A Quantitative Risk–Benefit Analysis of Changes in Population Fish Consumption

Article Outline

• Abstract
• Introduction
• Methodology
• Relationship Among Fish Consumption, n-3 PUFA Intake, and MeHg Exposure
  • Measures to Modify Fish Consumption Among Women of Childbearing Age
  • Measures to Increase Fish Consumption
  • Summary of Dose–Response Relationships
  • Quantification of Health Impacts Using a Common Metric
  • Characterization of Uncertainty
• Results
• Discussion
• Acknowledgment
• References
• Copyright

Abstract

Although a rich source of n-3 polyunsaturated fatty acids (PUFAs) that may confer multiple health benefits, some fish contain methyl mercury (MeHg), which may harm the developing fetus. U.S. government recommendations for women of childbearing age are to modify consumption of high-MeHg fish, while recommendations encourage fish consumption among the general population because of nutritional benefits. To investigate the aggregate impacts of hypothetical shifts in fish consumption, the Harvard Center for Risk Analysis convened an expert panel (see acknowledgments). Effects investigated include prenatal cognitive development, coronary heart disease mortality, and stroke. Substitution of fish with high MeHg concentrations with fish containing less MeHg among women of childbearing age yields substantial developmental benefits and few negative impacts. However, if women instead decrease fish consumption, countervailing risks substantially reduce net benefits. If other adults (mistakenly and inappropriately) also reduce their fish consumption, the net public health impact is negative. Although high compliance with recommended fish consumption patterns can improve public health, unintended shifts in consumption can lead to public health losses. Risk managers should investigate and carefully consider how populations will respond to interventions, how those responses will influence nutrient intake and contaminant exposure, and how these changes will affect aggregate public health.

Back to Article Outline

Introduction 

Fish are nutrient-dense food, with many types of fish containing high concentrations of the n-3 polyunsaturated fatty acids (PUFAs), eicosapentaenoic acid (EPA), and docosahexaenoic (DHA). These n-3 PUFAs may protect against several adverse health effects, including coronary heart disease (CHD) mortality and stroke. DHA may also be critical to the developing brain in utero. Humans are unable to synthesize n-3 PUFAs and must therefore obtain this nutrient from their diet.

Although fish are a rich source of n-3 PUFAs, some may also contain methyl mercury (MeHg), a neurotoxin that may adversely affect the development of the brain and nervous system. In order to protect against in utero exposure to MeHg, the U.S. Food and Drug Administration (FDA) and the U.S. Environmental Protection Agency (U.S.EPA) issued a joint advisory in March 2004 recommending that “[p]regnant women, women who might become pregnant, young children, and nursing mothers” modify their fish consumption.¹

Because fish consumption confers both benefits and risks, the advisories issued by the U.S. federal government raise the possibility of a classic risk–risk trade-off²: by avoiding one risk (exposure to MeHg), consumers who follow these advisories may be incurring another (adverse health consequences associated with lower n-3 PUFA intake). Likewise, individuals who increase their consumption of fish because of this food's nutritional benefits may incur risks associated with MeHg exposure.

To help quantify this trade-off, the Harvard Center for Risk Analysis convened an expert panel (see acknowledgments), the work of which is described in five papers. This paper, which synthesizes the results of the other four,³, ⁴, ⁵, ⁶ quantitatively evaluates the risk–risk trade-off just described by considering the potential public health impact of advisories on fish consumption patterns. The advisories considered include those aiming to reduce fish consumption among pregnant women and others that might encourage additional fish consumption among other members of the population (e.g., middle-aged and older individuals who are at increased risk for CHD and stroke). This paper also considers the possibility that the advisories influence fish consumption in a manner not intended by the advisory (e.g., changing consumption among individuals not belonging to the group targeted by the advisory).

Health effects considered in this analysis include fish consumption and both CHD mortality and stroke, and the influence of either DHA intake or MeHg exposure on prenatal cognitive development. The analysis considers only the health impacts, measured in terms of a common metric that facilitates quantitative comparison. This metric, the quality-adjusted life year (QALY)⁷ reflects life expectancy, weighted by quality of life. This study does not take into account resource (monetary) costs associated with these health effects.

An earlier analysis by Ponce et al.⁸ also considered the trade-off addressed in this paper. However, Ponce et al.⁸ did not consider the beneficial impact of fish consumption on cognitive development following prenatal intake, or on stroke risk in adults. Their consideration of the literature on fish consumption and CHD risk was limited to a single study,⁹ and their assessment of the impact of MeHg on cognitive development was limited to consideration of age at first speech. This paper relies on the more comprehensive set of analyses of the health effects associated with fish consumption that also appear in this issue.

Note that this analysis does not exhaustively address all effects associated with shifts in fish consumption patterns. For example, it omits consideration of other contaminants, including various organochlorines, that may bioaccumulate in fish.¹⁰ However, as explained in the discussion portion of this paper, these particular contaminants, which have been prominent in the news,¹¹ are unlikely to be important compared to the effects that this analysis does consider. Various studies have considered a range of other health effects, including increased risk of carcinogenicity, immunotoxicity, and renal toxicity in the case of MeHg exposure,¹⁰ and protection against asthma, type 2 diabetes, and rheumatoid arthritis in the case of n-3 PUFA intake.¹², ¹³ This paper limits attention to those effects that are most likely to contribute substantially to the aggregate health impacts associated with changes in fish consumption and for which there is adequate evidence to conduct a quantitative analysis. Of course, other effects could be incorporated into the framework presented here if the evidence warrants doing so at a later date.

