Abstract

BACKGROUND: In the past, systematic reviews of randomized trials about consumption of fish oil found protective effects against all-cause mortality. But results from the largest trial available were excluded from these analysis' biasing associations toward a protective effect. Randomized trials about fish consumption showed no association with mortality and until now, no systematic review of prospective studies was published examing the association between fish consumption and all-cause mortality.
OBJECTIVES:

• To review all prospective studies which published information about total fish, or specific types of fish, in relation with all-cause mortality/survival.
• To define the amount of consumption found to be related with these effects.
• To define possible effect modification by confounders.

DATA SOURCE: The Pubmed database was searched (No start date - Dec 13, 2010) for relevant articles using the keyword fish, omega 3, polyunsaturated fat, or seafood, combined with prospective, cohort, follow-up, or longitudinal and a fair amount of other keywords. The exact search term is described Here.
Prospective studies published in the English language were included. Reference lists were searched for additional articles.
RESULTS: 43 articles were found which provided information about 31 different cohorts. Of these, 0 articles were excluded.
Results are described when any evidence for an association - as defined in the Methods - was found. In addition, data about total fish, fatty fish, and lean fish is described.

• Data about total fish was provided by 30 cohorts, including 48,411 deaths. Significant protective effects were found in 4 cohorts and in 2 subcohorts. And nonsignificant protective effects were found in 3 other cohorts. These 9 cohorts included 38% of all cases. Suggestive evidence was found that total fish consumption decreases all-cause mortality (RR = 0.91).
• Stratified analysis showed that a protective effect might be found among US populations (8 cohorts: RR = 0.88) and Asian populations (4 cohorts: RR = 0.89), but not among European populations (17 cohorts: RR = 0.98). Effects were similar among men and women.
• Protective effects were inconsistently found at different levels of consumption, varying from 1 to 5.7 servings/week. RR's for < 1, 1, 2, and ≥ 3 servings/week were 0.98, 0.94, 0.94, and 0.91, respectively.
• Inconsistent findings were done among subjects with prevalent CHD, prevalent diabetes, and high-risk individuals based on blood pressure and/or serum cholesterol.
• No associations were found for fatty fish (3 cohorts) or lean fish (1 cohort) intake. Little data was available about other types of fish.

CONCLUSION: Suggestive evidence was found for a protective effect of total fish consumption against all-cause mortality (- 9% risk). The effect was strongest for the highest level of intake (≥ 3 servings/week). Stratified analysis showed that protective effects were found among US and Asian cohorts, but not among European cohorts. Inconclusive evidence was found for an association among subjects with prevalent disease. No/inconclusive evidence was found for an association between consumption of any type of fish and mortality.
LIMITATIONS: Hardly any data was available about specific types of fish, precluding evidence based recommendations to consume any type of fish.
PERSPECTIVE: Until now, only 2 randomized trials examined the relation between fish consumption and mortality. Both were European trials and both failed at finding an association, which is in line with results from prospective European studies.

Introduction.

Results from previous systematic reviews: A meta-analysis of randomized trials has shown that fish oil significantly decreases all-cause mortality by 17% (Mozaffarian D. 2006). Participants were randomized to fish consumption in 2 of these trials, and randomized to fish oil supplementation in the remaining trials. In one of the studies about fish consumption, (Burr ML. 1989) a significant protective effect was found (RR = 0.71; 95% CI = 0.54-0.92). 224 subjects died and the trial contributed fairly strong to the overall effects size (18.7% weight). But in 2002, another article was published about the same cohort, including a longer period of follow-up and 1,083 deceased subjects (Ness AR. 2002). No significant effect remained after this longer period of follow-up: RR = 0.94 (0.84-1.06). Excluding results from this article from the meta-analysis strongly biased the effect size in the meta-analysis by Mozaffarian et al.
In the same year, another meta-analysis of randomized trials was published (Hooper L. 2006). Marine omega 3 fats from fish or fish oil supplements decreased all-cause mortality, but this effect did not reach significance: RR = 0.86 (0.70-1.04). Again, the article from Ness et al was excluded from the meta-analysis, biasing the RR toward a stronger effect.
Randomized trials relating fish consumption to all-cause mortality: 3 articles provided information about 2 different trials. In both trials, UK men were randomized to ≥ 2 weekly portions of fatty fish, or to the control group. A significant protective effect was found in the DART study after 2 years of follow-up (Burr ML. 1989). But no significant association remained after 17 years of follow-up: RR = 0.94 (0.84-1.06) (Ness AR. 2002). In the DART 2 trial, also no significant association was found with total mortality: HR = 1.13 (0.94-1.37) (Burr ML. 2003).

