Posted inExercise / Home

Correlation analysis between physical activity and tendency to depression among occupational groups: an isotemporal substitution approach BMC Public Health


The Asia Best Workplace Mainland China (ABWMC) program was a cross-sectoral study to support companies in building healthy workplaces through policy, infrastructure, culture and healthy employees. The ABWMC program was designed by Peking University and organized by the American International Assurance Company. We invited companies to join the program through the method of dedicated selection. The criteria for the inclusion of participating companies were as follows: (1) legal companies registered in China; (2) at least 100 full-time employees; and (3) consent to participate in the program (30).


The analyzes were based on data from the ABWMC 2021 program. We recruited a total of 79 companies in four provinces: Shanghai, Jiangsu, Guangdong and Beijing. The total sample size was 11,903. The HR departments of each company delivered the questionnaires to all employees. All employees who were (1) 18 years of age or older and (2) full-time employees were invited to participate in this program.

Data collection

Experts from Peking University designed standardized questionnaires, including sociodemographic data, PA-related behavior, and other covariates. We then generated an online questionnaire system and specific questionnaire links based on Ipsop. All questionnaire completion logic (e.g. skipping questions or mutually exclusive options) was set up via the online questionnaire system and a function was set up to exclude certain unqualified questionnaires (e.g. questionnaires with less than 3 minutes of response time were automatically excluded) to guarantee data quality . The human resources departments of each participating company delivered an Internet link to all employees. Participants were required to read an online informed consent form before beginning to answer. All collected data were examined by researchers from Peking University, and respondents were contacted for clarification if any problems were discovered. The study was approved by the Ethics Committee of Peking University Health Sciences Center (IRB00001052-21086). All methods were performed in accordance with relevant guidelines and regulations.


Depressive tendencies

The Center for Epidemiological Studies Depression Scale (CES-D) was used to measure the depressive tendencies of workers (31). We used the revised Chinese version (CESD-9) with 9 questions (32). Participants will be asked to recall how often the feelings described in each item occurred in the past week. Scoring is as follows: less than 1 day = 0, 1–2 days = 1, 3–4 days = 2, and 5–7 days = 3. All items included four response categories that indicated frequency of depressive symptoms. Of the nine items, seven focused on positive symptoms, while the other two (items 5 and 8) assessed negative symptoms of depression. A score is assigned by combining all items (after reversing positive mood items). A total of 27 points on the scale with a score of 10 or more indicated a tendency towards depression. The scale has been tested in previous studies on large Chinese populations, with a Cronbach’s α coefficient of 0.95 and a high sensitivity of the scale (32)

Physical activity

PA behavior was measured using the Taiwanese short version of the International PA Questionnaire (IPAQ) (33). Subjects were asked to recall the frequency and duration of walking and moderate vigorous PA in the past 7 days. We calculated the total time employees spent each week exercising at each intensity level (frequency*duration) and then calculated the average daily PA time (divided by 7). Additionally, we assessed each individual’s total physical activity level based on their weekly PA duration and corresponding MET values ​​for each type of PA (walking: 3.3 METs/min, MPA: 4.0 METs/min, VPA: 8.0 METs/ min). According to the sum of the MET values ​​of each type of PA, the individual’s physical activity level is classified as low (< 600 MET-min/tjedno), umjerena (600–1500 MET-min/tjedno) ili visoka (> 1500 MET-min/week) (33).

Sedentary behavior

The duration of SB was also self-reported by the workers based on the IPAQ questionnaire on average daily SB time (33). We asked respondents to recall the average daily time spent 1) sitting while working or studying, including writing, working on the computer at work, and answering the phone; 2) recreational sitting after work, including resting, reading, playing on the phone or computer, and talking; and 3) time spent in driving or transportation in the past week. The sum of the three types of static sitting time was considered as the average individual daily duration of SB. In the descriptive analysis, we categorized sedentary behavior based on the accumulated sitting time per day into two groups: normal (≤ 8 h/day) and prolonged (> 8 h/day).

Duration of sleep

Participants were asked to recall the average number of hours they slept per day during the past week. We defined eight hours of sleep as enough sleep, while the opposite was considered insufficient sleep.

