Chapter  116:  Effects of Omega-3 Fatty Acids on Mental Health

Treatment Options

Treatment options for patients diagnosed with depression include psychotherapy, pharmacotherapy, and in some instances, electroconvulsive therapy. For individuals with severe depression, antidepressant medication is the treatment of choice, whereas psychotherapy alone may be sufficient to treat individuals with mild to moderate depression.

The most commonly used antidepressant medications include the selective serotonin reuptake inhibitors (SSRIs) and tricyclic antidepressants. The monoamine oxidase inhibitors (MAOIs) are used less frequently. Herbal therapy, including St. John's wort, has also been suggested as being helpful in the treatment of depression. The treatment of choice for bipolar disorder remains lithium or divalproex. The SSRIs and tricyclics act by slowing the reuptake of neurotransmitters, thus making them more available. The SSRIs work specifically on the neurotransmitter serotonin, whereas the tricyclics and MAOIs work on both serotonin and norepinephrine. In general, the SSRIs appear to demonstrate fewer side effects than do the tricyclics or MAOIs.

In spite of the availability of these medications, it has been estimated that 29% to 46% of patients are treatment-resistant, that is, they show no clinical response or only a partial response to the antidepressant medications. One approach to dealing with treatment-resistant depression is the use of combination therapy or the addition of a “booster drug” to augment the effects of the primary medication(s). Natural compounds, including omega-3 fatty acids, have recently been touted as potential augmentors of antidepressants' effects in treatment-resistant depression.^{18, 19}

Treatment Options

As with other mental health disorders, treatment for anxiety disorders, including obsessive-compulsive disorder, typically involves pharmacologic and psychotherapy treatment approaches. Pharmacologic treatment options have included benzodiazepines, tricyclic antidepressants and MAOIs, but more recently include antidepressant medications such as the SSRIs. Psychotherapeutic strategies that help patients cope with their anxiety include the cognitive-behavioural therapies. For individuals with obsessive-compulsive disorder, complete remission is rare, with most people requiring longterm medication.

Treatment Options

The goal of treatment of individuals with anorexia is weight gain. To achieve this, physicians must restore healthy eating patterns and to address thoughts and feelings concerning body image. This usually requires individual and/or family psychotherapy, and in some instances the use of antidepressant medications.

Treatment Options

According to the American Academy of Pediatrics, children with AD/HD should be treated with a stimulant medication such as methylphenidate (Ritalin®), dextroamphetamine and/or behavior therapy.²³ A relatively recent complementary or alternative approach, called EEG-centered biofeedback, aims to teach the child to modulate their own attentional states so that they may adapt more readily to varying environmental expectations regarding behavior.

Treatment Options

Pharmacological treatment for aggression includes the full spectrum of psychotropic medications including antidepressant medications, neuroleptics, and mood stabilizers. Although these agents have been used successfully in the clinic or in clinical trials, the Food and Drug Administration (FDA) has yet to approve any agent specifically for the treatment of aggression. Approaches to dealing with anger, hostility or violence have also ranged from psychotherapy to incarceration.

Treatment Options

Treatment options for individuals with alcoholism include self-help programs and psychosocial therapy. Pharmacotherapy is often used as an adjunct to psychosocial therapy, with the former including medications such as naltrexone that block the alcohol-brain interaction(s).

Treatment Options

Borderline personality disorder does not appear to respond well to existing pharmacotherapy approaches. In general, antidepressants and mood stabilizers are used to treat some of the defining symptoms, such as depression or psychosis. A new form of psychotherapy called dialectical behavior therapy, developed specifically to treat patients with borderline personality disorder, has shown promising results.²⁶

Treatment Options

The hallmark of schizophrenia treatment remains antipsychotic medications. These help to reduce symptoms, thus allowing patients to function better and improve their quality of life. Although these medications have been available since the mid-1950's, many have demonstrated significant side effects. A new class of antipsychotic, the atypical antipsychotics, have been available since the late 1980's. These medications, which include clozapine, risperidone, olozapine, quetiapine, ziprasidone and aripiprazole, may be somewhat more effective while producing fewer side effects than the earlier neuroleptic mediations. However, side effects with these medications still occur and often it is necessary to alter dosages or add additional drugs to find the most effective approach.