Nor does this analysis consider the nutritional impacts of other sources of energy and protein that may be consumed if fish intake rates drop (e.g., carbohydrates and other types of meat). For example, an increase in the consumption of foods higher in saturated fat may increase cardiovascular disease risks.¹⁴, ¹⁵ These foods may also contain other contaminants.¹⁶ Finally, this analysis does not take into account the ecologic impacts associated with more intensive harvesting of natural or farmed fish stocks¹⁷ that might be necessitated by increases in fish consumption.

Back to Article Outline

Methodology 

The impact of fish consumption on health can be modeled as the product of exposure to the active agent (micrograms per day of MeHg, grams per day of n-3 PUFA, or servings per week of fish), and the dose–response relationship for that active agent (incremental risk per microgram per day of MeHg, or reduction in risk per gram per day of n-3 PUFA, or servings per week of fish). The following section of this paper quantifies the impact of changes in fish consumption on MeHg exposure, n-3 PUFA intake, and the average number of fish servings consumed per week. The four papers³, ⁴, ⁵, ⁶ that also appear in this issue characterize dose–response relationships for the health effects considered (MeHg exposure and cognitive development, DHA intake and cognitive development, fish consumption and stroke risk, and fish consumption and CHD mortality risk). This paper summarizes those findings, and in order to make the disparate health impacts comparable, expresses them in terms of a common metric, known as the QALY.⁷ Finally, the characterization of uncertainty is described.

Back to Article Outline

Relationship Among Fish Consumption, n-3 PUFA Intake, and MeHg Exposure 

This analysis focuses on how changes in fish consumption influence typical n-3 PUFA intake rates and MeHg exposures in the U.S. population. It does not describe what may happen to specific populations (e.g., individuals whose consumption is dominated by locally available fish that may be contaminated by emissions from distant or nearby sources).

To estimate the impact of changes in fish consumption on MeHg exposure, n-3 PUFA intake, and the average number of fish servings consumed per week, this analysis uses a fish consumption model developed by Carrington and Bolger¹⁸ that they later modified.¹⁹ Briefly, the authors estimated consumption rates for 42 types of fish based on data from the U.S. Department of Agriculture's (USDA) Continuing Survey of Food Intake by Individuals (CSFII)²⁰ and the Third National Health and Nutrition Evaluation Survey. They developed estimates of the MeHg concentrations from FDA surveillance data and from information published by the National Marine Fisheries Survey.²¹

Carrington and Bolger¹⁸ implemented their exposure model in Visual Basic running under Microsoft Excel. They kindly provided us a copy of their most recent model (March 2004). Using data from the USDA Nutrient Data Laboratory (www.nal.usda.gov/fnic/foodcomp/srch/search.htm), this paper modifies the Carrington and Bolger model so that it can also predict n-3 PUFA intake for the U.S. population. The modified model reports intake for the two key n-3 PUFAs in fish—EPA (20:5n-3) and DHA (22:6n-3). Estimates of n-3 PUFA concentrations were not available for six types of fish included in the Carrington and Bolger model. This analysis effectively assumes that n-3 PUFA concentrations are zero for these species, which represent only 2.6% of total fish consumption. Because the Carrington and Bolger model omits individuals who eat no fish (which they estimate to represent between 10% and 20% of the population), this analysis assumes that modeled changes in fish consumption affect exposures among 85% of the population. Among the remaining 15% of the population, this analysis assumes that fish consumption remains unchanged (zero).

In characterizing changes in fish consumption, our purpose is not to focus on the means of influencing fish consumption (e.g., via taxes, warning labels, fishing bans, and so forth). Rather, this analysis considers a representative range of reactions among different segments of the population in response to two categories of risk management measures. In the first category are three hypothetical scenarios representing potential population responses to measures (e.g., an advisory) that encourage women of childbearing age to moderate consumption of fish containing relatively high MeHg concentrations. In the second category are two hypothetical scenarios representing potential population responses to measures (e.g., consumer education) that encourage other adults to consume more fish.

Measures to Modify Fish Consumption Among Women of Childbearing Age 

Scenario 1 represents the “ideal” (optimistic) response to such measures. It assumes that women of childbearing age eliminate consumption of fish containing “high” or “medium” MeHg concentrations, but that they maintain the same overall level of fish consumption by instead consuming fish with “low” MeHg concentrations. Here, the analysis uses the MeHg fish categories designated by Carrington et al.¹⁹: “high” means MeHg concentrations >0.6 μg/g, “medium” means MeHg concentrations ranging from 0.14 to 0.6 μg/g, and “low” means MeHg concentrations ≤0.13 μg/g). Scenario 1 is considered optimistic because it allows for the possibility that women will reduce their MeHg exposure while maintaining comparable n-3 PUFA intake rates.

