METHODS

The data for this study were obtained from the Avon longitudinal study of parents and children (ALSPAC), a geographically based cohort study that has been described in more detail elsewhere.^18,19 All pregnant women resident within a geographically defined area of south west England with an expected date of delivery between April 1991 and December 1992 were eligible. Ethical approval of the study was obtained from the ALSPAC ethics committee, and the three local ethics committees covering the area.

Assessment of fish/n-3 consumption

Information on food consumption in pregnancy was obtained by a self completion food frequency questionnaire, sent out at 32 weeks’ gestation. Three questions inquired about fish consumption. These were “How many times nowadays do you eat”: (a) white fish (b) other fish (c) shellfish. Each question had five pre-defined response categories: Never or rarely, once in 2 weeks, 1–3 times per week, 4–7 times per week, more than once a day. These were converted to assumed weekly frequencies of consumption as follows: Rarely or never = 0, once in 2 weeks = 0.5, 1–3 times per week = 2, 4–7 times per week = 5.5, more than once a day = 10. The answers to these questions were used to estimate the n-3FA intake from fish. This was done by assuming a standard composition for each portion of white fish, oily fish, and shellfish, which would yield 0.32 g n-3FAs from a portion of white fish, 0.89 g from a portion of oily fish, and 0.34 g from a portion of shellfish.²⁰ The answers to the question on oily fish consumption were validated by comparison with the erythrocyte fatty acid composition of blood samples obtained during pregnancy—the docosahexaenoic acid content of the erythrocytes increased significantly with increasing frequency of consumption of oily fish (p<0.001).²¹

The women were divided into six groups according to their calculated weekly intake of n-3FAs (QUANT0 to QUANT5), with women in QUANT0 consuming no seafood and the other five groups being fifths of the remaining women. The estimated mean daily fish intake across all women was 32.8 g, corresponding to a mean daily n-3FA intake of 0.147 g. The mean daily fish intakes calculated for the six groups were: QUANT0–0 g, QUANT1–9.7 g, QUANT2–15.6 g, QUANT3–33.8 g, QUANT4–45.4 g, QUANT5–77.4 g. Table 2 shows the mean daily n-3FA intakes in each group.

The women were also divided into four groups on the basis of their combined frequency of consumption of white and oily fish. One group consisted of women who ate no fish (FREQ0), the other groups were as near as possible thirds of the remaining women (FREQ1-3). This second method of categorisation was used as it did not require most of the assumptions about portion size and n-3 content needed to calculate n-3FA intake. Only women who consumed no shellfish (81% of women) were included in analyses using this variable. This was because shellfish tends to be consumed in much smaller portions than other types of fish and the sociodemographic characteristics of women who ate shellfish differed from those of women eating other types of fish (data not shown). The mean weekly frequency of fish consumption for the four fish frequency groups were as follows: FREQ0–0 portions, FREQ1–0.74 portions, FREQ2–2.29 portions, FREQ3–4.44 portions.

Outcome variables

Birth weight, sex of the child, and gestation were abstracted from hospital records. Gestational age was assessed on the basis of date of last menstrual period. However, if there was a discrepancy with ultrasound assessment or other clinical indicators of two weeks or more, the clinical records were reviewed and a best estimate made by an experienced obstetrician. The factor given the greatest weight in deciding gestational age was the ultrasound assessment. Preterm birth was defined as delivery before 37 weeks’ gestation. Birth weight was obtained from hospital records. Low birth weight (LBW) was defined as a birth weight of less than 2500 g. Birth weight was adjusted for gestational age and sex (BWGA) using the residuals method. We defined IUGR as a BWGA that was below the 10th centile. This was done with reference to our own sample, not to an external standard—as we had such a large population based sample this was felt to be appropriate.

Confounding variables

Information on maternal smoking in pregnancy (at 32 weeks’ gestation), alcohol drinking in pregnancy (assessed during first three months), highest educational qualification, whether or not the mother was living with a partner, height, parity, and pre-pregnant weight was obtained by questionnaire. Parity was measured as the number of previous pregnancies that the mother had had that resulted in either a live birth or a stillbirth. Maternal age at delivery was derived by subtracting the mother’s date of birth from the child’s date of birth. These variables were classified as follows: smoking (0, 1–9, ⩾10 cigarettes per day), drinking (no alcohol, <1 glass per week, 1+glasses per week), maternal age at delivery (⩽19, 20–29, 30–39, ⩾40 years), parity (0, ⩾1), height (⩽159, 160–169, 170–179 and ⩾180 cm), pre-pregnant weight (⩽49, 50–59, 60–69, 70–79, and ⩾80 kg), education (low—CSE or no qualifications; medium—vocational qualifications or O level; high—A level or degree), and cohabitant status (that is, living with a partner, yes/no).

