MATERIALS AND METHODS

Study population

The study was carried out within the DNBC, which has been described in detail elsewhere.^20–22 Candidates for the cohort were all pregnant women in Denmark who, at their first visit to their general practitioner, wanted to carry their pregnancy to term and who spoke Danish well enough to take part in the telephone interviews. Participants were given the possibility to indicate a preferred day and time for the interviews covering normal working hours, evenings, or weekends. Approximately 60% of all pregnant women accepted the invitation to join the cohort from almost 60% of the general practitioners who took part in the recruitment. From 1 March 1998 to 1 May 2000, 39 913 pregnant women (40 635 pregnancies) were enrolled in the DNBC, giving their first telephone interview (12–16 completed gestational week), and still being pregnant at that time. They were asked to state their occupation and work schedule. We identified 30 467 pregnancies in women who had a job and had stated their occupation and work schedule. We excluded pregnancies in women with endometriosis, ovarian cancer, or cervical cancer. We excluded unplanned pregnancies, partly planned pregnancies, and pregnancies that occurred after infertility treatment from the main analysis. The proportions of unplanned pregnancies were higher among shift workers. Some women participated in the cohort with two pregnancies during the study period; we excluded the second reported pregnancies to achieve statistical independence. We ended up with 17 531 daytime workers, 493 fixed evening workers, 177 fixed night workers, 1572 rotating shift (without night) workers, and 1665 rotating shift (with night) workers, who were eligible for the main analyses (table 1).

Exposure assessment

In the first interview, the women were asked in detail about their occupation, including the number of jobs, their job title, type of work, working hours, work schedule, and number of night shifts. Their partners’ occupations were also recorded. We grouped the women by occupation according to the Danish version of the International Standard Classification of Occupation (DISCO-88)²³ into: managers and professionals (0110–2470), technicians (3111–3480), service and sales workers (4111–6210), and industrial workers (7111–9333). We classified weekly working hours into two categories: <35 hours and 35+ hours. The work schedule question was asked as follows: “Do you primarily work during the day, during the evening or during the night, or do you have changing working hours?” The mutually exclusive answering categories were: 1, daytime; 2, fixed evening; 3, fixed night; 4, rotating shift (without night); 5, rotating shift (with night). We grouped the frequency of night shifts into 1–8 and 9+ times per month. We furthermore identified two specific job types with the largest numbers for an internal contrast in work schedules: nurses (n = 1624) and nursing assistants (n = 899). Their partners’ jobs were classified in the same categories, except when they reported having no job.

Measurement of outcome and potential confounders

TTP was recorded according to the following questions: “How long did you try to become pregnant, before you succeeded? 0–2 months, 3–5 months, 6–12 months, or >12 months?” Additionally, the woman was asked if the pregnancy was planned, partly planned, or not planned. Only planned pregnancies were included in the analyses presented in the tables.

Potential confounders were categorised as shown in table 2. They included age at conception, gravidity, prepregnancy body mass index (BMI), smoking, and alcohol consumption. Maternal age at conception was computed by subtracting the woman’s birth date from the last menstrual period and adding 14 days. Prepregnancy BMI was calculated on the basis of the women’s report on height and weight before pregnancy. We used the question whether they had smoked at any time in the first trimester to categorise them as either smokers or non-smokers. Alcohol consumption was classified in categories as described elsewhere.²² In brief, we added beer, wine, and spirits to one variable of total alcohol consumption per week, according to the woman’s report on her drinking habits before pregnancy. One bottle of beer contains 11.6 g of alcohol, and 12 g of alcohol is an approximate average for one unit of wine or spirits in Denmark. If less than one unit per week was reported by the woman, 0.5 units were coded. We then grouped them by these levels: 0, 0.5–7, 7.5+ units (12 g alcohol per unit) per week. It has previously been shown²¹ that neither interviewer’s habits nor her attitudes towards smoking and alcohol consumption during pregnancy had consequences for responses obtained. Likewise, the education, age, or parity of the interviewer did not correlate with the answers obtained. We had limited data on the partner and no information on sexual activity or sperm quality.