Treatment Options

There is currently no single treatment approach for individuals with autism. Most healthcare professionals agree that early intervention is important. Pharmacotherapy is sometimes used to treat associated behavioral problems (e.g., aggression, self-injurious behaviour) so that the individual can function more smoothly at home and school. The possible importance of nutrition has also been speculated upon.

Search Strategy

The search strategy for this project was designed to be comprehensive and achieve the highest possible recall of relevant clinical studies. The electronic search strategy was developed by an information specialist in consultation with clinical content experts in mental health. Because of the number of conditions falling under the rubric of mental health, the mental health subject tree and index terms for suicidal, aggressive and impulsive behavior was used, rather than terms appearing in free text. For those with less robust subject indexing in the area of mental health, supplemental free text terms were added to the electronic search strategy (CDSR, CAB Health). The mental health search concept was combined with the core omega-3 fatty acids search strategy established in collaboration with the project librarians, biochemists, nutritionists, and clinicians from the three EPCs involved in the 2-year, Health Benefits of Omega-3 Fatty Acids task order. Consultation among these sources provided the biochemical names and abbreviations of omega-3 fatty acids, names of commercial omega-3 fatty acids products, and food sources of omega-3 fatty acids.

The following electronic databases were searched: Medline (1966 - November Week 2 2003 and updated to April Week 3 2004), Embase (1980 to 2003 Week 48 and updated to 2004 Week 18), the Cochrane Library including the Cochrane Central Register of Controlled Trials (3^rd Quarter 2003), PsycInfo (1982 to December Week 1, 2003) and CAB Health (1973-Sept 2003). All databases were searched via the Ovid interface using Search Strategy 1 (Appendix A), except CDSR where we used Search Strategy 2 (Appendix A ^*) and CAB Health, which was searched through SilverPlatter using Search Strategy 3 (Appendix A ^*). Searches were not restricted by language of publication, publication type, or study design, except with respect to the MeSH term “dietary fats,” which was limited by study design to increase its specificity. A total of 1606 bibliographic records were downloaded, with 410 duplicate records identified and removed using citation management software (Reference Manager®).

Reference lists of included studies, book chapters, and narrative or systematic reviews retrieved after having passed the first level of relevance screening, were manually searched to identify additional unique references. Through contact with content experts, attempts were made to identify both published and unpublished studies. On behalf of the three EPCs investigating the evidence concerning the health benefits of omega-3 fatty acids, a letter was written to industry representatives to obtain additional evidence (Appendix B ^*). Unsuccessful attempts were made to contact the lead author of a recent Cochrane Collaboration systematic review of PUFA supplementation for schizophrenia to obtain unpublished data they claimed to have received from investigators.⁶¹ These supplementary efforts identified an additional 16 records that were added to the collection for review. A final set of 1,212 unique references was identified.

Eligibility Criteria

Published and unpublished studies, written in any language, were eligible for inclusion. Excluding grey literature from systematic reviews of interventions can lead to the overestimation of effect sizes.⁶² Substantial bias in the results of a systematic review pertaining to a complementary/alternative medical (CAM) intervention can ensue from the exclusion of data from reports written in languages other than English.⁶³ AHRQ and ODS consider omega-3 fatty acids to be a CAM exposure.

Data from live human study populations or subpopulations (e.g., genetic, minority, ethnic: e.g., diabetic) of any age were required to maximize generalizability. Study populations in treatment studies, as well as in those investigating the possible association of the onset, continuation or recurrence of psychiatric disorders or conditions with either the intake of omega-3 fatty acids or the fatty acid content of biomarkers had to have been assessed using any formal psychiatric diagnostic criteria (e.g., DSM-IV) or established psychiatric research instruments (e.g., Hamilton Depression Rating Scale). Our TEP requested that we investigate both psychiatric disorders and psychiatric conditions (i.e., behaviors, symptoms: e.g., dysphoric mood), recognizing that while the latter are necessary to identify a psychiatric disorder, alone they are insufficient to signal the presence of one (e.g., major depression). Any and all types of comorbid condition were eligible. Studies conducted in any era of psychiatric practice were considered candidates for inclusion.

