Inconsistency can reflect the true causal pattern

Consistency is a condition that we accept as supporting causal interpretation in epidemiological studies. However, when two opposing forces, beneficial nutrients and harmful contaminants, coexist in the same food and, in particular, contaminants contained in fish really affect health in human, inconsistent human data on fish consumption can be expected for several reasons discussed below.

First, it is well known that spatial and temporal trends of contaminants in fish or marine mammals vary greatly.15–17 In this situation, different results about fish consumption would be observed among studies in different countries, or even in different areas in one country, or at different times in the same area. In the face of such inconsistency, meta-analysis or pooling analysis would only hide the real heterogeneity. Studies with bigger sample sizes cannot be a solution, either. Considering the accepted benefits of omega-3 fatty acids,1 2 failure to consistently observe beneficial effects of fish consumption may even be regarded as indirect evidence on harmfulness of contaminants in fish.

Besides heterogeneity of contaminants in space and time, the fact that many chemicals that accumulate in fish are endocrine disruptors18 can be another critical source of inconsistency among epidemiological studies. POPs are well-known endocrine disruptors18 and heavy metals are also hormonally active.19 Unlike the toxicological model in which effects are observed at high dose in a dose–response relation, endocrine disruptors may exert their hormonal effects at low dose and weaken or disappear above a certain level.20 21

It is generally regarded that high-exposure populations are better to explore health effects of fish consumption than low-exposure populations like the participants of Mendez et al's study.12 However, if chemicals in fish act as hormone disruptors at a low dose, the selection of a reference group with truly little exposure becomes critical for valid risk estimates. When the risk gradient is steep across low to moderate exposure levels and flat or decreasing across moderate to high exposure levels, low-exposure populations can give more valid risk estimates than high-exposure populations. Although the prevailing opinion has been that chemical exposure is safe if it fails to show any association in high-exposure populations, this conclusion may be incorrect for endocrine disruptors. Reports of non-monotonic dose–response relationships with strong low-dose effects for endocrine disruptors continue to challenge the basic assumptions of risk assessment.18

The study of Mendez et al12 may suffer from this fallacy since very few of the women studied had very low fish consumption. The authors classified the consumption of specific fish consumption into only two groups. Thus, subjects with high consumption (>1 servings/week) were compared with those with low to moderate consumption (≤1 serving/week). POPs contained in fatty fish may be involved in the increased risk of SGA through a low-dose effect, yet the high-exposure population and the coarse categorisation used in this study may not be proper to evaluate the effects of fatty fish consumption.

Study of fatty fish can be more problematic

Mendez et al12 could evaluate effects of different categories of fish due to a high fish intake in this cohort. This approach can be important because both kinds and levels of environmental contaminants vary across types of fish. Several recent studies have reported specific associations of fatty fish or crustaceans, but not of total fish consumption, with SGA.13 22 23 In fact, among various types of fish, fatty fish presents the greatest dilemma because of a high content of both beneficial omega-3 fatty acids and POPs. The primary hypothesis of Mendez et al12 was that fatty fish would be related to SGA due to these chemicals but their finding was that crustaceans and canned tuna, which are not fatty fish, were associated with SGA.

This finding may be interpreted that beneficial effects of fatty fish outweigh possible harmful effects in this cohort; however, the finding may also be related to an intrinsic limitation of the food frequency questionnaire used. Although there are no cooking methods that will reduce mercury levels in fish since mercury binds to proteins in muscle, levels of lipophilic chemicals such as POPs are influenced by various cooking methods.24 For example, removing the skin, fat and internal organs before cooking can reduce the exposure to these chemicals. If there is substantial variability of cooking methods, the estimated amount of fatty fish consumption based on frequency and portion size from the traditional food frequency questionnaire would not accurately reflect the amount of exposure to lipophilic chemicals due to fatty fish consumption, leading to substantial misclassification of exposure.

A lifetime approach is needed

Mendez et al measured several POPs in maternal serum.12 However, they did not directly present results on associations of POPs with SGA. The information on POPs was used only to see if POPs were involved in the associations between fish consumption and SGA through statistical adjustment. Since adjustment for POPs did not change the associations between fish consumption and SGA, they suggested that other mechanisms for high SGA, not POPs, must be considered.

However, one important drawback of most epidemiological studies of POPs is that, due to cost and required amount of blood sample, only a few selected POPs can be measured and considered in analyses. POPs include several hundred chemicals that may have different health effects. Although there are substantial correlations among serum concentrations of various POPs in the general population because of the simultaneous exposure to mixtures of POPs through food consumption, these correlations can weaken when within a narrow age range such as that of pregnant women. Although adjustment for polychlorinated biphenyls, hexachlorobenzene and p,p′-dichlorodiphenyldichloroethylene did not meaningfully change results in their study,12 it may be premature to conclude that mechanisms other than POPs explain the associations between some fish consumptions and SGA. POPs not measured in this study, or mixtures of POPs, may explain the associations of fish consumptions with SGA.

Despite general positive associations between serum concentrations of POPs or MeHg and the amount of fish consumption in pregnant women,25 26 fish consumption during pregnancy can explain only a small part of the variation of serum concentrations of POPs.27 POPs build up over the lifetime. The fetus is affected by the total body burden of POPs in pregnant mothers, which is determined by various factors, including a lifetime diet history. One study among pregnant women showed that serum concentrations of POPs were more strongly associated with fish consumption during adolescence and duration of breast feeding during infancy rather than the fish consumption during pregnancy.27 POP levels are unlikely to reduce significantly during temporary diet changes during pregnancy because of their highly lipophilic and persistent nature. Under this situation, the usefulness of epidemiological studies that focus on the amount of fish consumption during pregnancy may be substantially limited and cannot give sensible evidence on safety or harmfulness of fish consumption. Therefore, direct measurement of POPs in pregnant women may be necessary. Also, if POPs are confirmed as harmful, any recommendation on diet should be done from a lifetime viewpoint.