Potential founders

Potential confounders included gender, age, education level, marital status, type of occupation and workplace, body mass index (BMI), and smoking status (current smoker or nonsmoker). BMI is categorized into three groups: normal (< 24), prekomjerna tjelesna težina (24-28) i pretili (> 28). We entered all these variables into a regression model to control for their confounding effects on depression susceptibility.

Data handling

ISM assumes that an increase in the duration of one behavior in a day leads to a decrease in the time spent on another behavior. Therefore, we adopted the following data cleaning procedures to ensure the reliability of the analysis results. According to the IPAQ data processing and analysis guidelines, (34) outliers of time spent per day for different types of behavior (e.g., SB time, calculated as three standard deviations, with a lower limit of 3.8 and an upper limit of 18) were filtered by the data exploration function, followed by outliers of walking, MPA and VPA times (< 10 min/day) were recorded as 0, extreme sleep outliers were recorded as 7, and SB were recorded as missing. After that, the sum of each independent time was calculated. Considering recall bias associated with self-report questionnaires and concerns about data stability, we excluded individuals with total activity time less than 12 hours and greater than 28 hours, resulting in a final sample size of 10656 (89.5%) individuals. Data cleaning was carried out as follows:

  • SB time for outliers (NOT= 36)

  • Total time > 28 hours (NOT= 149)

  • Total time < 12 h (NOT= 1062)

The final set of analytical data, NOT= 10656 (89.5%).

Statistical analyses

Participants’ depressive tendencies were first described by number and frequency. Chi-square test was used to detect significant differences among different groups.

After that, in multivariate analysis, we first performed a traditional logistic regression analysis as model 1, which included sleep duration, SB, MPA, VPA and other covariates such as workplace and age, to calculate the independent effect of each behavior on depressive tendencies .

Model 1 is shown as follows:

$$\mathrm{Logit}\left(\mathrm{P}\right)= {\upbeta }_{0}+ {\upbeta }_{1}{\mathrm{X}}_{1} \left( \mathrm{sitting}\right)+ {\upbeta }_{2}{\mathrm{X}}_{2} \left(\mathrm{Sleeping}\right)+ {\upbeta }_{3}{\ mathrm{X}}_{3} \left(\mathrm{walking}\right) + {\upbeta }_{4}{\mathrm{X}}_{4} \left(\mathrm{MPA}\right )+ {\upbeta }_{5}{\mathrm{X}}_{5} \left(\mathrm{VPA}\right)+ {\upbeta }_{6}{\mathrm{X}}_{ 6} \left(\mathrm{other covariates}\right)$$

The coefficient β for one type of activity represents the effect of increasing this type of activity while keeping other activities constant in this model.

The ISM was then used to assess the association of time spent in different activities in exchange for equivalent time spent in other activities with risk of depression. This model assumes that the time an individual spends on any one behavior during a fixed 24-hour day results in an isochronous switch to the other behavior, while the total time spent on both behaviors remains constant (35). For example, to estimate the effect of replacing SB with 30 minutes of walking, SB should be removed from the model based on the total time the behavior remained constant.

Model 2 is shown as follows:

$$\mathrm{Logit}\left(\mathrm{P}\right)= {\upbeta }_{0}+ {\upbeta }_{2}{\mathrm{X}}_{2} \left( \mathrm{sleeping}\right)+ {\upbeta }_{3}{\mathrm{X}}_{3} \left(\mathrm{walking}\right)+ {\upbeta }_{4}{\ mathrm{X}}_{4} \left(\mathrm{MPA}\right) + {\upbeta}_{5}{\mathrm{X}}_{5} \left(\mathrm{VPA}\right )+ {\upbeta }_{6}{\mathrm{X}}_{6} \left(\mathrm{other covariates}\right)+ {\upbeta }_{7}{\mathrm{X}}_ {7} \left(\mathrm{Total}\right)$$

Coefficients β in this model represent the effect of 30-minute replacement of SB with one of the types of activity (LPA or MVPA) while keeping the other types of activity and total spending time constant.

All statistical analyzes were performed using SPSS version 22.0 software, and the level of significance was set at p< 0.05.

Leave a Reply

Your email address will not be published. Required fields are marked *