The specific types of population required to address each of the basic research questions are described with reference to the analytic framework and those details are not repeated here. As one point of clarification, our TEP asked that, within the context of assessing the possible association between omega-3 fatty acid intake and psychiatric disorders or conditions (Question 2) we should review studies examining the possible protective effects of omega-3 fatty acid intake on the development of maladaptive behavior in populations presumed to be “healthy” (e.g., college volunteers), yet who might, for example, develop evidence of disrupted well-being when subjected to stressful circumstances. Excluded populations were those with degenerative (e.g., Alzheimer's) and peroxisomal (e.g., Zellweger's) disorders since each was addressed in SC-RAND's year-2 review of the evidence concerning omega-3 fatty acids in neurology.

Treatment studies, as well as those investigating the possible association between omega-3 fatty acid intake and the onset, continuation or recurrence of specific psychiatric disorders or conditions, had to specifically investigate foods or supplements known to contain omega-3 fatty acids of any type (e.g., EPA, ALA), from any source (e.g., fish, walnuts, seed oil), any serving size or dose, delivered in any fashion (e.g., capsules, liquid, PUFA-rich diet), and for any length of time. In all studies, some method had to have been employed to suggest the presence of omega-3 fatty acid content in the exposure, if not its actual amount (e.g., g/d). Studies investigating “PUFAs” or “ LC PUFAs,” or even types of diet one might presume would contain marine or land sources of omega-3 fatty acids (e.g., “Mediterranean diet”) at minimum had to highlight at least one source of the omega-3 fatty acid content (e.g., oily fish servings). No restrictions were placed on the types or doses of pre- or on-study cointerventions (e.g., medication, omega-6 fatty acid intake, other dietary supplements).

Controlled studies were required to address questions of intervention efficacy or effectiveness, with randomized controlled trials (RCTs) being the gold standard method to investigate these questions (Question 1).⁶⁴ Any definition of control, or comparator, was permitted. RCTs exhibit a greater inherent potential to deal with potentially serious biasing influences (e.g., selection bias) although a poorly designed or conducted RCT can produce results whose interpretability is no less complicated by the presence of confounding influences, for example, than observations derived from a well-constructed and conducted study employing a design with a lesser intrinsic capacity to control for these biases (e.g., non-RCT; prospective cohort study). For example, not all RCTs succeed, either through an explicit experimental plan or the process of randomization per se, to equally distribute known confounding influences (e.g., background diet; energy/caloric intake from the intervention; types and doses of psychotropic medication) across study arms in intervention studies. That said, our TEP asked that we identify all excluded uncontrolled studies with respect to questions of intervention efficacy/effectiveness so that future synthesis work could begin with these data. We achieved this by adding a third level of screening, which yielded a listing of citations for these excluded studies (see Study Selection Process section for details).

Any type of research design other than noncomparative case series or case studies was deemed appropriate for questions concerning the possible association between the intake of omega-3 fatty acids and the onset, continuation or recurrence of psychiatric disorders or conditions (Question 2). Often, but not exclusively, relevant data were generated by cross-sectional surveys involving a single sample. A special interpretative emphasis was placed on results from prevention RCTs and other controlled prospective designs.

Controlled studies were required to address the questions of the possible association between the fatty acid content of biomarkers and the onset, continuation or recurrence of psychiatric disorders or conditions (Question 3). Evidence of the possible role played by the fatty acid content of biomarkers in the etiology of schizophrenia, for example, requires derivation from controlled designs although not all of these designs are equal in their capacity to generate data directly pertinent to Question 3. A special interpretative emphasis was thus placed on results from prospective controlled designs, with cross-sectional studies yielding the least direct evidence.