Scenario 2 is somewhat less optimistic, assuming that all women of childbearing age decrease their total fish consumption by 17%. While doing so will decrease MeHg exposure, it will also decrease n-3 PUFA intake. The 17% change is based on an Oken et al.²² study of fish consumption patterns among pregnant women before and after the FDA's 2001 fish advisory was issued. Oken et al.²² reported that consumption of canned tuna, dark meat fish, shellfish, and white meat fish averaged 7.7 servings per month before the advisory, and 6.4 servings per month afterward. Scenario 2 assumes that the 17% reduction in fish consumption occurs across all species equally. Although the subjects in the Oken et al.²² study may have preferentially reduced consumption of fish with high MeHg concentrations, this possibility cannot be assessed from the information available. It must also be recognized that the 17% reduction is illustrative only of potential changes in consumption.

Scenario 3 is the most pessimistic of the scenarios involving the measures to influence fish consumption among women of childbearing age. This scenario assumes not only that the target population reduces their fish consumption by 17%, but also that other members of the population reduce their fish consumption by the same amount.

Measures to Increase Fish Consumption 

Scenario 4 represents the “ideal” (optimistic) response to a measure aiming to increase fish consumption among other adults. In particular, this scenario assumes that all females not of childbearing age and all males increase their fish consumption by 50%. Scenario 5 represents an alternative case in which women of childbearing age also increase their fish consumption despite the fact that they are not targeted by the measure.

Note that the analysis re-expresses MeHg intake (micrograms per day) in terms of its impact on the concentration of MeHg in maternal hair at childbirth (micrograms of total Hg per gram of maternal hair, which is approximately equal to micrograms of MeHg per gram of maternal hair²³), the units used in the dose–response relationship for MeHg exposure and IQ. Here, the analysis assumes that maternal body weight is 60 kg, and uses the relationships outlined by the National Research Council.¹⁰ That report estimated that 1 μg/kg/day of MeHg intake in a 60-kg adult female would result in a total Hg concentration in hair of 10 μg/g. Hence, 1 μg/day of MeHg intake (0.017 μg/kg/day) corresponds to approximately 0.17 μg/g of total Hg in maternal hair.

Summary of Dose–Response Relationships 

Table 1 summarizes the findings from the four dose–response relationship papers³, ⁴, ⁴, ⁵, ⁶ that accompany this risk–risk trade-off analysis. For each relationship, the table lists the central estimate and characterizes uncertainty. Note that two of the dose–response relationships specify both a “threshold” effect component and an incremental effect component. The threshold component represents protection conferred on individuals who consume some fish compared to those who consume no fish. The incremental component represents additional protection against risk resulting from further increases in fish consumption.

Table 1. Summary of dose–response relationships

Health effect	Relationship	Central estimate	Uncertainty
Fish consumption and CHD mortality	ΔRR for some fish consumption vs no fish consumption (<1 serving/month)	−17%	95% CIa: −8.8% to −25%
	ΔRR per additional serving/week	−3.9%	95% CIa: −1.1% to −6.6%
Fish consumption and stroke incidence	ΔRR for some fish consumption vs no fish consumption (<1 serving/month)	−12%	95% CIa: +1.0% to −25%
	ΔRR per additional serving/week	−2.0%	95% CIa: +2.7% to −6.6%
MeHg exposure and cognitive development	ΔIQ per μg/g total Hg in maternal hair	−0.7pts	Bounds: 0 to 1.5 pts
DHA intake and cognitive development	ΔIQ per g/day maternal intake of DHA	1.3pts	Bounds: 0.8 to 1.8 pts

CHD, coronary heart disease; CI, confidence interval; DHA, docosahexaenoic acid; MeHg, methyl mercury; pts, points; RR, relative risk.

Realistically, an increase or decrease in a population's average fish consumption probably reflects in part changes in the number of individuals who consume at least some fish. However, because the stylized scenarios outlined above envision percentage changes in fish consumption, they (unrealistically) do not lead to a predicted change in the number of individuals consuming any fish. In order to approximate this phenomenon within the framework of the scenarios defined here, the analysis assumes that individuals who consume less than one serving of fish per month (100 g of fish per month) (chapter 10 in U.S.EPA²⁴) effectively consume no fish for the purpose of characterizing their health risks. Many of the epidemiologic studies used to develop the dose–response relationships for stroke and CHD mortality had comparison groups with fish consumption rates defined in a similar manner. These groups effectively represent the “no-fish-consumption” population in our analysis.

Finally, the analysis assumes that there is no delay between a change in consumption and a change in risk. For example, if an individual increases his fish consumption for 1 year, the analysis assumes that during that year, relative risk for CHD mortality decreases by an amount that depends on the dose–response relationship and the individual's baseline risk. For the purpose of evaluating trade-offs, the analysis assumes the change in consumption, and hence the attendant effects persist indefinitely. Effectively, the analysis reports the change in the rate at which health benefits are accrued (i.e., net gain or net loss annually).

Quantification of Health Impacts Using a Common Metric 

So that disparate health effects can be aggregated across the population, this analysis re-expresses them in terms of a common metric, the QALY.⁷ Details of this conversion appear in the Technical Appendix, Technical Appendix, Technical Appendix, Technical Appendix, Technical Appendix, Technical Appendix (found online at www.ajpm-online.net). The QALY is a measure of health damages that takes into account changes in longevity and quality of life. Total QALYs accrued by an individual amount to the number of life-years lived, with each life-year assigned a weight between 0 (health condition equivalent to death) and 1.0 (perfect health). The health impact associated with any scenario is quantified in terms of the change in QALYs accrued in the population relative to the baseline scenario.