|Additional references:
Burr ML. Effects of changes in fat, fish, and fibre intakes on death and myocardial reinfarction: diet and reinfarction trial (DART). Lancet. 1989 Sep 30;2(8666):757-61. Abstract
Burr ML. Lack of benefit of dietary advice to men with angina: results of a controlled trial. Eur J Clin Nutr. 2003 Feb;57(2):193-200. Abstract
Hooper L. Risks and benefits of omega 3 fats for mortality, cardiovascular disease, and cancer: systematic review. BMJ. 2006 Apr 1;332(7544):752-60. Full text
Mozaffarian D. Fish intake, contaminants, and human health: evaluating the risks and the benefits. JAMA. 2006 Oct 18;296(15):1885-99. Full text.
Ness AR. The long-term effect of dietary advice in men with coronary disease: follow-up of the Diet and Reinfarction trial (DART). Eur J Clin Nutr. 2002 Jun;56(6):512-8. Full text |

Methods.

Defining fish consumption: "Total fish" included data about fish consumption with or without seafood/shellfish, or without using a definition for fish. Preferably, the definition excluded seafood/shellfish.
Data about "omega-3 fatty acids" was included as a surrogate for fish consumption, if these fatty acids reflected (primarily) non-supplemental dietary intakes; were defined as "marine omega-3 fatty acids", "omega-3 fatty acids from fish (oil)", or "dietary fish oil"; and did not include omega-3 from other dietary sources.
Preferably, fish consumption was chosen over consumption of marine omega-3 fatty acids.
Conversions for different units of consumption: Four systematic reviews provided data about conversion of grams to servings. In 2005, authors from two of these reviews (Bouzan C; König A) assumed that 100 g of fish was equivalent to one serving. In 2004, He K. published two systematic reviews. In both cases, the assumption was made that 105 g of fish was equivalent to one serving, based on the derived average portion size in the Health Professionals Follow-Up Study.
For my own systematic reviews, 105 g of fish was converted to one serving/portion. And for frequency intakes per week, the assumption was made that consumption of fish once/week is equal to consumption of one serving fish/week.
Dose response methodology: Fish consumption was categorized into standardized levels of intake. If one of these standardized levels of intake fell into two categories of intake in a single study, the average RR from these two categories was found to reflect the RR for this standardized level of intake. If > 1 standardized levels of intake fell into the same category of intake in a single study. The RR for this category was found to reflect the RR for all these standardized levels of intake. Studies were considered if they included ≥ 3 different levels of consumption. Studies examining fish oil as a proxy for fish consumption were not included.
Data from the author "Hirayama T.": Dr. Hirayama examined the effects of a small amount of food groups in relation to a large amount of mortality end points in a Japanese cohort of very large size. An extended review of his work was published as a book in 1990 (Hirayama T [7]). Data about this cohort is seldom included in current systematic reviews about the relations mentioned. Dr. Hirayama published a lot of articles stating that vegetables and meats were related to several disease end points, adjusted for age and sex. However, the book included one page showing effects after multivariate analysis including cigarette smoking, meat, green-yellow vegetables, and alcohol. This analysis showed that a large amount of previously published effects completely changed when these variables were taken into account.
Since a) Dr. Hirayama himself only published sex-, and age-adjusted results in the English language, while results following multivariate analysis often were completely different b) the results were published as a book and not in a peer-reviewed journal, and c) Dr. Hirayama was the only researcher examining this cohort, results from his cohort are debatable. Results will be presented including effects from his work, but his work will not be included when the evidence for a possible effect is judged.
Data from the author "Knoops KT": Dr. Knoops examined the effect from consumption of several dietary variables on mortality from all causes in The HALE Study (Knoops KT [28]). But 2 inconsistencies attract the attention:

• HR's are given for the total population, as well as for 2 subpopulations ("Northern European" subjects vs "Southern European" subjects). But HR's for the total population do not always cover the HR's for the subpopulations. This is the case for grains, saturated fat, and complex carbohydrates.
• For saturated fatty acids, the HR = 1.25 (1.10-1.41) for the total population, for low consumption of saturated fat (< 10 en%) vs high consumption. HR's for both subpopulations are nonsignificant (HR = 0.98; 95% CI = 0.61-1.56, and HR = 1.19; 95% CI = 0.95-1.48). But the authors conclude that "the intake of saturated fatty acids was positively associated with mortality".