Overall, any and all clinical outcomes were considered relevant, including symptom severity or control, response rate, incidence, prevalence or diagnostic status (e.g., case-control or cross-sectional studies). As markers of omega-3 fatty acid metabolism, the following fatty acid compositions or concentrations, from any source (e.g., red blood cell [RBC] membranes, plasma phospholipids), were considered relevant in intervention studies (i.e., exclusively as an exploratory focus on the possible covariation of clinical and biomarker effects, or correlations between these factors) or as possible predictors of the onset, continuation or recurrence of psychiatric disorders or conditions: EPA, DHA, AA/EPA, AA/DHA, AA/EPA+DHA. Studies exclusively evaluating the role of other biomarkers (e.g., cytokine production, eicosanoid levels) were not included. These decisions were made with the assistance of our TEP.

Study Selection Process

The present review employed specific electronic functionality in the form of an internet-based software system, housed on a secure web site. It brings appreciable efficiencies to the systematic review process and the management of a systematic review team. Electronic yields of literature searches are posted to the system for review. Reviewers then submit all of their results of relevance screening, data appraisal or data abstraction directly to the system. The software system automatically conducts an internal comparison of multiple reviewers' responses to screening questions, to determine the eligibility/relevance of a bibliographic record or a full report. As well, the software captures responses to specific requests to abstract pre-specified data (e.g., mean age of study participants; the assessment of a study's internal validity) from pertinent reports. One large advantage associated with using this software is that review team members are able to complete their work from wherever they have internet access.

Following a calibration exercise, which involved screening five sample records using an electronic form developed and tested especially for this review (Appendix C ^†), two reviewers independently screened the title, abstract, and key words from each bibliographic record for relevance by liberally applying the eligibility criteria. A record was retained if it appeared to contain pertinent study information. If the reviewers did not agree in finding at least one unequivocal reason for excluding it, it was entered into the next phase of the review. The reasons for exclusion were noted using a modified QUOROM format (Appendix D ^‡).⁶⁵ The screening process also aimed to identify the exact mental health question a record addressed, in addition to determining whether it might also or instead pertain to any of the other topics being systematically reviewed by the three EPCs in year 2 of the omega-3 fatty acids project.

Print or electronic copies of the full reports for those citations having passed level one screening were then retrieved. After completing a calibration exercise which involved evaluating five sample reports using the same eligibility criteria (Appendix C ^*), the rest of the reports were independently assessed by two reviewers. Reports were not masked given the equivocal evidence regarding the benefits of this practice.⁶⁶ To be considered relevant at this second level of screening, all eligibility criteria had to be met. Implementing the recommendations of our TEP, a third level of dual-reviewer screening was used to exclude, yet at the same time to identify, studies addressing questions of intervention efficacy/effectiveness employing uncontrolled research designs.

Disagreements arising at either screening levels 2 or 3 were resolved by forced consensus and, if necessary, third party intervention. Excluded studies at each of these levels are noted as to the reason for their ineligibility in listings found at the end of this report.

Overview

The evidence is presented in three ways. Evidence tables in the Appendices offer a detailed description of the included studies (e.g., study design, population characteristics [e.g., diagnosis], intervention/exposure characteristics [e.g., omega-3 fatty acid types and doses], cointervention [e.g., background diet, concurrent medication]), with a study represented only once. These tables are organized by research design (Table 1: experimental studies [e.g., treatment RCTs]; Table 2: observational studies [e.g., cross-sectional studies examining the possible association of the omega-3 or omega-6/omega-3 fatty acid content of biomarkers with the onset, continuation or recurrence of a specific psychiatric disorder or condition]; Table 3: cross-national ecological analyses [e.g., studies addressing the possible association of omega-3 fatty acid intake with the onset, continuation or recurrence of a specific psychiatric disorder or condition), with studies arranged alphabetically within each of the three table/design categories.

Question-specific summary tables embedded in the text describe each study addressing a given question in abbreviated fashion, highlighting some key characteristics, including sample size (as measure of the “weight” of the evidence and possible precision of the results), dose and type of omega-3 fatty acids, and comparators' (i.e., comparison groups') specifications. This affords a comparison of all studies addressing a given question. A study can appear in more than one summary table since it can address more than one research question. Also question-specific is each summary matrix, situating each study in terms of its study quality and its applicability.