Statistical analyses

Only singleton liveborn babies were included in these analyses. Women who reported taking fish oil supplements at any point during pregnancy were excluded from the analyses.

Univariate analyses

Mean birth weight and gestation, and the frequency of LBW, preterm delivery and IUGR in the different n-3FA intake and fish consumption groups were compared using one way analysis of variance and the Kruskal-Wallis test for continuous variables, and χ² analysis and binary logistic regression for categorical variables.

Multivariate analyses

In the multivariate analyses, the ORs for LBW, preterm delivery, and IUGR in the different n-3FA and fish intake groups were calculated using logistic regression, adjusting for sex of the child and the confounding variables listed above. The adjusted ORs were also calculated separately for non-smokers only, as one study has suggested that the relation between n-3FA intake/status and birth outcome might be restricted to non-smokers.⁶

RESULTS

The 32 week questionnaire was returned by 12441 women out of 14 150 pregnancies reaching 32 weeks (87.9%), of whom 12 200 (86.2%) had completed the questions on fish intake. Excluding still births reduced the sample to 12 174 (86.0%), excluding multiple births reduced the sample to 11 851 (83.8%), and excluding women who took fish oil supplements reduced the sample to 11 585 (81.9%). Information on all confounding variables was available for 10 040 (71.0%) women—this formed the sample for the multivariate analyses.

Table 1 shows the frequency with which white fish, oily fish, and shellfish were consumed. White fish was eaten at least once in two weeks by 81.6% of women, as compared with 57.4% eating oily fish, and 19.3% eating shellfish.

Table 2 shows the relation of the confounding variables with n-3FA intake. The proportion of smokers, less educated mothers, primiparas, single women, short women, teenage mothers, and non-drinkers were all lower among women with higher n-3FA intake. The proportion of women with a low pre-pregnant weight also varied significantly with n-3FA intake, although the relation was not linear.

Table 3 shows the occurrence of LBW, preterm delivery, and IUGR by quantile of n-3FA intake, as well as the mean birth weight, gestation, and BWGA. As n-3FA intake increased there was a significant fall in the frequency of LBW and IUGR. There was also a small but significant increase in mean birth weight and mean BWGA, which increased by about 70–80 g between the children of mothers eating no fish or shellfish and those in the highest quantile of n-3FA intake. However, there was no association with the frequency of preterm birth or mean gestation. On adjusting for potential confounders the associations between n-3FA intake and mean birth weight, LBW, BWGA, and IUGR lost significance.

Similar results were obtained using grouped frequency of white and oily fish intake (table 4). Mean birth weight and mean BWGA increased and the frequency of IUGR declined as frequency of fish intake increased, and there was a small non-significant (p = 0.064) fall in the frequency of LBW. However, the frequency of preterm birth and mean length of gestation was unassociated with frequency of fish intake. On adjustment the association between mean birth weight or BWGA and frequency of fish intake lost significance. However, the reduction in the proportion of infants with IUGR with increasing fish intake remained significant in the adjusted analysis.

Table 5 shows the crude and adjusted ORs for preterm birth, LBW, and IUGR by quantile of n-3FA intake. There was no association between n-3FA intake and the likelihood of preterm birth, in either adjusted or unadjusted analyses. In unadjusted analysis there was a significant drop in the OR for LBW and IUGR with increasing n-3FA intake. On adjustment these relations were considerably attenuated and lost statistical significance, for example the OR of LBW in the lowest compared with the highest quantile of n-3FA intake fell from 1.33 to 1.08, and the equivalent figures for IUGR were 1.69 to 1.20. However, there was still some suggestion of a graded reduction in the ORs for IUGR with increasing n-3FA intake.

The logistic regression analysis was repeated using frequency of fish intake as the predictor (table 6). There was no significant association between frequency of fish intake and the occurrence of preterm birth and LBW in either univariate or multivariate analysis. However, there was a significant decrease in the OR for IUGR as frequency of fish intake increased, with ORs significantly greater than one in the FREQ0 and FREQ1 groups. This relation retained statistical significance but was considerably attenuated on adjusting for potential confounders, for example, the OR for IUGR in the lowest compared with the highest fish intake group fell from 1.85 to 1.37. The association between fish intake and IUGR was unchanged by the inclusion of ethnic group of the child as a confounding variable (5% of the cohort were of non-white ethnic origin).

