows: ((total sedentary time on a weekday * 5) + (total sedentary time on a weekend day * 2))/7. Although not explicitly mentioned in the paper of Busschaert and colleagues [26], but consistent with the IPAQ protocol, we decided to exclude participants with SB levels higher than 960min/day (i.e., 16h/day) from the analysis.
Patient and public involvement was not appropriate for this study.
Secondary school teachers from multiple geographical regions, urban and rural communities and different education networks were recruited for this study. Participants could report their sex, diploma and ethnicity. The author team included early, middle and late career researchers with balance from people who identify as male and female.
All data were analysed using R (R core Team, 2019; R Studio version 3.6.2) and SPSS (version 27). P-values<0.05 were considered statistically significant, whereas p-values between 0.05 and 0.10 were considered marginally significant. Representativeness of the sample at baseline (T0) was assessed by conducting two proportions z-tests. Drop-out analyses between baseline (T0) and the primary endpoint (T1) were conducted to assess possible selection bias of the retention group. In the first analysis, participants of whom we had data at T0 and T1 (i.e., retention group) were compared to participants of whom we only had data at T0 (i.e., drop-out group). As the generalized mixed models that we used typically include all available observation points, we decided to perform a second analysis in which we compared participants of whom we had data at T0 and T1 to participants of whom we only had data at T1. Independent samples t-tests, Mann-Whitney U tests and chi tests were conducted to detect possible differences between the drop-out group and retention group regarding total PA, total SB, sex, age, ethnicity, marital status, having children, smoking status, diploma, having an extra job, education network and BMI.
Multilevel models were used for data analysis. Preliminary analyses checked if a three level model was advised (repeated measures clustered within participants, participants clustered within schools) using graphical representations and by inspecting the amount of variance explained by each cluster. If necessary, one (or both) levels were dropped. Possible confounders, such as age and sex, were checked, but seemed to have no significant effects, and therefore no adjustments were made in the statistical models. The PA scale scores were non-normally distributed with continuous, positively skewed non-negative values. The SB scales also contained non-negative continuous values, but with less severe skewness. For both outcome variables, Gamma and Gaussian generalized linear mixed models were constructed using the R package lme4 [27]. To decide upon the model (i.e., Gamma or Gaussian) and link functions (i.e., log, inverse or identity), Bayesian Information Criterion (BIC) values were compared and a likelihood ratio test was performed (lrtest() function of the R package lmtest [28]). The model selection procedure of each outcome is explained in Additional file 1. For both PA and SB outcomes, the Gamma model with the log link function was selected. In total, five separate models (i.e., total PA, PA intensities, PA domains, total SB, SB domains) were analysed. In order to assess the effect of the lockdown on total PA and SB, a model with total PA and one with SB as outcome variable and time as predictor variable was fitted. To inspect the lockdown effect in the different domains (PA and SB) or intensities (PA), the same model was fitted but with the domains or intensities as a categorical predictor variable together with an interaction term between time and domains or intensities. Significance of main and interaction effects of the categorical variables consisting of more than two categories were checked using Wald Chi tests (Anova function from the R package car [29]). Contrasts were constructed (test Interactions function from the R package phia [30]) to inspect statistical differences between T0 and T-2, T-1, T1, T2 of each domain and intensity, respectively. Data visualisation was performed using the R packages ggplot2 [31] and sjPlot [32], based on the predicted values of the response variable. More detailed information on the statistical analysis procedure can be found in Additional file 2.
Continue reading here: