Overwhelming evidence demonstrates that even modest fish consumption (1–2 servings per week) has extensive health benefits, particularly for cardiovascular health. A recent systematic review reported that this relatively low level of fish consumption was associated with a 36% decreased risk of fatal coronary heart disease (CHD) and a 17% decreased risk of total mortality compared to <1 serving/week of fish consumption. Health benefits of modest but regular fish consumption occur within only months, but greater benefits occur with longer duration of intake. Oily fish (e.g., salmon, herring, sardines) may be especially beneficial, as they are relatively higher in two long-chain n-3 polyunsaturated fatty acids (n-3 PUFAs), eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA). EPA and DHA have been identified as the active constituents for cardiovascular benefits [1]. Furthermore, in many western societies it is likely that fish consumption displaces and thus reduces consumption of red and processed meat, and this reduction is associated with many health benefits [2‐10].

Despite these health benefits of fish intake, there is concern about the potential adverse effects of mercury from fish; fish is a major source of mercury [11], which, at extremely high levels, has been shown to result in serious health complications including cardiovascular disease [1]. Several studies investigating the relationship between fish intake, mercury exposure and cardiovascular disease in adults have generated inconsistent results [12‐19]. As a result, government agencies and health organizations have identified mercury exposure from fish consumption as an area needing further investigation [1, 20‐23]. The present study seeks to clarify whether current Mediterranean diet recommendations of fish consumption (1–2 servings of fish per week), elevate mercury exposure to levels that increase risk of cardiovascular disease.

Extensive evidence supports that toenail concentration of methylmercury is an excellent biomarker for long-term mercury exposure [24‐29]. A nested case-control study in the United States of 3427 cardiovascular disease cases and 3427 matched controls assessed mercury exposure by measuring toenail mercury concentration. They found no adverse effects of mercury exposure on coronary heart disease, stroke or total cardiovascular disease [11]. Another analysis using the same data found no adverse effects of mercury exposure on hypertension. This was true for levels 2.5-fold higher than the U.S. Environmental Protection Agency reference dose [30].

A 2012 review of mercury exposure and human health concluded that there is no clear relationship between mercury exposure and cardiovascular disease [31]. Thus, further evidence is needed before altering any policies or recommendations regarding mercury exposure through fish intake. We prospectively analyzed the associations between mercury exposure through fish intake and cardiovascular disease in PREDIMED, a large clinical trial of nutritional intervention for the primary prevention of cardiovascular disease. Our study population has unique characteristics for studying this association because our participants are at high risk of cardiovascular disease and they tend to consume large amounts of fish, and thus likely have higher levels of mercury exposure through fish consumption; per capita fish consumption in Spain is approximately twice the average of European countries [32], and fish consumption was even higher among PREDIMED participants randomized to the Mediterranean Diet.

Our results indicate that mercury exposure, as assessed by an objective biomarker, was not associated with increased cardiovascular disease among Spanish adults age 55–80 years despite their being at high risk for cardiovascular disease and having relatively high fish consumption. There was no significant association in the total population sample used for our nested case-control study, among the subset randomized to the Mediterranean diet, or among the subset randomized to the control (low-fat) group. There was also no association between toenail mercury levels and cardiovascular disease in our total population sample when we analyzed mercury exposure in conjunction with baseline fish intake or change in fish intake between baseline and one year of follow-up.

Null results from all sensitivity analyses demonstrate that our findings are robust. Lack of association after excluding those diagnosed with cardiovascular disease more than five years after toenail collection showed that potential misclassification of exposure due to variation over time did not attenuate results. We found no association after excluding participants with high toenail selenium levels, suggesting that the potentially protective effect of selenium against mercury toxicity did not influence our associations. Similarly, we found no associations after stratifying on low versus high fish intake or adjusting for fish intake, further supporting that our lack of association between mercury exposure and cardiovascular disease risk was not attributable to the potentially protective effect of fish consumption. The lack of association after restricting to participants at very high risk of cardiovascular disease suggests that these levels of mercury exposure from moderate fish consumption probably do not represent reason for concern regarding cardiovascular risk even for the highest-risk, most susceptible individuals. We found no association after stratifying on baseline aspirin use (yes/no), suggesting that the potentially protective effect of aspirin on cardiovascular disease [45] also does not influence this relationship.

