|Home | About | Journals | Submit | Contact Us | Français|
MB, KTP and DC designed the study protocol, developed search strategies and established selection criteria. MB, KTP and MC performed the data collection and applied quality score, MB and KTP performed the meta-analyses; MB and CA wrote the manuscript under the supervision of MMF and DC. MC, DC, and KTP contributed to the data interpretation and critical revision of the manuscript.
Varying work schedules are suspected of increasing risks to pregnant women and to fetal wellbeing. In particular, maternal hormonal disturbance arising from sleep deprivation or circadian rhythm disruption might impair fetal growth or lead to complications of pregnancy. Two independent meta-analyses (from 2000 and 2007) reported a small adverse effect of shift work on the risk of preterm delivery (PTD). However, these reviews were based on few high quality studies.
To provide an updated review of the associations of shift work with PTD, low birthweight (LBW), small for gestational age (SGA), and pre-eclampsia.
We conducted a systematic search of MEDLINE using combinations of keywords and MeSH terms.
for each relevant paper we abstracted standard details, used to summarize design features and rate methodological quality. We calculated pooled estimates of relative risk (RR) in random-effect meta-analyses.
We retrieved 23 relevant studies. The pooled estimate of RR for PTD was 1.16 (95%CI 1.00-1.33, 16 studies), but when five reports of poorer methodological quality were excluded, the estimated RR reduced to 1.03 (95%CI 0.93-1.14). We also observed increased RRs for LBW (1.27, 95%CI 0.93-1.74) and for SGA (1.12, 95%CI 1.03-1.22), which varied little by study quality. Little evidence was found on pre-eclampsia.
These findings suggest that overall, any risk of PTD, LBW, or SGA arising from shift-work in pregnancy is small.
Several authors have suggested that varying work schedules (including rotating shifts and night-work) may present special risks to pregnant women, including neuroendocrine changes as a consequence of sleep deprivation or disrupted circadian rhythms affecting fetal growth and the timing of parturition. In support of this, some observational studies have found a higher risk of preterm delivery (PTD) or low birthweight (LBW) among women exposed to shift work.3 In addition, one study has suggested that shift work during pregnancy carries an increased risk of pre-eclampsia,4 whereas another suggested that it did not.5 These pregnancy-related complications are major contributors to perinatal mortality and morbidity so identification of modifiable risk factors such as working conditions is an important priority in maternity care. Moreover, given the mounting evidence linking LBW with an increased risk of chronic diseases in adulthood,6-8 research into the effects of workplace exposures on pregnancy is a priority also for public health. In 2007, we published a systematic review of the epidemiological evidence relating common occupational exposures (prolonged working hours, lifting, standing, heavy physical workload and shift work) to adverse pregnancy outcomes, in which a pooled meta-analysis indicated a small but significantly increased risk of PTD in association with shift work (RR=1.20, 95%CI 1.01-1.42, p=0.002).9 This was in accord with the finding of a previous partially overlapping meta-analysis, published in 2000, which gave a pooled RR estimate of 1.24 (95% CI 1.06-1.46).3 However, because of the possibility of bias and residual confounding, these meta-analyses cannot be considered conclusive.
We have now updated our earlier review of the associations of shift work with PTD and indices of fetal growth retardation, including a further 5 years (Feb 2005-Feb 2010) of published epidemiological evidence.
We conducted a systematic review of the epidemiological evidence relating shift work and four important pregnancy outcomes: PTD, (LBW), small for gestational age (SGA) and pre-eclampsia. We systematically searched the Medline bibliographic databases for the period 1966 to February 2010, using medical subject headings (MeSH terms). Preterm birth, gestational age, small for gestational age, fetal growth restriction, pregnancy complications, pre-eclampsia and reproductive health were used to represent outcomes of interest; while work schedule tolerance, work and workload were selected to represent relevant exposures. We also supplemented our search using “shift work” as a simple search term. We limited our search to papers with an abstract, written in English, relating to humans. We examined the abstracts, eliminated irrelevant references, and obtained paper copies of all primary reports and reviews. We also cross-checked the references of all selected papers to identify other potentially relevant reports. Finally we included all papers reporting at least one risk estimate for one or more of the specified outcomes, comparing women exposed to shift work with working women who were unexposed (or less heavily exposed).
For each paper that met our inclusion criteria, we abstracted a standard set of information: year of publication, study location, timing of investigation, study design, strategies for exposure assessment and period of pregnancy in which exposure occurred, method of outcome assessment, and risk estimates with associated 95% CIs. If 95% CIs were not reported, we extracted the information that was needed to calculate them.
