Participants were selected between May and December 2013 using a multistage cluster and stratified random sampling. All cities in Xinjiang with administrative bureaus and petrochemical companies listed in the China Petroleum and Petrochemical Species Classification Catalog were identified, and one city was selected randomly. Five petrochemical companies in one city were randomly selected, and all the employees in each company were divided into 4 groups, giving a total of 20 groups. All groups were numbered, and 10 groups were randomly selected according to the random number table method. A total of 3400 were selected for study. They underwent health examination in 2013 at Karamay City Center for Disease Control and Prevention in Xinjiang and filled out a questionnaire with basic information. Those eligible for the study were employees of the Karamay City Petroleum Administration and Petrochemical Company who had been employed in that position for 1 year, were 20–60 years old, and had signed an informed consent form. Patients with dyslipidaemia at baseline (n = 1037), diseases affecting blood lipids, medications affecting blood lipids, or diet (n = 52), or hair shorter than 3 cm (n = 11) were excluded. Those who answered less than 80% of the questions in the questionnaire (n = 48), left work, and were unavailable during follow-up (n = 69) were also excluded. The survey was launched in May 2013 and included a 6-year follow-up period during which participants did not change their shift work. Participants were followed up with questionnaires and occupational health examinations at the Karamay Center for Disease Control and Prevention in Xinjiang from May to December 2014, 2015, and 2019. The study cohort included 2170 participants, 1021 men and 1149 women; 1348 were shift workers and 822 were non-shift workers.
In the early stage of the study, we regarded the shift population as the exposed group, with an incidence rate of p1 = 0.238, and the general adult population as the non-exposed group, with an incidence rate of p0 = 0.186 [32, 33]. Take the test level α = 0.05, the power of the test was 1- β (take β = 0.10). The formula for calculating the sample size was as follows:
The required sample size was calculated to be 1297. The sample size in this study met these requirements.
We used a self-reported questionnaire to obtain information on shift work patterns, family medical history, and personal information such as smoking and drinking. Employees who regularly worked fixed-day shifts from 8:00 am to 5:00 pm were considered non-shift workers. Employees who worked night shifts were considered shift workers and were divided into two, three, and four shifts as described below. “Two shifts” included two 12-hour shifts and two groups of workers alternating weekly; “Three shifts” included two 12-hour shifts with three groups of workers alternating weekly, with one of the groups resting; “Four shifts” included three 8-hour shifts (morning, mid, and evening) with four groups of workers working alternately and with one at group at rest. Shift work was thus divided into four groups: fixed day shift, two shifts, three shifts, and four shifts.”
Blood lipid data was obtained at annual occupational health examinations. Dyslipidemia was determined by measuring the concentration of cholesterol in the four lipoproteins, total cholesterol (TC), triglycerides (TG), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL-C) . In participants, dyslipidemia required one of the following results in two assays performed 2 weeks apart, TC > 5.18 mmol/L, TG > 1.7 mmol/L, and HDL-C < 1.04 mmol/L within 2 weeks meet conditions .
During the baseline period, we randomly divided 2170 participants into 70 groups of 31 subjects each, and randomly selected 11 of these groups, to collect hair samples from a total of 341 subjects to collect hair samples. Researchers reported that natural hair color had no effect on hair cortisol concentrations , and Sauvé et al. found that chemically treated hair (dyed hair) to have significantly lower hair cortisol concentrations than untreated hair . Finally, after deleting 5 maximum and 4 minimum values and 1 unnatural hair color, the hair cortisol concentrations of 331 subjects were included in the analysis.
Hair samples (2–3 cm, 20–30 mg) were collected from the hair roots of the participants. Pretreatment of hair samples was performed according to the experimental protocol described in the patent, “Pretreatment method for detecting cortisol content in hair” . The hair sample was soaked with 2–3 ml of isopropyl alcohol for 5 minutes, washed and peeled, then frozen in liquid nitrogen for more than 4 hours and then pulverised. The pulverised hair sample was placed in a centrifuge tube, mixed with 5 ml of methanol solution and 3 ml of ether solution, and placed in a water bath at 50.8 °C for 16 h for extraction and incubation. During analysis, the hair fragments were mixed by multiple inverting and centrifuged at low speed at 3500 rpm for 15 min. The supernatant was transferred to a 4 ml Eppendorf tube, and the extracted mixture was dried with a nitrogen blower. After the addition of 2 ml of phosphate buffer solution, the sample was stored at − 4 °C in a refrigerator until the day of testing. HCC was detected using an automated radioimmunoassay.
