Methods

Our study used data from the Southampton Women’s Survey (SWS), a longitudinal study that aimed to recruit all women aged 20–34 years, who were resident in the city of Southampton, England during 1998–2002.13 At entry to the SWS, each of the 12 583 participants was interviewed and examined by a research nurse, and those who subsequently became pregnant were followed up with further interviews at 11 and 34 weeks of gestation. Details of the birth were then abstracted from obstetric records, and anthropometric measurements were made on the new-born babies.

The analysis presented in this paper focused on singleton pregnancies leading to a live birth between 1 June 1999 and 31 December 2003, after more than 34 weeks of gestation, in Caucasian women who did not have diabetes or pre-eclampsia, and who undertook paid employment at some time during pregnancy. Where a woman had more than one baby during this period, only the first eligible pregnancy was included.

From the interview and examination at entry to the SWS, we obtained information about the mother’s date of birth, alcohol consumption, smoking habits, education, parity, height and weight. Alcohol consumption was classified into two bands (<14 and ⩾14 units per week), and education was graded in six levels according to the highest academic qualification obtained. Height was measured to the nearest millimetre with a stadiometer, and weight to the nearest 0.1 kg with an electronic scales (after removal of shoes and heavy clothing). These measurements were used to calculate body mass index (BMI).

Occupational activities during pregnancy were ascertained through the interview conducted at 34 weeks of gestation. Women were asked whether they were in paid work at each of 11, 19 and 34 weeks of gestation; and where they had held jobs at one or more of these times, they were asked how many hours per week they worked, whether the work included night shifts, and whether or not an average day at work involved each of four physical activities (standing or walking for more than 4 h in total, kneeling or squatting for more than an hour in total, standing or sitting with the trunk bent forward (illustrated in a diagram) for more than an hour in total, and lifting or carrying weights of 25 kg or more by hand).

Data abstracted from hospital records at birth included the baby’s sex and birth weight. We collected detailed menstrual and ultrasound data in early pregnancy13 and set the estimated date of conception using an algorithm that combined menstrual and ultrasound data. Deliveries were classed as preterm if they occurred before 260 days (37 weeks) of gestation. Birth weights were measured with digital scales, and babies were classed as SGA if, after account was taken of gestational age and sex, the weight was in the lowest 10% of the standard British distribution in 1990.14

In addition to the abstraction of these routinely recorded clinical data, a research nurse from the study team measured the baby’s head circumference and upper abdominal circumference (at the level of the xiphisternum) within 48 h of birth. These measurements were made to the nearest millimetre with a tape measure, using a standardised technique. Head circumference was assessed against the standard British distribution in 1990, according to gestational age and sex.14 For abdominal circumference there were no satisfactory external reference values, and the expected sex-specific distribution by gestational age was therefore derived by applying random-effects regression to internal SWS data using the method of Royston.15 Small head and abdominal circumference were defined as being below the 10th centile of the reference population.

Statistical analysis was carried out with Stata 10.0 software. Risk factors for each of four adverse outcomes of pregnancy (preterm delivery, SGA, small head circumference and small abdominal circumference) were examined by logistic regression, and associations were summarised by odds ratios (ORs) and associated 95% confidence intervals (CIs). We first explored associations with non-occupational risk factors, using a single regression model for each adverse outcome. Variables that were associated with an outcome (p<0.2) were then carried forward as factors of adjustment in subsequent analyses of occupational risk factors.

Having assessed the risk of each outcome according to whether or not the mother was in any form of work at 34 weeks of gestation, we examined associations with exposure to specific occupational activities at three different points in pregnancy (11, 19 and 34 weeks). In these analyses, we compared women who were exposed to an activity with those who were in work at the same stage of pregnancy but unexposed to the activity. Each activity was analysed independently with adjustment for potential non-occupational confounders.

Finally, to address the possibility that some women may have ceased one or more occupational activities because of early complications of pregnancy, we looked for evidence of healthy worker selection by comparing mothers who carried out an occupational activity at 34 weeks of gestation with mothers who were in work at 34 weeks of gestation but had not been exposed to the activity at any stage of pregnancy.

Results

A total of 1327 pregnancies met the criteria for inclusion in our analysis (table 1). The ages of the mothers ranged from 21 to 38 years (mean 30 years), and 54% were primiparous. Almost all were in paid work at 11 and 19 weeks of gestation, and 797 (60%) were working at 34 weeks of gestation. Preterm delivery occurred in 46 (3%) pregnancies, and 89 (7%) babies were SGA.