Study Quality

Study quality refers to the internal validity, or methodological soundness, of a study. A systematic review can be faced with great variability in the quality of its included studies. Our approach is not to use a minimal level of quality as an inclusion criterion since this precludes assessing the possible impact of study quality on study results.

A study with low quality can make it difficult to clearly and meaningfully interpret its results, that is, to unequivocally attribute a significant observed benefit exclusively to an intervention/exposure (as opposed to other factors). Since definitions, or standards, of study quality can depend on the type of research design, different constructs were selected to evaluate, from study reports, the quality of RCTs and studies employing other types of research design. After a calibration exercise involving two studies with an RCT design, two assessors independently evaluated study quality. Disagreements were resolved via forced consensus. In the case of designs other than RCTs, a single experienced quality assessor performed the evaluations. Time did not permit their dual assessment.

Four fundamental quality constructs from two instruments were used to rate the internal validity of RCTs. These tools were chosen collectively by the three EPCs involved in the 2-year task order because they have been validated. The Jadad items⁶⁷ assess the reporting of randomization, double blinding, and, withdrawals and dropouts (Appendix C ^§). Total scores range from 0 to 5, with a score less than 3 indicating low quality. The reporting of the concealment of a trial's allocation to treatment⁶⁸ yields three grades (A = adequate; B = unclear; C = inadequate) (Appendix C ^*).

The assessment of the quality of studies using designs other than RCTs is complicated by the dearth of validated instruments and the variety of such designs (e.g., non-randomized controlled trials; uncontrolled studies). Nevertheless, a recent systematic review by Deeks et al. identified a number of “best tools” for use with these designs.⁶⁹ Among them was a published instrument developed by Downs and Black⁷⁰ and an unpublished one derived by experts in Newcastle and Ottawa (NOS).⁷¹ The former validated both design-specific and design-neutral items.

Where case-control and cohort studies were included in the review, the validated NOS was employed. Items applicable to other designs such as non-RCTs, cross-sectional designs, cross-sectional surveys and others were taken from the Downs and Black instrument; or, if the required constructs were not operationalized in this instrument, they were developed as modifications of existing Downs and Black items (e.g., for multiple-group cross-sectional designs), NOS items (e.g., single prospective cohort studies), borrowed from Jadad's assessment tool (e.g., description of withdrawals/dropouts), or developed outright. For example, items needed to be created to evaluate cross-national ecological analyses (Appendix C ^*).

It should be noted that the items defining the case-control and cohort study assessment tools from the NOS were each used as a whole, although specific guidelines as to which design-specific total scores indicate low or sound quality are unavailable. Likewise, no guidelines exist to mark low or sound study quality based on any subset of Downs and Black's 27-item instrument. As already asserted, an Jadad total quality score of less than 3 indicates low quality. To permit the entry of these quality data into a summary matrix, cutpoints for each type of design were set somewhat arbitrarily to establish three levels of internal validity (see Summary Matrix).

It was decided by our review team that, given the limitations of space, especially in print-based study reports, and the amount of detail that would likely be required to provide all of the details we needed to fully establish that only appropriate methods had been used to extract, prepare, store and analyze lipid content, it was reasonable to appraise these methods by focusing instead on identifying extant descriptions of inappropriate methods. On occasion, the inappropriateness of methods had to be determined by reference to standard protocols.

Pilot-tested exclusively for their ease of use within the data abstraction form were questions designed to informally assess the successful control of study confounding from variables identified by content experts as potential threats to the internal validity of studies pertinent to the review. In their view, these variables required experimental or statistical control to permit an uncomplicated interpretation of study results (Appendix C ^*). The two major categories of threat in controlled designs came from having study groups vary in terms of key prestudy or baseline characteristics (e.g., background diet; psychotropic medication; severity of a disorder), or from having certain on-study changes (e.g., unexpected stressors; changes in medication type or dose) unrelated to the exposure or intervention, occur unequally across study groups to produce confounding. Even RCTs are not immune from being affected by these threats to internal validity.