Many previous studies have investigated the relationship between mercury exposure and cardiovascular disease [11‐18, 30]. The majority of findings are null, consistent with our own results. The few studies that observed positive associations had critical limitations [12, 13, 18, 19]. One case-control study that found a positive association with nonfatal myocardial infarction was retrospective and had higher rates of participation among cases. Thus, selection and recall bias may have affected their results [12]. A prospective study found a positive association with incident coronary disease, but with no dose-response relationship, and no association with overall cardiovascular mortality [13]. Two cross-sectional studies found positive associations – one with cardiovascular disease [18], and another with arterial stiffness [19]. Yet cross-sectional designs cannot establish causality, and both of these studies measured blood rather than toenail mercury concentration. Blood samples are not as well time-integrated as nail samples; red blood cells typically have a 17-week turnover, compared to 26–52 weeks for toenails [46]. However, these limitations do not entirely discount their credibility; policy and dietary recommendations regarding fish intake, mercury exposure and cardiovascular disease should be made in light of the entire body of evidence and should also take into account other non-cardiovascular harmful effects of mercury toxicity.

Our findings should not impact policies surrounding mercury pollution. The relationship between very high mercury exposure and increased risk of many health complications has been consistently demonstrated in other studies. The main application of our findings should be dietary recommendations for moderate to high fish consumption as part of ideal diets for cardiovascular disease prevention.

However, our findings do suggest that existing concerns for potential cardiovascular risk surrounding mercury exposure from fish consumption are probably unwarranted. Several aspects of our study population make these null results particularly meaningful. First, all participants are already at high risk of cardiovascular disease. Thus, any increased risk from mercury would be very likely to cause cardiovascular disease events. Second, our participants consume very high levels of fish. Per capita seafood consumption in Spain is approximately twice the average for European countries [32]. The intervention group is encouraged to consume even more fish than what is typical in Spain. The mean fish intake in our study population is approximately 92.7 g/day, compared to approximately 31.6 g/day [11], 43 g/day [13], 30.4 g/day [14], and 39.3 g/day [19] in previous studies. The observed lack of increased cardiovascular risk in this population provides evidence that consuming even the highest levels of fish consumption do not increase mercury exposure to dangerous levels. The observed null findings in this population provide evidence that mercury exposure from relatively high fish consumption (e.g. 3 servings/week) is unlikely to increase risk of cardiovascular disease. Third, our population has much higher levels of mercury (0.66 μg/g) compared to three out of four of the populations that observed a positive association between mercury exposure and cardiovascular disease (0.26 μg/g [12], 0.45 μg/g [18], and 0.30 μg/g [19]). This suggests that higher mercury exposure is not associated with increased risk of cardiovascular disease in the context of a healthy Mediterranean diet with high fish intake.

These null results are likely generalizable to populations with lower cardiovascular disease risk and lower fish consumption, since any harmful effects of mercury exposure would have even less impact among these people. Toenail mercury concentration is an objective, stable and time-integrated biomarker. Our prospective study design and our exposure assessment using an objective biomarker prevents recall bias.

In a recent report we showed the benefits of long-chain w-3 fatty acids from fish in the total data base of the trial.⁴⁷ The hazard ratios for intake of corresponding to the levels of long-chain n-3 polyunsaturated fatty acids recommended by the International Society for the Study of Fatty Acids and Lipids were 0.84 (95% CI 0.67–1.05) for all-cause mortality, 0.61 (95% CI 0.39–0.96) for fatal cardiovascular disease, 0.54 (95% CI 0.29–0.99) for fatal coronary heart disease, and 0.49 (95% CI 0.22–1.01) for sudden cardiac death. In addition, there is extensive available evidence supporting the cardiovascular benefits of fish consumption discussed above. However, it should not be surprising that in the participants included in this sub-study we didn’t find associations between fish intake and cardiovascular disease. Virtually all established dietary guidelines recommend at least 1–2 servings of fish per week (1 serving = approximately 75 g) for cardiovascular benefits. Furthermore, evidence suggests that higher levels of fish intake does not confer added benefits.¹ Our study population consumes an average of 95 g of fish per day. It is likely that we have too small a sample size and an insufficient range of exposure (i.e. average fish consumption is too high) to observe the cardiovascular benefits of fish consumption within this case-control substudy of mercury exposure.