Preterm delivery was consistently defined by the authors of primary reports, according to the World Health Organization definition, as the birth of a baby at <37 completed weeks of gestation. The definition of LBW was a birth weight <2500 × g. One study also reported risk estimates for birth weight <3000 × g.10 Authors investigating birthweight seem not to take into account specific variables such as maternal weight gain and fetal sex in defining ‘normal birthweight’, as suggested by Gardosiet al.11
The definition of SGA was a baby with a birth weight below the 10th percentile for gender and gestational age. One study considered also babies below the 5th percentile.12 Distributions of birthweight by sex and gestational age were usually nationally or regionally based. Hence thelimit at which a baby was classed as SGA varied between studies.
Pregnancy-induced hypertension was defined as hypertension in a previously normotensive woman, with different thresholds among studies: hypertension was defined alternatively as having blood pressure≥140/90 mmHg on at least two occasions occurring from 20 weeks of gestation onwards,13 or having an increase of at least 20 mmHg in the mean arterial blood pressure between the mother’s first and last visit.5 Pre-eclampsia was defined as gestational hypertension in association with proteinuria/albuminuria, either occurring at some time from 20 weeks of gestation13 or diagnosed more than once during the pregnancy;4 no cutoffs were specified.
We classed studies as high quality where reporting was complete and they were free from important bias or confounding. Completeness of reporting was evaluated as proposed by Ariens et al.14 and by Van der Windt et al.15 with modifications for studies of pregnancy outcomes,9 using a nine point scale covering the following items: (1) study design; (2) sampling procedure; (3) inclusion/exclusion criteria; (4) characteristics of the study population (age, social class); (5) study numbers and response rate; (6) exposure definition and assessment methods; (7) outcome assessment methods; (8) methods of statistical analysis; and (9) quantitative risk estimates with 95%CIs (and not only p values).
As reported in our previous review 9 we judged studies as prone to important bias where there was potential for recall bias that could lead to overestimation of risks. This was the case for retrospective studies investigating the association between shift work and outcomes that were both self-reported and clearly adverse (such as pre-eclampsia or PTD).
For each examined outcome, we classified reported associations as susceptible to confounding if important established risk factors were not taken into account. The importance of an established risk factor was based on the size of the relative risk that it carried, its prevalence in the general population and the likelihood that it might vary importantly between different occupations.
For studies with sufficiently similar definitions of exposure and outcome, we calculated pooled risk estimates with a random effect meta-analysis, assuming heterogeneity of effects, following the maximum likelihood method.16 Each risk estimate was weighted by the inverse of its variance. To increase comparability across studies, where authors presented RRs for exposures during more than one period of pregnancy, we focused on exposures during the first trimester as they were the most widely investigated. We also conducted a sensitivity analysis, checking the impact on pooled risk estimates of excluding lower quality studies, and we explored possible publication bias using funnel plots. All statistical analyses were performed using STATA software version 11.0 (Stata Corp College Station, TX, USA).
As set out in Tables 11--3,3, our search identified 17 original studies that investigated the association between shift work and PTD, including three published after 2005; ten studies concerning SGA (four published after 2005); six studies concerning LBW (one published after 2005) and three studies investigating the relationship between shift work and pre-eclampsia and/or gestational hypertension (one published after 2005). Tables 1--33 summarise quantitative risk estimates and quality assessments, with studies grouped by health outcome and then by type of study.
Sample size varied substantially across studies - from fewer than 400 pregnancies 17,18 to more than 35,000.19 Also, the proportion of women exposed (a determinant of the number of exposed cases) differed between studies, ranging from <5 to 20%.10 As demonstrated by Croteau et al.,20 the proportion of women exposed to shift work decreases over the course of pregnancy. Consequently, studies investigating the effect of exposure late in pregnancy tend to include fewer exposed cases than those investigating the effects of shift work earlier in pregnancy.
In eight cohort studies 10, 12, 18, 19, 21-24 exposure was ascertained prospectively during pregnancy, whereas in 13 studies (case-control or cross-sectional) information about occupational exposure was ascertained after delivery. Eight studies did not report the timing of exposure during pregnancy; most of the remainder focused on the first and second trimesters.
The types of shift work considered relevant varied between studies, some comparing night workers with day workers 13, 21, 23 and others investigating workers on rotating shifts.19, 24 Moreover, several papers did not specify what kind of shift work was investigated, or classed women as exposed if they worked either at night or in rotating shifts.