For example, if in a placebo-controlled RCT test of the supplemental treatment efficacy of omega-3 fatty acids, only certain treatment group members' background diets changed appreciably from what was observed at baseline (e.g., decreased fish intake and thus an increased omega-6/omega-3 ratio in the background diet), at which point the two study groups' baseline diets had been deemed comparable, then this on-study inequality could influence study outcomes. Because of this change in background diet, one study group might all of a sudden be receiving a different ratio of omega-6/omega-3 fatty acid intake than what had been set in the study protocol. This would amount to a change in the planned, on-study between-group difference in omega-6/omega-3 fatty acid intake; and, it is this intake ratio which could have the greatest influence on clinical outcomes. In general, contraventions of planned on-study between-group equivalences (e.g., caloric/energy intake; background diet; medication types and doses; severity of disorder; current smoker status; alcohol consumption) or of planned, on-study between-group differences (e.g., amount of omega-3 fatty acid intake) related to events other than the intervention/exposure (e.g., stressors, which can alter the severity of the disorder in addition to the patterns of eating, smoking and alcohol consumption), that is, in variables with the potential to affect mental health outcomes (and biomarker levels), could either “mask” or incorrectly “reveal” clinical benefits of the intervention depending on the groups in which these unexpected changes occurred. Then, unless statistical adjustments are made, such a scenario will complicate the meaningful interpretation of outcomes.

These informal assessment items were modified to assess single group studies since on-study changes involving the same key variables can also complicate the interpretation of their study results. However, no quality scores were derived from the data abstractors' responses to these questions pertaining to controlled or uncontrolled studies.

Study Applicability

As specified in the scope of work for this series of evidence reports on the health benefits of omega-3 fatty acids, the primary focus is on the US population. Given the geographical location of the UO-EPC, however, the definition of study applicability was expanded slightly to include Canada as part of a larger North American context. This study's reference point became the “typical” North American.

Also known as external validity, or generalizability, the construct of applicability refers to the degree to which a given study's sample population is sufficiently representative of the population to which one wishes to generalize its results. In the present review, two schemes operationally defined applicability (Appendix C**). One assessed studies involving at least one target population identified with a psychiatric disorder or condition, with the other evaluating studies involving a target population with or without a known elevated risk for a psychiatric disorder or condition.

With regards to the highest level of applicability (Level I) in the first scheme, the broadest definition of the population of interest is the otherwise “healthy” North American (or similar individual) identified with a psychiatric disorder or condition, diagnosed using a standard North American strategy and methodology/nomenclature (e.g., DSM-IV) or identified using at least one established psychiatric instrument, presenting with or without comorbid psychiatric conditions while possibly receiving “typical” North American medications for the primary diagnosis, is drawn from a somewhat broad socio-demographic spectrum (i.e., gender, race), and eats a diet “typical” of a broad spectrum North American population (e.g., with an estimated omega-6/omega-3 intake ratio of at least 15: see below for references). For Level I applicability in the second scheme, the broadest definition of the population of interest is the otherwise “healthy” North American (or similar) individual, presenting with or without a known elevated risk for onset of a psychiatric disorder or condition, representing a somewhat broad socio-demographic spectrum (i.e., gender, race), and eating a diet “typical” of a broad spectrum North American population (e.g., with an estimated omega-6/omega-3 intake ratio of at least 15).

Together, these level I definitions represent the respective reference points, with applicability decreasing as the definition of the sample study population narrows in terms of the factors represented in the two schemes. With respect to the scheme applied to studies with diagnosed participants, we identified what are likely the two most important variables as being the method of diagnosis and the background diet of participants leading up to the study, if not also during the study. Each defines the study population. When the second scheme is applied to studies where the participants have not yet been diagnosed with a psychiatric disorder or condition, background diet is the key variable.

The method of diagnosis is an important factor since not all countries employ the diagnostic methods or nomenclatures used most frequently in North America (e.g., DSM). Psychiatric populations identified using different approaches, even when diagnostic labels are the same, can vary in terms of what these labels refer to.¹³ At the same time, different labels (e.g., “AD/HD” in North America versus “hyperactivity” in the UK) can refer to the same clinical entity. That said, the most frequently employed approaches employed in North America are considered the reference point.