• 
  • View Large Image
  • Download to PowerPoint

• Technical Appendix.

• 
  • View Large Image
  • Download to PowerPoint

• Technical Appendix.

• 
  • View Large Image
  • Download to PowerPoint

• Technical Appendix.

• 
  • View Large Image
  • Download to PowerPoint

• Technical Appendix.

• 
  • View Large Image
  • Download to PowerPoint

• Technical Appendix.

• 
  • View Large Image
  • Download to PowerPoint

• Technical Appendix.

Many health effects have long-term consequences. For example, a change in IQ can affect productivity over many years, while an avoided stroke can affect quality of life over many years. As a result, consequences that occur over time must be aggregated. Consistent with standard practice in the field of health economics,⁷ QALYs are discounted in a manner similar to the typical discounting of financial returns (an annual rate of 3% is used here). This practice results in greater weight being placed on health effects that occur sooner than on health effects affecting the quality of life over a more extended period. Ultimately, this approach converts a stream of consequences to a present value equivalent in the same way that investors place a present value on an income stream. This analysis also reports health impacts in terms of undiscounted lost life years not weighted to account for quality of life, and in terms of so-called “natural units,” including incremental CHD mortality risk, stroke risk, and changes in IQ.

Characterization of Uncertainty 

In addition to listing central estimates for the dose–response relationship parameters, Table 1 also characterizes the distribution of plausible values for each of these parameters. The analysis uses Monte Carlo simulation to estimate the range of plausible results (QALYs gained or lost across the population) implied by these distributions.

Because the analyses producing the dose–response relationships for CHD (König et al.³) and stroke (Bouzan et al.⁴) used regression to estimate the dose–response parameters, this paper's analysis assumes the threshold relative-risk reduction component (which corresponds to the regression's intercept coefficient) and the incremental relative risk reduction component (corresponding to the regression's linear coefficient) are normally distributed and (negatively) correlated. The estimated correlation coefficient for CHD mortality is −0.77, while for stroke, the corresponding correlation was −0.73 (correlations estimated using SAS procedure REG, SAS version 8.2 for Windows, SAS Institute, Cary NC, 2001).

Our analysis describes the plausible set of slope values for the two cognitive ability dose–response relationships (IQ and MeHg exposure, and IQ and DHA intake) as triangular distributions with bounds equal to those specified in Table 1 and modes equal to the central estimates in that table.

In addition to the probabilistic uncertainty analysis just described, this paper also reports results for four sensitivity analyses. The first considers the possibility that the dose–response relationships for CHD mortality and stroke incidence have threshold effects, but that there is no incremental benefit associated with further fish consumption (i.e., the analysis uses the central estimates for the regression intercepts but assumes the linear coefficients are zero). The second analysis omits the threshold effect (i.e., it assumes that the intercept coefficients are zero) and retains only the incremental benefit (i.e., uses the central estimates for the dose–response linear coefficients). The third analysis considers the possibility that the QALY value of an IQ point ranges from one fourth to twice its central estimate value. This range relates to assumptions about the monetary value of a QALY (see the Cognitive Effects section of the technical appendix). The final sensitivity analysis assumes that the impact of MeHg exposure on cognitive development amounts to 0.2 IQ points per microgram of MeHg per gram of maternal hair, rather than the central estimate value of 0.7 points. This lower estimate corresponds to the assumption that the dose–response relationship for MeHg exposure and IQ is linear, rather than supralinear (i.e., concave down and hence steeper at low levels of exposure). This effect estimate is also closer to the slope factor of 0.13 IQ points per parts per million of MeHg in maternal hair developed by Ryan²⁵ for the U.S.EPA's Regulatory Impact Analysis of the Clean Air Mercury Rule (see table 9-4 in U.S.EPA²⁶). Cohen et al.⁶ discuss this issue further.

Back to Article Outline

Results 

Table 2 details modeled changes in fish intake, MeHg exposure, and DHA intake for each of the five scenarios considered in this analysis. Note that the analysis omits from consideration CSFII sampling error uncertainty and numerical uncertainty introduced by the Carrington et al.¹⁸ simulation of population fish consumption patterns. Both of these sources of uncertainty are small (see Table 2 footnote) compared to the uncertainty associated with the dose–response relationships (Table 1).

Table 2. Projected changes in MeHg exposure and n-3 PUFA intake by scenario^a

CSF II, Continuing Survey of Food Intakes by Individuals; DHA, docosahexaenoic acid; g, grams; Hg, mercury; MeHg, methyl mercury; n-3 PUFA, n-3 polyunsaturated fatty acids.

Table 3 details central estimates for the modeled changes in CHD mortality, stroke mortality, nonfatal stroke incidence, and cognitive development in newborns, expressed as the aggregate change in IQ for the population. Table 3 also reports the magnitude of these impacts in terms of undiscounted life years and discounted QALYs.