Statistical analysis

In general, fecundability is defined as the probability to conceive in a menstrual cycle. In our study, we estimated the probability of obtaining a clinically recognised pregnancy in a waiting time interval among women not pregnant in the previous interval, conditionally that they did get pregnant. Fecundity odds ratio (OR) measures the odds of a conception within each waiting time interval among the exposed divided by the odds among those not exposed. This measure will correlate with fecundability under most conditions, but it is not a measure of fecundability. ORs were calculated with 95% confidence intervals (95% CI) using the discrete time survival analysis techniques performed by logistic regression, which is a non-proportional hazard model,²⁴ to estimate the effects on shift workers compared with daytime workers. TTPs in our study were measured in discrete times (four intervals). We first broke down each individual’s TTP into a set of intervals that were treated as distinct observations. After pooling these observations, the next step was to estimate a binary regression model (logistic model) predicting whether a conception occurred in each interval and a variable to indicate that the interval was fitted into the model while covariates were allowed. We repeated all analyses for only those women who tried to become pregnant for the first time, since a previous TTP may modify risk behaviour in subsequent attempts in such a way that may be impossible to adjust for. All analyses were restricted to women who had a job at the first interview. We estimated the effect of night shifts on TTP by including only fixed night workers and rotating shift (with night) workers in the analysis. Potential confounders were included in the model, as well as a variable to indicate waiting time interval and the couple’s occupational characteristics. Analyses were performed using SPSS 10.0.

RESULTS

The proportion of shift work was 19.1% among women who had a job. Table 2 shows the primary characteristics of the study population. Characteristics of rotating shift workers (with or without night) were similar to that of daytime workers, except that most rotating shift workers were technicians and most daytime workers were service and sales workers. Fixed evening workers and fixed night workers were heterogeneous populations compared with others. They were younger, more often parous, obese, and smokers, and they more often worked as industrial workers, as did their partners. Their median weekly work hours were 28 hours and 30 hours, respectively, while daytime and rotating shift (with or without night) workers worked 37 hours per week. By agreement, full time employment in Denmark equals 37 hours a week. The median number of night shifts was 14 a month for fixed night workers and four for rotating shift workers (with night).

Compared with daytime workers, fixed evening workers and fixed night workers had a longer TTP, while rotating shift (with or without night) workers had similar TTP. The crude ORs were 0.83 for fixed evening workers, 0.75 for fixed night workers, 1.00 for rotating shift (without night) workers, and 1.11 for rotating shift (with night) workers. The adjusted risk estimates never changed more than the 10% limit (table 3). The adjusted OR for working at night more than nine times per month was 0.97 (95% CI 0.70 to 1.34) compared with 1–8 times per month.

The estimates did not change when we included only the first pregnancies (data not shown). When we restricted the analysis to nurses, none of the shift workers had a statistically significant low fecundity; the adjusted ORs were 0.99 (95% CI 0.71 to 1.38) for fixed evening work, 0.67 (95% CI 0.39 to 1.15) for fixed night work, 0.96 (95% CI 0.73 to 1.24) for rotating shift work (without night), and 0.99 (95% CI 0.81 to 1.22) for rotating shift work (with night), compared with daytime work (data not shown). When we did an analysis for nursing assistants, only fixed evening work (OR 0.73; 95% CI 0.54 to 0.99) showed a statistically significantly decreased fecundity (data not shown).

We performed separate analyses for smokers and non-smokers, since smoking was more prevalent among fixed evening work and fixed night work and correlated with TTP (OR 0.82; 95% CI 0.78 to 0.87). Shift work did not show any statistically significant association with fecundity among non-smokers. In smokers, fixed evening workers and fixed night workers had a decreased fecundity: OR 0.70 (95% CI 0.54 to 0.90) and 0.63 (95% CI 0.44 to 0.89), respectively (data not shown).

DISCUSSION

We did not find an association between rotating shift work (with or without night) and reduced fecundity. This finding was consistent for all, for nulliparous women, for nurses and nursing assistants, for non-smokers, and for smokers. The median number of working nights per month was only four for rotating shift workers (with night). If night shifts have only transient effects (say, few days), we may have limited possibility to detect it. We found no difference in TTP among women working at night more than 9 times per month and women working at night 1–8 times per month after adjusted for potential confounders.