Table 2 shows the relationship of birth outcomes to the non-occupational risk factors examined. The strongest associations were for primiparity with SGA (OR 2.62, 95% CI 1.55 to 4.42) and small abdominal circumference (OR 2.22, 95% CI 1.46 to 3.37). In addition, weaker associations (but with p<0.2) were observed for maternal age, pre-pregnancy BMI and smoking with SGA; primiparity and maternal educational level with small head circumference; and maternal age and smoking with small abdominal circumference. These variables were therefore carried forward as factors of adjustment when examining occupational risk factors.

In comparison with mothers who had stopped working before 34 weeks of gestation, those who remained in employment showed no statistically significant differences in the risk of adverse birth outcomes. After adjustment for potential confounding factors, the odds ratios were 0.76 (95% CI 0.41 to 1.37) for preterm delivery, 1.25 (95% CI 0.79 to 1.99) for SGA, 1.04 (95% CI 0.72 to 1.51) for small head circumference, and 1.21 (95% CI 0.82 to 1.77) for small abdominal circumference.

Risk of preterm delivery was elevated in women whose work entailed sitting or standing with the trunk bent forward for more than an hour per day (table 3), and this association was particularly strong when the activity occurred at 34 weeks of gestation (OR 2.92, 95% CI 1.27 to 6.70). Other occupational activities showed no clear relation to preterm delivery.

Tables 4–6 summarise the relationships of occupational activities to SGA, small head circumference and small abdominal circumference. No statistically significant associations were seen with SGA or small abdominal circumference, all ORs being less than 1.4. However, small head circumference was more common in babies born to women who worked for >40 h per week (ORs 1.51 to 1.72). In addition, there was a suggestion of an increased risk of small head circumference in relation to standing or walking for >4 h per day during the first two trimesters (ORs 1.36 to 1.40), and in relation to lifting weights of >25 kg by hand, and especially in the last trimester (ORs 1.64 to 2.52). However, these findings did not reach statistical significance.

To explore possible healthy worker effects, we repeated the analyses for women who were in work at 34 weeks of gestation, comparing those who carried out an activity late in pregnancy with those who had not been exposed to the activity at any of 11, 19 or 34 weeks of gestation.

This analysis confirmed the association between preterm delivery and sitting or standing with the trunk bent forward (OR 2.66, 95% CI 1.15 to 12.6). A relationship between small head circumference and working more than 40 h a week was again observed but did not quite reach statistical significance at the 5% level (OR 1.71, 95% CI 0.94 to 3.11).

Discussion

Our findings suggest an increased risk of preterm delivery in women whose work entails trunk bending for >1 h per day, especially after 34 weeks of gestation, and of low head circumference in women working >40 h per week during pregnancy. However, in keeping with our earlier review, preterm delivery showed little association with prolonged working hours, lifting or standing, and the RR for shift work was similar to the previous pooled estimate of 1.20 to 1.26.2

The present study had the advantage of a prospective design, with exposures assessed pre-delivery and outcomes ascertained objectively, with reasonable precision of measurement and blinding to occupational history. Relevant personal risk factors that could confound associations (such as smoking and maternal BMI) were ascertained before pregnancy and controlled for where relevant in subsequent analysis. Risk estimates were also presented separately for exposures in each trimester of pregnancy, whereas in our earlier review we rarely found reporting at this level of detail.16 17 18 19 And finally, we studied some clinically important outcomes for which there is almost no occupational risk information (head and abdominal circumference), as well as at least one occupational exposure for which research on adverse pregnancy outcomes appears very sparse (trunk bending).

One limitation of our investigation was the reliance on self-reported exposures, with the possibility of either differential or non-differential measurement error. The former appears less likely, however, as exposures were ascertained before outcomes. The latter may be greater for exposures that are difficult to self-estimate, such as weights lifted or time spent standing, than for those that are simpler to judge (eg, working hours, night shift working). However, non-differential measurement error is unlikely to explain the positive associations that we observed.

A second limitation was our capacity to assess exposures only by questions at the 34-week interview, without tracking job changes and their reasons. Healthy worker selection bias could arise if women with complications of pregnancy dropped out of work altogether, or changed the work they did, to avoid certain activities. The likely impact would be to obscure risks of work later in pregnancy, and once again such bias to the null would not explain the positive associations of preterm delivery with truck bending. Similarly, the elevated risk of small head circumference, which was apparent in relation to prolonged working hours in each of the trimesters, requires an alternative explanation.