Table 3. Projected health impacts by scenario compared to baseline^a

CHD, coronary heart disease; DHA, docosahexaenoic acid; MeHg, methyl mercury; QALY, quality-adjusted life year.

Consumption Scenario 1 (women of childbearing years shift to consumption of low-MeHg fish) results in a substantial improvement in cognitive development. The gain of 410,000 IQ points across the population corresponds to approximately 0.1 IQ points per child born. Because the reallocation of fish consumption leads to a slight decrease in daily fish intake in these groups (an artifact of the FDA exposure model), our analysis predicts a slight increase in CHD mortality, stroke incidence, and stroke mortality. These impacts are small when expressed in terms of QALYs, compared to the impact on cognitive development. On a per capita basis, stroke incidence and annual CHD mortality each increase by <1 per 1 million among women between the ages of 35 and 44. Per capita changes in other groups are smaller.

Scenario 2 assumes that women of childbearing age reduce fish consumption. The aggregate impact on cognitive development remains positive but is substantially smaller (a gain of 92,000 IQ points in this scenario vs a gain of 410,000 IQ points in Scenario 1). This reduction reflects the more modest reduction in MeHg intake associated with decreased fish consumption (compared to the impact on MeHg exposure of switching consumption to fish with low MeHg concentrations), and the offsetting impact of decreased DHA intake. On an individual basis, the remaining benefit is very small, amounting to approximately 0.02 IQ points per child born. The lower fish consumption (and resulting decrease in n-3 PUFA intake) results in higher CHD mortality and stroke incidence. However, because the baseline rates for these effects are low among women of childbearing age, the aggregate impact (in QALYs) is small. For women of childbearing age with the highest baseline risks for stroke and CHD mortality (aged 35 to 44), these changes are small on a per capita basis, amounting to 2 per 1 million for CHD mortality, <1 per 1 million for fatal stroke, and approximately 3 per 1 million for nonfatal stroke.

Because Scenario 3 assumes that other members of the population also decrease their fish consumption, and because the baseline rates for CHD mortality and stroke incidence are higher among many of these population segments than among women of childbearing age, this scenario results in a substantially greater impact for these effects. The impact on cognitive development in Scenario 3 is the same as it was for Scenario 2 because both scenarios assume the same change in fish consumption among women of childbearing age. On a per capita basis, the greatest losses are experienced among elderly men. For men aged between 75 and 84, annual CHD mortality risk increases by approximately 2 in 10,000.

Scenario 4 assumes that fish consumption increases among all adults with the exception of women of childbearing age. This scenario therefore does not affect prenatal cognitive development. However, CHD mortality and stroke incidence both decrease because of the increase in fish consumption among other members of the population. Again, elderly men are affected to the greatest extent on an individual basis. For men aged between 75 and 84, annual CHD mortality risk decreases by 5 in 10,000.

Scenario 5 extends the increase in fish consumption to women of childbearing age. As a result, there is a slight reduction in population risk for the CHD and stroke when compared to Scenario 4. The increase in fish consumption among women of childbearing age also has a negative impact on cognitive development, resulting in a net loss of 270,000 IQ points. This loss amounts to a modest decrement of approximately 0.07 IQ points per child born.

Figure 1 illustrates the extent to which uncertainty in the parameter estimates for the dose–response relationships (see the far right column in Table 1) contributes to uncertainty in the model's predictions for each of the five scenarios considered. The box and whisker plots denote the distribution 5th and 95th percentiles (whiskers), lower and upper quartiles (bottom and top edges of box), and median (line within box). The asterisk denotes the arithmetic mean for the corresponding distribution.

• 
  • View Large Image
  • Download to PowerPoint

• Figure 1. 

  Range of plausible values for total quality adjusted life years (QALYs) gained: uncertainty associated with dose–response parameter estimates. Asterisks denote distribution mean; whiskers denote 5th and 95th percentiles; box boundaries denote first and third quartiles; and line in box denotes median.

Table 4 summarizes the impact of alternative assumptions considered in our sensitivity analyses on aggregate QALY gains for the population. Results for the first sensitivity analysis indicate that eliminating the assumed incremental benefit of each additional fish meal has a substantial impact on the aggregate QALY estimates. On the other hand, results for the second analysis indicate that the intercept terms for the CHD and stroke dose–response relationships are less important. Assuming that the intercepts are zero for the CHD and stroke dose–response relationships does not affect the sign of the estimated population aggregate QALY impact, and has only a modest impact on its magnitude. The third sensitivity analysis (QALY value of an IQ point is one fourth to twice the central estimate value) has a substantial impact on the predicted results for Scenarios 1 and 2, but only a modest impact on Scenarios 3 through 5. The final sensitivity analysis decreases the assumed impact of MeHg exposure on IQ by a factor of approximately 3, causing the impact of DHA on IQ to exceed the impact of MeHg. As a result, the net impact of decreasing total fish consumption among women of childbearing age becomes negative (Scenario 2).