There are limitations to our study that warrant continued investigation of this issue. We only have one measure of toenail mercury concentration, but toenail mercury concentrations have high reproducibility over time and low within-person to between-person variation ratio. This suggests that our single measure is a strong indicator of long-term intake [26]. However, true associations may still be attenuated. For example, it is possible that measurement error contributed to our null results. We did not have biological measurements for n-3 fatty acids. Although the FFQ has been validated in this population, fish vary in their n-3 fatty acid content [47], and the FFQ is still an imprecise measurement of n-3 fatty acids. Thus it is possible that imprecise measurement might have created residual negative confounding, as n-3 fatty acids would likely be positively associated with fish intake, and thus mercury levels, but inversely associated with cardiovascular disease. To address this, we conducted a bias analysis to assess the degree of plausible bias due to residual confounding by n-3 fatty acid intake (dichotomized at the median) based on our observed OR of 0.66 in Table 2, comparing the highest to lowest quartile of toenail mercury (Additional file 1: Table S7). We found that the odds ratio for the association between a dichotomous confounder (n-3 fatty acid) and cardiovascular disease would need to be as low as 0.2, and difference in prevalence of the confounder (high n-3 fatty acid intake in the exposed (highest quartile of mercury intake) compared to unexposed (lowest quartile of mercury intake) must be very extreme (e.g. 0.2 in exposed, 0.8 in unexposed) to render the association as detrimental. Acknowledging that toenail mercury concentration is also a biomarker for fish intake, and fish intake is associated with reduced cardiovascular disease risk, there could be negative residual confounding by fish intake or n-3 PUFA intake. This previously observed inverse association between fish intake and cardiovascular disease was not apparent in our data, probably because even in the lowest categories of fish consumption in our participants, the amount of fish consumed was already sufficiently high. While we believe it is unlikely that fish intake may strongly confound our results, because we already adjusted for fish intake, we acknowledge that residual confounding may yet exist and that our bias analysis results are subject to different interpretations. Based on previous evidence, we believe that it is unlikely, that the cardiovascular benefits of n-3 PUFAs are strong enough to account for a residual confounding of such magnitude as to explain the lack of association observed between mercury levels and cardiovascular disease. High n-3 PUFA consumption would need to be associated with approximately 0.11 the odds of cardiovascular disease compared to low n-3 PUFA consumption assuming a high association (OR = 5) between mercury and n-3 intake (Additional file 1: Table S7). A 2011 review included summarized 15 different meta-analysis RCTs and prospective cohort studies analyzing n3 PUFAs and various cardiovascular disease outcomes, including cardiovascular death, coronary heart disease, stroke and others. The lowest RR of these 15 meta-analyses was 0.62 (0.46–0.82) [48]. Despite this, we still manipulated the prevalence of the confounder to be 0.11 and higher (0.23, 0.43, 0.62, 0.81). Even at a confounder-disease OR of 0.43, which is likely still too low given that the higher RRs from the extensive meta-analyses [48], the corrected RR would be 0.90. Under the 0.23 confounder-disease OR, almost certainly too extreme, it would be 1.13. While we believe that these results indicate that residual confounding from n-3 PUFA is probably insufficient to explain our results, we admit that this interpretation may be subjective and that under some of the scenarios shown in Additional file 1: Table S7, the association mercury-CVD could be detrimental.

To minimize participant burden, our FFQ did not ask about consumption of specific types of fish. Thus, we could not obtain direct measurements of association between type of fish intake and toenail mercury concentrations. Seafood varies in mercury content [49], although mercury content of popular Spanish seafood have relatively consistent mercury levels [50]. Lastly, our sample size is small and many models are adjusted for many covariates, thus limiting our power, particularly when we adjust rather than match for sex and age in unconditional logistic regression models. However, it has been empirically demonstrated that the the rule of thumb usually suggested (adjusting for one confounder for every 10 events) can be relaxed [51]. However, the unique characteristics of our population described above, and the inverse association suggested by our point estimates, make it unlikely that we would have missed a material positive association between mercury exposure and cardiovascular disease. The role of other heavy metals in cardiovascular disease in this population merits further studies.

Mercury exposure from fish intake did not appear to increase cardiovascular disease risk in our sample of high risk participants exposed to high levels of fish consumption and consequently mercury concentrations, though under some assumptions on residual confounding by n-3 fatty acids, a slightly increased risk could persist. We did not observe any substantial risk in this unique setting of elderly participants with high cardiovascular disease risk, high fish intake and relatively high toenail mercury concentrations. These findings have important public health implications, as they suggest that the mercury content in fish is not very likely to discount the well-known cardiovascular benefits of fish consumption. Based on our findings, dietary recommendations should continue to include high fish intake as part of a healthy diet. Our findings, coupled with previous extensive evidence, support the dietary recommendation of at least 1–2 servings of fish per week for the prevention of cardiovascular disease.