Of more concern in the context of healthy worker selection bias is the interpretation of null findings, such as the lack of association between standing and preterm delivery. In the main analyses we compared women who were exposed to a given activity with those who were in work at the same stage of pregnancy but not exposed to that activity. To explore the impact of selection effects, we also conducted a sensitivity analysis comparing mothers who carried out an occupational activity at 34 weeks of gestation with mothers who were in work at that stage but had not been exposed to the activity at any stage of pregnancy. The associations between trunk bending and preterm delivery and between working hours and small head circumference were little changed, suggesting that healthy worker selection did not have a strong influence on the findings. In principle, our comparison group (women in work at the same stage of pregnancy but unexposed to a given occupational activity) could have been exposed to other occupational activities that confer risk. If so, then the risk of preterm delivery with trunk bending in late pregnancy would be underestimated. In practice, however, any bias is likely to be small given evidence from our previous systematic review2 and from the current analysis that the exposures in question are relatively weak risk factors.

A third limitation of our study concerns the power to detect associations in a relatively healthy cohort (outcome prevalences 3–10%) among whom exposures such as night shift work and lifting were reported by ⩽10% of women. This meant that for some of the exposures investigated, 95% CIs were relatively wide. Thus, for example, although night shift work at 11 weeks of gestation was associated with a 14% increase in risk of preterm delivery (five exposed cases), the 95% confidence limits were compatible with a 57% protective effect or a near tripling of risks. Other exposures, such as standing or walking and trunk bending, were more common (24–45%), however, allowing risks to be estimated with greater precision.

The stepwise selection strategy used in our analysis, which was based upon statistical significance testing, could in theory discard factors showing strong but imprecise associations, but no factor with these characteristics was omitted (table 2) and in practice, such a factor would most probably have a low prevalence and a limited potential to confound.

Finally, it should be noted that our entry criterion for analysis, which required interview data at 34 weeks of gestation, precluded exploration of the risks of early fetal loss, as well as reducing the rate of preterm deliveries in our study population.

Notwithstanding these limitations, our findings on occupational activity and SGA, and on working hours, standing, lifting, shift work and preterm delivery, are broadly consistent with earlier research. Previously, in relation to SGA, we described seven studies on working hours, five with RRs close to unity, and eight reports on standing, seven with RRs ⩽1.4; and in relation to preterm delivery, 11 of 12 studies on lifting carried RRs <1.4, while the pooled estimates of RR for prolonged working hours, standing and shift work ranged from 1.2 to 1.3.2

In this study we found that trunk bending in late pregnancy (in occupations such as nursing, work with small children (child minders, nannies, teachers), catering, cleaning and shop work) carried a nearly threefold elevation in risk of preterm delivery. Although linked in a few reports with a greater risk of spontaneous abortion,20 21 22 we know of no other study of trunk bending and preterm delivery.

This omission seems important to rectify. Raised intra-abdominal pressure, consequent on trunk bending, is a biologically plausible risk factor for preterm labour, especially in the last trimester when space within the abdominal cavity is maximally constrained. Moreover, obesity, a non-occupational cause of raised intra-abdominal pressure, has also been linked with preterm birth, albeit inconsistently in a small literature.23 Hence, our observations on trunk bending in late pregnancy require confirmation in other studies.

Additionally, after adjustment for educational attainment and primiparity, we found associations of prolonged working hours with small head circumference. We are aware of only one other occupational study concerning this outcome, in low-income mothers from Boston.24 No significant association was found between standing at work and small head circumference at birth although the odds ratios were elevated and close to statistical significance for exposures in the first two trimesters. Contrary to our finding, an inverse relationship between maternal physical activity (not specifically work activity) in the first two trimesters and small head circumference has been reported in villagers from rural India.16 Such inconsistencies and the paucity of reports underscore the need to interpret our data with caution. If true, however, our observations could be clinically relevant, as small head circumference at birth has been linked with a doubling of risks of prevalent coronary heart disease in middle age,10 a significantly higher cardiovascular mortality among working-aged men,11 and a significantly higher adult systolic blood pressure.7 Moreover, in a survey of 50-year olds from Preston, UK, those with impaired glucose tolerance or non-insulin dependent diabetes mellitus had a significantly smaller head circumference at birth.9 In most of these studies, definitions of restricted head circumference were less stringent in percentile terms than our own.

The mechanism by which prolonged working hours and other occupational activities might affect birth anthropometrics is unknown, and in the absence of strong prior expectations our findings on head circumference should be viewed only as hypothesis-generating at this stage. Further research is warranted. However, a higher priority is the need to assess further the risks of preterm labour in mothers who bend their trunk late in pregnancy. Our other findings, on preterm delivery and on occupational physical activities and SGA, add to a growing body of evidence that is broadly reassuring to pregnant workers.