Covariates included sex, age, body mass index (BMI, kg/m2), ethnicity, marital status, education level, family history of hypertension, coronary heart disease, stroke, diabetes, income level (Yuan), job tenure (years), type of work, smoking, drinking, and exercise. Participants were stratified by age (youth group: 20–29 years, young and middle-aged group: 30–39 years, and middle-aged and elderly group: 40–60 years), BMI (Chinese standard BMI value: low body weight: < 18.5 kg/m2, normal weight: 18.5–23.9 kg/m2, overweight: 24–28.0 kg/m2, and obese ≥28 kg/m2 ). Ethnicity was divided into “Han”, “Uygur” and “other minority”. Marital status was divided into ‘not married’, ‘married’, and ‘other’ (divorced, widowed, or remarried, respectively). The educational level was divided into ‘high school or below’, ‘junior college education’, and ‘college or above’. A family history of hypertension was subdivided into ‘yes’, ‘no’, or ‘unknown’. A family history of coronary heart disease was subdivided into ‘yes’, ‘no’, or ‘unknown’. Family history of stroke was classified as ‘yes’, ‘no’, or ‘unknown’. Family history of diabetes was divided into ‘yes’, ‘no’, or ‘unknown’. The income level (Yuan) was divided into ‘< 3000/$422’, ‘3000-5000/$422–$736’, and ‘>5000/$736’ Yuan. Job tenure was divided into ‘< 10’, ‘10–20’, and ‘≥ 20’ years. The type of work was divided into ‘oil’, ‘oil recovery’, ‘refining’, and ‘other’. Smoking was divided into ‘often’ (≥1 cigarette/day), ‘occasional’ (< 1 cigarette/day), ‘quit smoking’, and ‘nonsmoking’. Drinking was divided into ‘often’ (≥ 8 g/day), ‘occasional’ (< 8 g/day), ‘quit drinking’, and ‘nondrinking’. Physical exercise was divided into ‘no exercise’, ‘< 3 times/week’, ‘≥ 3 times/week’, and ‘irregular’.
EpiData3.0, the questionnaire’s double-track data entry software, and STATA13.0 were used to organise and analyse the data. Measurement data were described as mean average (¯X) ± standard deviation (SD) or median and interquartile range [M(Q1-Q3)] and geometric mean concentrations (GM) ± the GSD to improve statistical power. Comparison of measured data was performed using the t-test or analysis of variance, and comparison of count data was performed by χ2 test. Four models were established to perform logistic regression analysis between indicators. Model 1 represented associations between indicators without adjustment for confounders, and Model 2 was adjusted for gender, age, ethnicity, marital status, education level, type of work, length of service, and average monthly income. Model 3 was adjusted for smoking status, drinking status, physical exercise, and BMI based on Model 2. Model 4 was adjusted on the basis of Model 3 for hypertension, coronary heart disease, stroke, and family history of diabetes. Linear regression was used to analyse the association between HCCs and changes in blood lipid levels. HCC values showed a skewed distribution. Make HCC values normally distributed by log transformation. Shift work was divided into five groups according to shift pattern and a fixed day shift as a reference group.
We conducted a mediating-effect analysis to understand the mechanism by which one variable affects another. The coefficient between shift work and dyslipidemia was the overall effect. When HCC was the mediator, the coefficient between shift work and dyslipidemia represented a direct influence. The mediation effect was calculated by subtracting the direct effect from the total effect . Previous studies have shown that excessive HCC may have an effect on dyslipidemia . Methods described by Karlson, Holm, and Brin  were used to verify the significance of the HCC effect. If both the overall effect and the indirect effect were significant and the direct effect was not, then HCC was considered to regulate the relationship between shift work and dyslipidemia . However, if all the effects were significant, then HCC was considered to have played a role in mediating the outcome . We used this method to estimate the percentage of the total effect mediated by HCC.
All participants signed an informed consent form after receiving information about the study. This study was approved by the Nantong University Ethics Committee (2013-L073).