Reporting was generally satisfactory, but we judged it to be incomplete for four studies. A report by Hartikainen-Sorri and Sorri.17 did not provide information about methods of exposure assessment, inclusion criteria or the period of pregnancy for which exposure was assessed; McDonald et al. 25 did not report confidence intervals for risk estimates (only that the observed elevation of risk was statistically significant) and did not state the total number of women studied or the times during pregnancy to which exposures to shift work related. In one paper, Saurel-Cubizzolles and Kaminski.26 reported only a crude analysis comparing night workers with others, indicating nonsignificant differences in the prevalence of LBW and PTD, and gave no adjusted risk estimates. Finally, a recent study by Abeysena et al.12 evaluated the risk of SGA in women exposed not only to shift work but also to “other occupational exposures” that were not clearly specified. Risk estimates for shift work exposure in the absence of other occupational hazards were not reported. None of the above mentioned studies was included in the meta-analysis, which was restricted to high quality studies.
In two studies, both exposure and relevant health outcomes (PTD or pre-eclampsia) were self-reported after delivery.27, 28 Hence, we classified these studies as susceptible to important recall bias and excluded them from our sensitivity analysis.
Various maternal characteristics have been recognised as risk factors for PTD (extremes of maternal age, maternal weight, height and ethnicity, socio-economic status, smoking and alcohol intake, substance abuse, multiparity, primiparity, diabetes, pre-eclampsia, bacterial infections during pregnancy, and other maternal diseases). However, of these, only smoking and socio-economic status (or proxies such as maternal education or income) are believed to be associated with more than small risk increases in risk (RR>1.5) and are likely to differ importantly in prevalence between occupations. Three studies 17, 29, 30 failed to control for both smoking and social class, and were thus classified as having a higher potential for confounding. Both LBW and SGA share most of the risk factors for PTD, but small maternal stature and low pre-pregnancy maternal weight are associated with higher RRs (>1.5). We therefore classified three studies 5, 12, 26 as having higher potential for confounding because they failed to control for smoking and at least one of social class, maternal height and pre-pregnancy weight. Finally, the only known risk factors for pre-eclampsia and gestational hypertension that we considered to be both common and carrying a substantial RR were primiparity and maternal weight. One study took into account neither of these variables and was therefore classified as a poor quality study.5
The possible association between shift work and PTD was examined in 17 studies (seven of which were prospective cohorts in which exposure was ascertained before delivery). Most papers reported several risk estimates covering different time windows of exposure during pregnancy, so that there were a total of 24 time-specific-risk estimates.
Most of the studies,10,18,20,21,23-25,27-29,31 indicated elevated risks, but only two found significantly elevated RRs of 2 or higher.24, 27 One study gave a RR estimate >5, but was weakened by possible recall bias, both exposure and outcome being self-reported retrospectively.
Studies showing more than moderate adverse effects (RR>1.5) 10, 23, 24, 27 tended to be smaller than the negative ones, with the four largest studies finding RRs close to one 5, 19, 20, 30, suggesting that publication bias may have occurred with exclusion of small studies with negative results.
We pooled 16 studies in a random effect meta-analysis (see Figure 1), and we obtained a pooled RR estimate of 1.16 (95%CI= 1.00-1.33, test for heterogeneity p=0.006).
When we excluded five studies that did not meet our criteria for high quality,17, 26-29 the pooled estimated risk reduced and was no longer statistically significant (RR= 1.03, 95%CI 0.93-1.14, test for heterogeneity p= 0.023).
Ten studies analysed the possible association between shift work and the risk of delivering an SGA baby, including five prospective cohort investigations. Studies tended to rule out a more than moderate effect, with RRs ranging from 0.8 to 1.5, an exception being a recent investigation by Abeysena et al.12 in Sri Lanka which found an RR>3.0 for women exposed to the combination of shift work and other occupational hazards.
A pooled risk estimate (based on ten studies) was calculated as 1.12 (95%CI 1.03-1.22, test for heterogeneity p= 0.39) (Figure 2). When the one poor quality study was removed12 in a sensitivity analysis, the revised risk estimate was 1.10 (95%CI 1.00-1.20).
Risk of delivering an LBW baby was analysed in six studies. Among these, only one of the three cohort studies showed a significantly elevated risk. When all of the results were pooled in a meta-analysis, the combined risk estimate was 1.27 (95% CI 0.93-1.74, test for heterogeneity p= 0.39). Exclusion of the study by Saurel-Cubizolles and Kaminski,26 which did not present an RR estimate adjusted for important potential confounders, did not change this finding materially.
We found only two studies investigating pre-eclampsia and two investigating pregnancy-induced hypertension. None of them was prospective. No studies showed a significantly increased risk. Formal meta-analysis was not performed because of the small number of retrieved papers.