Table 4. Sensitivity analysis results: total QALYs gained compared to baseline

Sensitivity analysis	Scenario
	1	2	3	4	5
Base case assumptions (see Table 3 last row)	49,000	9,700	−41,000	120,000	90,000
1: Linear coefficients zero for stroke and CHD dose–response relationships	50,000	11,000	−6,600	25,000	−6,200
2: Intercept coefficients zero for stroke and CHD dose–response relationships	50,000	10,000	−23,000	97,000	68,000
3: IQ QALY value is twice its central estimate	99,000	21,000	−30,000	120,000	58,000
IQ QALY value is one fourth its central estimate	12,000	1,300	−49,000	120,000	110,000
4: MeHg: IQ dose–response relationship is linear, rather than concave down	17,000	−2,500	−53,000	120,000	120,000

CHD, coronary heart disease; MeHg, methyl mercury; QALY, quality-adjusted life year.

Back to Article Outline

Discussion 

Recent risk management efforts to address the issue of MeHg in fish recognize that fish is a rich source of nutrition, including protein and n-3 PUFAs. Addressing trade-offs between these contaminants and nutrients is an important public health undertaking. Although it may be possible to reduce global average MeHg concentrations in fish in the long term, short-term reductions are precluded by worldwide emissions, natural sources, and the environmental half-life of the compound.²⁷ Because of the nutritional value of fish, risk communication efforts have carefully targeted their message to women of childbearing age, encouraging them to alter the type of fish that they consume, rather than the amount (see, e.g., USDA²⁸ and Mahaffey²⁹).

Certainly, getting women to shift their consumption in this way without depressing consumption among other adults yields positive public health benefits, as suggested by the results for our Scenario 1. Accomplishing the aims of the USDA's 2005 dietary guidelines by getting people to increase their consumption of fish containing high concentrations of n-3 PUFAs²⁸ would accrue even greater benefits. For example, our model indicates that if somehow all adults (including women of childbearing age) consumed 8 oz of salmon each week (a fish with relatively low MeHg content and high n-3 PUFA content), annual CHD mortality would drop by nearly 20,000 cases, annual stroke mortality and nonfatal stroke incidence would each drop by 4000 cases, and aggregate IQ would increase by >2 million points, amounting to approximately 0.5 IQ points per child born. Annual gains would be equivalent to the saving of >400,000 QALYs (detailed results for this scenario are not shown).

However, as encouraging as this potential gain may be, risk managers must recognize the possibility that their recommendations will not be adhered to in exactly the manner intended. Indeed, women of childbearing age may associate all fish with the dangers attributed to MeHg, and therefore decrease their total fish consumption. The findings of Oken et al.²² demonstrate the very real potential for the balance in messages to be lost. Whereas effective adherence to the recommendations can result in positive health benefits (Scenario 1), unintended reactions in consumption among women of childbearing age may result in much smaller benefits for children, reducing aggregate gains from 400,000 IQ points (approximately 0.1 points per child) to 90,000 IQ points (approximately 0.02 points per child).

Warnings may also discourage fish consumption among other adults. Results for Scenario 3 show that if the rest of the population also reduces fish consumption by 17%, substantial public health losses (as measured in QALYs) may result. However, even more modest shifts may have a notable impact on the net result. The importance of this offsetting impact depends on what other population segments also reduce fish consumption. If fish consumption is reduced only among immediate family members of women of childbearing age, the offsetting health losses would not be substantial because it is likely that these individuals are young and hence have a low baseline risk for either stroke or CHD mortality. For example, if it is assumed women of childbearing age are married to individuals of approximately the same age, and furthermore that these individuals also reduce their fish consumption by the same amount as their spouses (17%), a further 1600 QALYs would be lost annually, reducing the overall annual benefit for Scenario 2 from 9700 QALYs to approximately 8100 QALYs. On the other hand, if the entire population reduces fish consumption by 3% to 4% due to perceived risks associated with this food, the net gain of 9700 QALYs in Scenario 2 would become negative. Given the potential for spillover effects, government agencies might consider measures other than advisories. For example, because consumption choices are influenced by price,³⁰ subsidies for fish with low levels of MeHg and high levels of n-3 PUFAs would tend to shift consumption preferences toward the desired species without decreasing overall fish consumption.

It is also important to consider the uncertainty associated with all of these projections, as illustrated in Figure 1. Public health benefits associated with increased fish consumption may not be nearly as large as our central estimate assumptions suggest, or they may be substantially larger (Scenarios 4 and 5). Results for Scenario 1 are also relatively uncertain. Given the sensitivity of the results to the form of the dose–response relationship assumed for MeHg exposure and IQ, further efforts to resolve this particular source of uncertainty would be useful (see fourth analysis in Table 4, which uses a slope factor closer to that estimated for the U.S.EPA's Regulatory Impact Analysis of the Clean Air Mercury Rule). Gathering better estimates for the effect on CHD mortality of incremental fish consumption would also be helpful (i.e., better resolving the value of the linear coefficient for this dose–response relationship). It is reassuring that the sensitivity analyses summarized in Table 4 indicate that changing many other parameters does not qualitatively alter the overall direction of the predicted aggregate health impact for any of the scenarios. That is, while the magnitude of the aggregate impact changes in these other cases, its overall direction does not.