The ORs for preterm delivery, LBW, and IUGR according to n-3FA intake or frequency of fish consumption among non-smokers only (tables 5 and 6) were similar to those obtained for all subjects. The effect of n-3FA/fish consumption was if anything more pronounced among the whole sample than among non-smokers only. These results did not suggest that any possible effect of n-3FA or fish intake on pregnancy outcome was restricted to or stronger among non-smokers than among smokers.

DISCUSSION

In this study we found no association between intake of n-3FAs or fish in late pregnancy and the duration of gestation or the frequency of pre-term birth in either crude or adjusted analyses. In univariate analysis fish/n-3FA consumption was positively associated with birth weight, with an increase of around 80 g between the lowest and highest n-3FA intake groups, and negatively associated with the frequency of IUGR. The relation with birth weight was lost on adjustment for potential confounders. The relation between n-3FA or fish intake and the frequency of IUGR was considerably attenuated on adjustment for potential confounders, however, the reduction in the odds of IUGR with increasing fish intake remained statistically significant. The fact that there was a significant association between fish intake and IUGR but not between BWGA and IUGR suggests that fish intake has an effect specifically on the right hand tail of the birth weight distribution, rather than shifting the whole distribution.

Overall it seems that most of the observational studies relating fish intake to pregnancy outcome, including our own, have found a moderate positive effect on fetal growth rate but little or no effect on gestation. In contrast, fish oil supplementation trials have tended to find a positive effect on gestation and little or no effect on fetal growth. Indeed, there has been some evidence from supplementation trials that n-3FAs might have a detrimental effect on fetal growth. There are several possible explanations for these differences. It may be that some constituent of fish other than n-3FAs is responsible for the association of fish intake with birth weight. Alternatively, there may be a threshold effect. In nearly all the supplementation studies the doses of n-3FAs given were considerably higher than would be consumed at even the highest levels of fish intake, and it is possible that effects on gestation are not observable at the lower intakes of n-3FAs achievable by eating fish. In several of the observational studies relating fish intake to birth weight there was a failure to control for socioeconomic status, which is associated with both birth outcome^22,23 and dietary intakes.²⁴

Limitations of this study

The study design was limited in its ability to detect a relation between fish intake and very preterm birth as the food frequency questionnaire was not sent out until 32 weeks’ gestation, thus any births before this time point were not included in the analysis. Conversely, ability to detect an effect of high fish intake on prolonging gestation might be affected by the current practice of inducing labour at 42 weeks’ gestation. Information on how labour started was available for a random sample of 1418 of the ALSPAC mothers. Labour was induced rather than spontaneous in 19% of all births, and in 22% of preterm births. It is possible that this may have obscured any associations between fish intake and preterm birth or mean duration of gestation.

The question on oily fish consumption did not distinguish between tuna and other types of oily fish—tuna has a much lower content of n-3FAs than other oily fish,²⁰ and in this respect is comparable to white fish. Including a separate question on tuna would have permitted more accurate calculation of probable n-3FA intakes, reducing measurement error and making any possible associations between n-3FA and birth outcome easier to observe. Another limitation of the study design is the limited ability to distinguish between levels of fish intake at the lowest end of the exposure scale—the lowest intake categories for each dietary variable were “never or rarely” and “once in 2 weeks”. In the recent Danish study⁹ the lowest categories for each dietary variable were “never”, “less than once a month” and “1–3 times a month” enabling the researchers to distinguish between low levels of fish intake. Furthermore, there is a possibility that some bias may have been introduced by missing data—of the 12 441 women who returned the questionnaire, only 10 040 had complete information on fish consumption and all confounding variables. It is possible that uncontrolled confounding might have obscured a true relation between fish intake and gestation, or created a spurious relation between fish intake and IUGR. One possible confounder for which we were unable to control was intrapartum weight gain, as these data were not available for most of our sample. Further observational studies of fish intake and pregnancy outcome in a range of populations are needed to help investigate this last possibility. This would allow the relation to be examined in populations with different levels of fish intake, and different patterns of association between fish consumption and other confounders.

The effect of seafood and n-3FAs on a range of other outcomes has recently received increased attention, for example the possible relations between n-3FA intake and eyesight development,²¹ mental health, and behavioural outcomes.^25,26 In the ALSPAC cohort it has been found that the children of mothers who ate oily fish during pregnancy were more likely to have achieved adult stereoacuity at the age of 3 years.²¹ Follow up of this and other cohort studies will show whether fish consumption in pregnancy has other long term effects on the fetus.