In our study we calculated pooled risk estimates for three clinically important pregnancy outcomes (PTD, SGA, and LBW) among women exposed to shift work, after systematically reviewing all available epidemiological studies. We found small elevations of risk (pooled RRs between 1.1 and 1.3).
Our risk estimate for PTD was based on a reasonable number of studies, but may have been exaggerated by publication bias, with more complete reporting of positive than non-positive findings from smaller studies. Moreover, when poorer quality studies were excluded, the pooled risk estimate did not differ significantly from unity. Regarding SGA and LBW, the number of studies identified was smaller, but results were more homogeneous across studies, and exclusion of poor quality studies did not materially change the pooled risk estimates. However, the pooled risk estimates were not sufficiently high to allow confident conclusions about a hazard, and spurious associations from bias or residual confounding cannot be ruled out. For pregnancy-induced hypertension and pre-eclampsia too few studies were available to allow any firm conclusion. Our findings are largely consistent with those of previous reviews 3, 9 and indicate that any effects of shift work on PTD, SGA and LBW are likely to be small.
Estimates of risk for PTD have tended to reduce with inclusion of more recent studies in meta-analyses: Mozurkewich3 in 2000 pooled risk estimates from six studies and found a 24% elevation of risk among exposed women; in our previous review9 (2007, based on 13 studies), we calculated a 20% increase in risk; and with addition of three further studies published over the past 5 years, the excess RR has fallen to 15%. If there is a hazard, then one explanation of this time trend could be that in most countries, precautionary legislation was introduced, allowing pregnant workers to be assigned to other tasks (without shift work) or to take earlier antenatal leave, so that fewer women continued shift work during the later stages of pregnancy. An occupational exposure such as night shifts may have been voluntarily suspended if women suspected that it could be detrimental to their pregnancy. In this situation, risks in women who continue to work shifts late in pregnancy might be underestimated because of a healthy pregnant worker effect,32 healthier women with uncomplicated pregnancies being less likely to modify their work schedules. In support of this, in two studies 20, 21 the risk of PTD associated with shift work was higher in women whose work conditions did not change in the course of pregnancy. Finally, in a study comparing European countries, significant associations of PTD with shift work were mainly observed in countries where long prenatal leaves were infrequent and legal support for preventive measures was weaker.30
An important limitation of the available epidemiological evidence is that many authors have not properly distinguished between different types of shift work. There is a possibility that rotating shifts (with or without work at night) may have varying level of job demand or could lead to a different degree of misalignment of maternal endogenous circadian rhythms compared with fixed night work. We did not calculate pooled estimates of risk for specific patterns of work-schedule because of the small number of studies available with sufficient information. However, Croteau et al20 presented risk estimates separately for night workers and for rotating shift workers (both compared with fixed day-time workers), and found that for both SGA and PTD risks were elevated only among rotating shift workers.
The mechanisms whereby shift work might result in adverse pregnancy outcomes are not entirely understood. Both direct (through disturbances of circadian rhythm) and indirect (through psychosocial stresses and sleep disruption) mechanisms have been proposed to explain a causal relationship between shift work and obstetric complications that are inherently multifactorial in nature. There is a sound basis for the notion that stressors may impact on PTD and LBW through several intersecting pathways, which include neuroendocrine, behavioral, immune, and vascular mechanisms.33, 34 Moreover, shift work modifies peak values and rhythm amplitudes of serum melatonin, whose function seems essential for successful pregnancy. 35 Recently, melatonin was found to act synergistically with oxytocin to increase membrane-bound phospholipase C activity and associated signaLling mechanisms, thereby enhancing myometrial contractility and gap junction-associated intercellular communication.36 Further investigation is needed to ascertain whether and to what extent disruption of sleep and circadian rhythms and daylight-night working cycles poses a significant threat to pregnant women and their fetuses.
On balance, the evidence currently available about the investigated birth outcomes does not make a compelling case for mandatory restrictions on shift-working in pregnancy. Further studies are needed to address the question whether adverse birth outcomes are related to different types of rotating work schedules (separating between night- and day-time shift), or to fixed night work. In the meantime, we suggest that, it would be prudent, insofar as job circumstances allow to permit pregnant women who wish to do so, to reduce their exposure to shift and night working.
MB presented an earlier version of this paper to the Scientific Symposium on the Health Effects of Shift Work, held in Toronto, Canada on April 12, 2010, when he was invited speaker on the topic shift work and pregnancy.
No funding was obtained for this research activity.
DISCLOSURE OF INTEREST
The authors declare that they have no competing interests.