As noted in this paper's introduction, this analysis omits consideration of effects related to polychlorinated biphenyl (PCB) exposure resulting from fish consumption. However, potential effects associated with this contaminant are unlikely to be important compared to those effects that were considered. Consider Scenario 5, which nets 120,000 QALYs annually as the result of a 9.3 g/day increase in fish consumption (Table 2, Table 3). Based on samples collected in nine U.S. cities, Hites et al.³¹ estimated that lifetime exposure to organic chemicals (PCBs, toxaphene, and dieldrin) associated with consumption of between 4 and 16 oz of farmed salmon per month (median of 8 oz per month) would result in a lifetime cancer risk of 1 × 10⁻⁵. Consumption of 8 oz of fish per month amounts to 7.5 g/day, which is similar to the 9.3-g/day increase envisioned in Scenario 5. Cohen et al.³² estimated that the average cancer death results in the loss of 14.6 life years. Treating this loss as approximately equal to the loss in QALYs, and assuming an average lifetime of 70 years, yields an annual QALY loss of 14.6 QALYs × 10⁻⁵ / 70 years per person, or approximately 600 QALYs per year for the US population. This calculation makes a number of uncertain assumptions (e.g., the assumption that organic contaminant concentrations in farmed salmon are similar to those in other fish consumed in the United States). However, the result suggests at first glance that risks from other organic contaminants (600 QALYs per year) are small compared to the benefits associated with fish consumption (net of 120,000 QALYs per year).

By identifying potential pitfalls associated with various risk-management measures, we do not mean to suggest that no effort should be made to communicate the risks associated with MeHg exposure resulting from fish consumption. Certainly, among populations exposed to highly contaminated local sources of fish, such measures may be especially important. However, the potential adverse effects of such efforts must be kept in mind.³³ As such, it is incumbent on risk managers to carefully and quantitatively evaluate how all members of the population will react to measures under consideration, how such reactions will affect exposure to contaminants and intake of nutrients, and how changes in contaminant exposure and nutrient intake will affect the probability of various adverse and beneficial health effects. Only by carefully considering these implications can we hope to improve the likelihood that our efforts will aid public health.

Back to Article Outline

The expert panel for this project was chaired by Steven M. Teutsch. The remainder of the panel consisted of the following co-authors of this paper: David Bellinger, William Connor, Penny Kris-Etherton, Robert Lawrence, David Savitz, and Bennett Shaywitz. This work was supported by a grant from the National Food Processors Association Research Foundation (NFPA-RF) and the Fisheries Scholarship Fund. The sponsor played no role in the design or conduct of this study or in the identification and interpretation of the literature, and the authors retained control over the final form of the manuscript. NFPA-RF did offer comments on an earlier draft of this manuscript, in response to which we made minor revisions (details are available upon request). Member companies of the NFPA-RF may be affected by the findings of research that funded my participation on the panel that wrote this paper.

