Design and study population
A cross-sectional study was performed from January 1, 2014 to December 31, 2014 in the occupational health services of the Languedoc-Roussillon region and in the town of Roanne in France. Eighty-three occupational health physicians (OHP) participated in the recruitment of workers.
According to French labor law, each woman having worked during pregnancy must benefit from a medical visit with an OHP after delivery and at the time of returning to work. This is a compulsory visit for employers and employees. Also, at the first postnatal visit, all women (no selection method) were invited to participate in the study by the OHP. The inclusion criteria were that they: (i) must have worked for an employer during their last pregnancy (our study did not include workers without employment contract (such as self-employed, craftsman, farmer, company head…) because these workers were not followed by French occupational health services); (ii) to be older than 18 years of age (legal majority in France) (iii) must have had a postnatal visit with an OHP in the year after delivery; or (iv) within three years after delivery if they had full-time parental leave; and (v) be sufficiently fluent in French to participate in the interview. Before the medical visit, the OHPs asked eligible workers to fill-in a self-administered questionnaire to obtain sociodemographic information; during the visit, a second occupational questionnaire was administered face-to-face by the OHP. The computerized medical record system in each occupational health service was declared to the French National Commission for Data Protection. All the volunteer participants gave their informed consent to be enrolled and data were collected anonymously. Because we used anonymised data from routine medical visits, in 2013 according to French law, ethics approval was not required.
Outcomes of interest
Information on SL during pregnancy was obtained from workers with the following questions by taking account of leaves for legal reasons and sickness: (i) for the first and second trimester of the pregnancy (“did you have one SL before a pathological or legal leave (regardless of the cause and duration)?, yes/no); (ii) on the duration (“what was the total duration of your SL, before a pathological or legal leave?, in days during pregnancy); (iii) “did you take a SL without returning to work before a pathological or legal leave ?, yes/no ? If yes, at what time of your pregnancy did you stop working completely? in which week of gestation (WG)”. In this study, we defined each trimester in the following way: first trimester when < 15 WG; second trimester from 15 WG to 28 WG. Early SL was defined as leaving job before 15 WG.
Exposure of interest
The exposure assessment of potential hazards for pregnancy was conducted by OHPs based on knowledge of workstations in early pregnancy. Seventeen potential hazards were selected [3]: biological hazards (working with very young children, sick persons, animals); chemical hazards; night work (between 9:00.p.m and 5:00 a.m.); physical hazards (standing > 1 h a day, stair climbing (several times a day), forward bending ≥ 1 h a day, difficult postures (upper and/or lower limbs), heavy lifting > 5 kg, repetitive tasks, vibration (driving), temperature (>30 °C, <10 °C), noise >80 dB, work on industrial machines); ionizing radiation and electromagnetic fields. The responses were based on a 4-point Likert scale ranging from 1: no; 2: very rarely (a few per month); 3: sometimes (a few times a week); 4: frequently (a few times a day or more). Then, all these variables were transformed into binary variables and were coded as either 0 (to indicate the reference category) or 1 (to indicate the category at risk). For all the variables, the category at risk was the “frequently” category (level 4) except for three variables: ionizing radiation (level from 2 to 4); night work (at least one night); Electromagnetic fields (level from 3 to 4). A cumulative index of occupational hazards for pregnancy in four classes (0, 1–2, 2–4, ≥ 5 risks) was built using these seventeen occupational variables.
Potential confounders
The choice of potential confounders was based on the literature data (age, deprivation, occupational data, number of children, assisted reproductive therapy (ART), and pregnancy at risk) except for smoking, alcohol consumption, and body mass index. We have considered that socioeconomic deprivation was a proxy measure for these last three variables because very positively correlated to them [25]. Socioeconomic deprivation was assessed using the Evaluation of Deprivation and Inequalities in Health Examination (EPICES) individual scale [25]. This is a reliable proxy in workplace settings for population-based measures of deprivation which is strongly correlated with the Townsend and Carstairs indices [25]. Workers with an EPICES score equal to or higher than 30 were classified as being deprived. Occupational skill level was classified according to the French standard classification of occupations (version 2003) from the French National Institute of Statistics [26]. The occupations were classified into four skill levels: managers/supervisors, intermediate occupations, employees, and manual workers. Women were asked if they had been followed-up for “pregnancy at risk of medical complications” (yes/no) and whether they had had assisted reproductive therapy (yes/no). Other explored factors were maternal age at delivery, company size at four levels (<10, 10–49, 50–199, ≥ 200 workers), type of contract (non-fixed term, fixed term), home-work commuting (duration, mode), working time (full-time, part-time), and job duration (<2, ≥ 2 years).
Several factors, no used in literature like confounders until now, related to concept “work-family conflict” were also used [27, 28]. This concept focuses on the difficulties that employees have in balancing their work and family responsibilities which could increase fatigue during pregnancy and the occurrence of SL [27, 28]. In our study, several predictors of work-family conflict were used as follows. An index of cumulated work-family conflict risks in four classes (0, 1, 2, ≥ 3 risks) was built using five dichotomous variables: preschool-age children at home (yes/no), home-work commuting > 50 min/d (yes/no), duration of working hours > 8 h/d (yes/no), irregular working hours (yes/no), and absence of two consecutive rest days in a week (yes/no).
Statistical analyses
In bivariate analyses, chi-square tests were used to compare binary variables. Also for continuous variables, the Kolmogorov-Smirnov test was used to evaluate the normal distribution. A Student’s t-test was used to analyze the normally distributed quantitative values, and the Mann–Whitney U test or the Kruskal-Wallis test was used to analyze the non-normally distributed ones. In multivariate analyses for binary outcomes, adjusted relative risks (RRa) with 95% confidence intervals (CIs) were calculated based on a generalized linear model with a Bernoulli distribution and a log link adjusted for selected confounders [29]. A stepwise forward procedure was conducted to identify the variables having a significant association with the outcome. For “at least one sick leave” variable, we have conducted analyses separately for each trimester of pregnancy because the impact of the factor may be different according to the pregnancy period. To compare the results between trimesters, we have presented the significant variables from final models of the stepwise forward procedure but also the non-significant variables adjusted for these significant variables. Also, we tested interaction terms between several variables. Notably, to answer our hypothesis, analyses between “number of occupational risks for pregnancy” and “pregnancy at risk” were carried out separately when a positive interaction was identified between these two variables. For count outcome, zero-inflated negative binomial regression (ZINB) was carried out to take into account over-dispersion and/or excess of the zero value in data [30]. For this outcome the purpose of the analysis was exploratory, we have presented the results without a selection procedure of the variables. A two-sided p-value of less than 0.05 was considered statistically significant in our study. Statistical power was estimated on the basis of a binomial test with unequal group sizes (ratio 1/4). In this case to detect an effect size of RR = 2 with 80% power at a significance level of 0.05 and a control-group proportion at 5%, 1,245 pregnant workers were required. As the responses were relatively complete, analyses excluded missing data. All the statistical analyses were performed using STATA statistical software, version 14.0 (Stata Corp, College Station, Texas, USA).