One of the authors replied to my question for clarity on this subject. She agreed that the published data was not correct. Sadly, she did not have access to the original data.
Data on at least 3 dietary variables seems to be published incorrectly, and the interpretation for the data about saturated fat seems to be incorrect. Therefore, results from this article are debatable. And results from an other publication about this cohort - including fewer subjects - were chosen to be included (Iestra J [28].

|Additional references:
Bouzan C. A quantitative analysis of fish consumption and stroke risk. Am J Prev Med. 2005 Nov;29(4):347-52. Abstract
He K. Accumulated evidence on fish consumption and coronary heart disease mortality: a meta-analysis of cohort studies. Circulation. 2004 Jun 8;109(22):2705-11. Full text
He K. Fish consumption and incidence of stroke: a meta-analysis of cohort studies. Stroke. 2004 Jul;35(7):1538-42. Full text
König A. A quantitative analysis of fish consumption and coronary heart disease mortality. Am J Prev Med. 2005 Nov;29(4):335-46. Abstract|

Total fish and all-cause mortality or survival.

Data about total fish was provided by 31 cohorts, including 103,934 cases. Survival was the end point in one cohort (Nube M [5]) and all-cause mortality was the end point in all other cohorts (Table 1).
Significant protective effects were found in 5 cohorts (Hirayama T [7], Albert CM [12], Yuan JM [17], Hu FB [19], Barzi F [20]) and in 2 subcohorts (Nube M, among men only [5], Tomasallo C, among referents only [31]). This analysis included 65,960 cases (63% of all cases).
After excluding the debatable results from Hirayama T [7], the analysis left 10,437 cases (22% of all cases). In addition, nonsignificant protective effects were found in 3 cohorts (Dolecek TA [8], Erkkilä AT [22], Yamagishi K [29], including 7,564 cases (16% of all cases).
But significantly increased risks were found in one cohort (Osler M [21]), and in one subcohort (Nube M, among women only [5]). The average RR = 0.83, but the effect size was cut in half after excluding the debatable results from Hirayama T: RR = 0.91.

Inclusion of intermediate levels of consumption:
Average RR's for intakes of < 1, 1, 2, and ≥ 3 servings/week were 0.98, 0.94, 0.94, and 0.91 after excluding results from Hirayama T. This analysis can be found here. Protective effects were found at different levels of consumption across the cohorts examined. Protective effects at any level of consumption were as follows:

• Nube M [5]: Significant at consumption 4-16 times/month among men only.
• Hirayama T [7]: Significant at consumption ≥ 1 time/month.
• Albert CM [12]: Significant at ≥ 1 serving/week.
• Gillum RF [15]: significant at consumption 1 time/week among white men only.
• Yuan JM [17]: Significant at consumption ≥ 3.3 servings/week.
• Barzi F [20]: Significant at consumption ≥ 2 times/week.
• Erkkilä AT [22]: Significant at consumption ≥ 3.8 servings/week.
• Nakamura Y [26]: Nonsignificant at consumption 3.5 servings/week.
• Yamagishi K [29]: Nonsignificant at consumption 5.7 servings/week.
• Tomasallo C [31]: Significant at consumption ≥ 1 time/week.

Effect modification: A detailed analysis can be found here. Briefly, stratified analysis showed the following effects:

• Effects did not differ between men (RR = 0.90) and women (RR = 0.88).
• Protective effects were stronger among US cohorts (RR = 0.88) and Asian cohorts (RR = 0.89) than among European cohorts (RR = 0.98).
• Hardly any other possible effect modifiers were examined. No associations were found.

Subjects with prevalent heart disease: Data about subjects with prevalent CHD/MI was provided by 6 cohorts (Table 2). A significant protective effect was found in one cohort (Barzi F [20]). And a nonsignificant protective effect was found in one cohort of very small size (Erkkilä AT [22]). No other associations were found. The average RR was not calculated because relevant data was missing for 2 cohorts (Folsom AR [25], Iestra J [28]).
Subjects with prevalent diabetes: Data about subjects with prevalent diabetes was provided by 3 cohorts (Table 3). A significant protective effect was found in one cohort (Hu FB [19]). No other associations were found.
High-risk individuals based on blood pressure and/or serum cholesterol: Data about high-risk individuals was provided by 2 cohorts (Table 4). A nonsignificant protective effect was found in the US cohort (Dolecek TA [8]), while a significantly increased risk was found in the European cohort (Osler M [21]).

Conclusion: Significant protective effects of total fish consumption were found in 4 cohorts and in 2 subcohorts. In addition, nonsignificant protective effects were found in 3 cohorts. These 9 cohorts included 38% of all cases. Suggestive evidence was found that total fish consumption protects against all-cause mortality (- 9%). The level of consumption for this effect could not be clearly defined, but the effect size was strongest for the highest level of consumption (≥ 3 servings/week).
Stratified analysis showed that any possible protective effect seemed to be restricted to US (- 12%) and Asian (- 11%) cohorts, while no evidence for an association was found among European cohorts (- 2%). Effects did not seem to differ between men and women. Inconclusive evidence was found for an association among subjects with prevalent CHD or diabetes.