Fixed evening workers and fixed night workers had different results. These workers were, however, very different from daytime workers and rotating shift workers. The proportions of unplanned pregnancies were much higher among these workers (table 1). Among unplanned pregnancies, the proportions of contraceptive failures were higher for fixed evening workers (21.8%) and fixed night workers (27.8%) than for daytime workers (17.0%). TTP is undefined for unplanned pregnancies, and because these unplanned pregnancies are excluded from the analysis of TTP, pregnancy planning bias could arise.^19,25,26 The highly fertile couples are more likely to have their children unintentionally (for example, as contraceptive failures) and thus never be eligible for the study of TTP. Meanwhile, the less fertile will be more likely to plan their pregnancies carefully, making themselves eligible for study. This self selection can produce an artefact by which shift work appears to impair fecundity, as seen in our study for fixed evening workers and fixed night workers. No effect for fixed evening work or fixed night work was seen when we restricted the analyses to nurses whose pregnancy planning status had no statistically significant difference among different work schedules. Infertile workers may also be overrepresented in the groups of fixed evening and night workers, as this type of work may be less adaptable to family life. Some families, however, argue that a fixed schedule is more conducive to family life, as it means better planning opportunities (you always know when to work) compared with the often irregular working schedules of the rotating staff. We also found that fixed evening and night workers were more often parous and had more children in the family.

Subfecundity is a couple’s characteristic, but we have limited data on the male partner. We adjusted for their partners’ occupations. Couples who have unprotected sexual intercourse around the time of ovulation are more likely to get pregnant in that cycle. Fixed evening and fixed night workers may have less opportunity for having sexual activity. However, both the Swedish study¹² and the European multicentre study¹³ showed that the frequency of sexual intercourse was evenly distributed among daytime workers, rotating shift workers, and fixed night workers. Working time may affect the timing of intercourse, which is more important than frequency for conception.

We do expect that the effect of previous use of contraceptive pills^27,28 would bias our results since we had limited data on it. We asked the women about the use of oral contraceptives during the past four months before pregnancy and found that among those having TTP 0–2 months, the proportions of the previous use of pills were lower for fixed evening workers (27.9%) and rotating shift (with night) workers (32.2%), but not for fixed night workers (41.3%), and for rotating shift (without night) workers (38.5%), compared with daytime workers (38.8%). However, when we only looked at getting pregnant during 0–2 months of waiting time and performed an analysis after adjusting for the previous use of oral contraceptives and other potential confounders, we obtained similar results. Menstrual cycle characteristics are also known determinants of female fecundity, but we had only limited data to analyse the importance of this factor.

Three previous studies^12–14 found an association between shift work and women’s fecundity, while three others^15–17 did not. The differences in working condition for shift workers in different populations and the pregnancy planning bias could be major reasons for inconsistent results. Although the European multicentre study¹³ concluded that data from the study were in favour of an association between shift work and prolonged TTP, they found no association between shift work and subfecundity in the population sample with first pregnancies (OR 1.0; 95% CI 0.7 to 1.5) and no effect of shift work on menstrual cycle length and irregular bleedings. They stated “it is likely that shift work is only a risk indicator”.

The DNBC recruited about 60% of invited pregnant women (30–40% of all pregnant women). The response rate could cause selection bias if the decision to participate was associated with both exposure to shift work and TTP. We believe this to be unlikely, since studying determinants of subfecundity was not specified as the aim of the birth cohort.

This study included only women who had a clinically recognised pregnancy and all effect measures were restricted to pregnancies that survived the first 12–16 weeks of gestation. If the exposures under study lead to an all or none effect or early fetal loss, we had no possibility of detecting such an effect. Most experiences so far show that only few exposures have such an effect.²⁵

We used telephone interviews to obtain information on TTP. Validation studies of TTP have shown that recall of TTP is accurate, even over longer time periods.^26,29 In our study they only had to remember a few months back in time.

We asked for work schedules in the first interview. We have no data on work status at the start of pregnancy planning. If women who experience unsuccessful pregnancy attempt to change their work schedules, the results may be misleading. We believe this to be a minor problem. The proportion of shift work in our study population was comparable with the previous reports in Denmark.³⁰ Secondly, it is unlikely that a woman would change daytime work to shift work after getting pregnant. In order to minimise the potential bias of behaviour modification and the female reproductively unhealthy worker effect, we restricted the analysis to the women who conceived their first pregnancy and obtained the same results.

To study determinants of subfecundity also assumes comparable persistence in pursuing a pregnancy attempt among the compared groups.³¹ We expect this problem to be of minor importance after correction for age and gravidity.³² Other potential confounders, such as prepregnant BMI, smoking, and alcohol consumption,^33–35 were also controlled in the analyses. Since BMI is associated to shift work and may be related to a biological effect due to hormone disturbances,³⁶ we performed the same analyses without this variable in the models and found that the estimates remained unchanged.

In conclusion, we found no evidence of a causal association between shift work performed in our study and subfecundity. The slightly reduced fecundity among fixed evening workers and fixed night workers may be mediated by pregnancy planning bias.