Back to Article Outline

References 

1. U.S. Department of Health and Human Services, U.S. Environmental Protection Agency. What you need to know about mercury in fish and shellfish, 2004. Available at: www.cfsan.fda.gov/∼dms/admehg3.html. Accessed December 2004.
   • View In Article
2. Graham JD , Wiener JB . Risk versus risk: tradeoffs in protecting health and the environment . Cambridge MA: Harvard University Press; 1995;
   • View In Article
3. König A, Bouzan C, Cohen JT, et al. A quantitative analysis of fish consumption and coronary heart disease mortality. Am J Prev Med 2005;29:335–46.
   • View In Article
4. Bouzan C, Cohen JT, Connor WE, et al. A quantitative analysis of fish consumption and stroke risk. Am J Prev Med 2005;29:347–52.
   • View In Article
5. Cohen JT, Bellinger DC, Connor WE, Shaywitz BA. A quantitative analysis of prenatal intake of n-3 polyunsaturated fatty acids and cognitive development. Am J Prev Med 2005;29:366–74.
   • View In Article
6. Cohen JT, Bellinger DC, Shaywitz BA. A quantitative analysis of prenatal methyl mercury exposure and cognitive development. Am J Prev Med 2005;29:353–65.
   • View In Article
7. Gold MR , Siegel JE , Russell LB , Weinstein MC . Cost-effectiveness in health and medicine . New York: Oxford University Press; 1996;
   • View In Article
8. Ponce RA , Bartell SM , Wong EY , et al.  Use of quality-adjusted life year weights with dose-response models for public health decisions: a case sstudy on the risks and benefits of fish consumption . Risk Anal . 2000;20:529–542
   • View In Article
   • MEDLINE
9. Daviglus ML , Stamler J , Orencia AJ , et al.  Fish consumption and the 30-year risk of fatal myocardial infarction . N Engl J Med . 1997;336:1046–1053
   • View In Article
   • MEDLINE
   • CrossRef
10. National Research Council. Toxicological effects of methylmercury. Washington DC: National Academy Press, 2000.
    • View In Article
11. Wilkinson T. Allure of Alaska's wild fish: why expensive Alaskan salmon have conquered the fish counters across the Lower 48. Christian Science Monitor, Boston, March 31, 2005.
    • View In Article
12. Schachter HM, Reisman J, Tran K, et al. Health effects of omega-3 fatty acids on asthma. Rockville MD: Prepared by University of Ottawa Evidence-Based Practice Center for the Agency for Health Care Research and Quality, 2004 (AHRQ publication 04-E013-1).
    • View In Article
13. MacLean CH, Mojica WA, Morton SC, et al. Effects of omega-3 fatty acids on lipids and glycemic control in type II diabetes and the metabolic syndrome and on inflammatory bowel disease, rheumatoid arthritis, renal disease, systemic lupus erythematosus, and osteoporosis. Rockville MD: Prepared by Southern California/RAND Evidence-Based Practice Center for the Agency for Health Care Research and Quality, 2004 (AHRQ publication 04-E012-1).
    • View In Article
14. Hu FB , Manson JE , Willett WC . Types of dietary fat and risk of coronary heart disease: a critical review . J Am Coll Nutr . 2001;20:5–19
    • View In Article
    • MEDLINE
15. Wolfram G . Dietary fatty acids and coronary heart disease . Eur J Med Res . 2003;8:321–324
    • View In Article
    • MEDLINE
16. Institute of Medicine. Dioxins and dioxin-like compounds in the food supply: strategies to decrease exposure. Washington DC: National Academy Press, 2003.
    • View In Article
17. Tidwell JH , Allan GL . Fish as food: aquaculture's contribution (Ecological and economic impacts and contributions of fish farming and capture fisheries) . EMBO Rep . 2001;2:958–963
    • View In Article
    • MEDLINE
    • CrossRef
18. Carrington CD , Bolger MP . An exposure assessment for methylmercury from seafood for consumers in the United States . Risk Anal . 2002;22:689–699
    • View In Article
    • MEDLINE
    • CrossRef
19. Carrington C, Montwill B, Bolger M. An intervention analysis for the reduction of exposure to methylmercury from the consumption of seafood by women of child-bearing age. Washington DC: U.S. Food and Drug Administration, 2004.
    • View In Article
20. U.S. Department of Agriculture. 1989–91 Continuing Survey of Food Intake by Individuals (CSFII) and 1989–91 Diet and Health Knowledge Survey. Hyattsville MD: Agriculture Research Service, 1998 (CD-ROM; PB96-501747).
    • View In Article
21. National Marine Fisheries Service. Survey of Trace Elements in Fishery Resource. Washington DC: National Oceanic and Atmospheric Administration, 1978 (NMFS SSRF-721).
    • View In Article
22. Oken E , Kleinman KP , Berland WE , Simon S , Rich-Edwards JW , Gillman MW . Decline in fish consumption among pregnant women after a national mercury advisory . Obstet Gynecol . 2003;102:346–351
    • View In Article
    • MEDLINE
    • CrossRef
23. Myers GJ , Davidson P , Cox C , et al.  Prenatal methylmercury exposure from ocean fish consumption in the Seychelles child development study . Lancet . 2003;361:1686–1692
    • View In Article
    • Abstract
    • Full Text
    • Full-Text PDF (143 KB)
    • CrossRef
24. U.S. Environmental Protection Agency. Exposure factors handbook. Volume II. Food ingestion factors. Washington DC: Office of Research and Development, National Center for Environmental Assessment, 1997 (EPA/600/P-95/002Fb; PB98-124233).
    • View In Article
25. Ryan LM. Effects of prenatal methylmercury on childhood IQ: a synthesis of three studies. Report to the U.S. Environmental Protection Agency. Boston MA: USEP, 2005.
    • View In Article
26. U.S. Environmental Protection Agency. Regulatory impact analysis of the Clean Air Mercury Rule. Research Triangle Park NC: Office of Air Quality Planning and Standards, Air Quality Strategies and Standards Division, 2005 (EPA-452/R-05-003).
    • View In Article
27. U.S. Environmental Protection Agency. Mercury study report to Congress. Washington DC: U.S. EPA, 1997 (1997EPA 452/R-97-003). Available at: www.epa.gov/oar/mercury.html. Accessed September 2004.
    • View In Article
28. U.S. Department of Agriculture. Dietary guidelines for Americans: 2005, 2004. Available at: www.health.gov/dietaryguidelines/dga2005/report/. Accessed November 2004.
    • View In Article
29. Mahaffey KR . Fish and shellfish as dietary sources of methylmercury and the omega-3 fatty acids, eicosahexaenoic acid and docosahexaenoic acid: risks and benefits . Environ Res . 2004;95:414–428
    • View In Article
    • MEDLINE
    • CrossRef
30. Huang KS , Haidacher RC . Estimation of a composite food demand system for the United States . J Business Econ Stat . 1983;1:285–291
    • View In Article
31. Hites RA , Foran JA , Carpenter DO , Hamilton MC , Knuth BA , Schwager SJ . Global assessment of organic contaminants in farmed salmon . Science . 2004;303:226–229
    • View In Article
    • CrossRef
32. Cohen JT , Beck BD , Rudel R . Life years lost at hazardous waste sites: Remediation worker fatalities vs. cancer deaths to nearby residents . Risk Anal . 1997;17:419–425
    • View In Article
    • MEDLINE
    • CrossRef
33. Arnold SM , Lynn TV , Verbrugge LA , Middaugh JP . Human biomonitoring to optimize fish consumption advice: reducing uncertainty when evaluating benefits and risks . Am J Public Health . 2005;95:393–397
    • View In Article
    • MEDLINE
    • CrossRef