Operationalized ideally in this review as the omega-6/omega-3 fatty acid ratio, background diet is an important factor in assessing both types of study population (i.e., diagnosed vs undiagnosed participants). Given the competitive relationship between omega-3 and omega-6 fatty acids, both for enzymes to yield key metabolites with specific effects in the human biosystem (see Chapter 1) and for positions in cell membranes from which to have these and other possible influences (e.g., clinical improvement or prevention), the absolute and relative intake of omega-3 and omega-6 fatty acids from all sources, and not just from the identified exposure, likely need to be taken into account when deciding whether populations assessed in different studies are comparable. The likelihood of biological and/or clinical effects in studies may turn out to vary depending on these absolute or relative intake values. A high background dietary omega-6/omega-3 fatty acid intake ratio—potentially reflected in a corresponding differential in these contents in cell membranes—may make it harder for omega-3 fatty acid supplementation to make a clinically meaningful difference,⁷² although already having considerable omega-3 fatty acid content in the background diet and in cell membranes because of a low omega-6/omega-3 fatty acid intake ratio may make it difficult for typically small amounts of omega-3 fatty acid supplementation to make a clinically meaningful difference (see Discussion).

Irrespective of which of these hypotheses may be eventually confirmed elsewhere, the fact that national, and sometimes regional, populations can vary in terms of their diet's omega-6/omega-3 fatty acid intake ratio strongly suggests that this potential confounding influence on study outcomes needs to be represented in the applicability schemes whereby the North American value is the reference point. The typical North American diet contains an omega-6/omega-3 fatty acid intake ratio of at least 15, while urban India and Japan's corresponding values are 38–50 and 4, respectively.^33–45

UK populations represent somewhat of a special case in that, while they often use the same diagnostic methods and research instruments to identify psychiatric disorders and conditions in populations, respectively, and while they can exhibit socio-demographic pictures similarly broad to the ones seen in North American study populations, their somewhat different lifestyle and background diet recommended an applicability value of “II.” However, if participants were drawn from a narrower UK population, then a “III” was assigned. Given their inclusion of multi-national populations, with or without representation from the U.S. or Canada, cross-national ecological analyses necessarily received a “III.” One experienced assessor evaluated study applicability.

Summary Matrix

For a given research question, and where possible (e.g., more than one study addressing the question), a summary matrix situates the pertinent studies in terms of their respective study quality (internal validity) and applicability (external validity) values. The Jadad total quality score defined RCTs' internal validity in summary matrices. A three-level format was derived from the range of possible RCT quality scores (A = Jadad total score of 4 or 5; B = Jadad total score of 3; C = Jadad total score of 0, 1 or 2). Given that allocation concealment scores have in the past tended to vary less widely than Jadad total scores, allocation concealment values were entered as superscripts in the summary matrices.⁷² A similar approach was taken for the studies employing other research designs. The following cutpoints were established, albeit without benefit of a validational exercise:

• comparative before-after study: A = total quality score of 8–11; B = 5–7; C = 1–4;

• case-control study (NOS): A = 8–10; B = 4–7; C = 1–3;

• (multiple-group) cross-sectional study: A = 8–10; B = 4–7; C = 1–3;

• single prospective cohort study (Modified NOS): A = 8–10; B = 4–7; C = 1–3;

• cross-sectional survey: A = 8–10; B = 4–7; C = 1–3; and,

• cross-national ecological analysis: A = 7–9; B = 4–6; C = 1–3.

The three-level applicability format was established by the 3 EPCs involved in the 2-year project for practical reasons, to permit the incorporation of quality scores within a summary matrix. Studies assigned an “X” (i.e., insufficient information to establish applicability) were excluded from summary matrices.

Qualitative Data Synthesis

An overarching qualitative synthesis describes the progress of each citation, then report, through the stages of the systematic review. It also highlights certain report and study design characteristics of included studies (e.g., distributions of research design by research question). Then, for each question, a separate qualitative synthesis is derived for included evidence, organized by broad categories of research design (i.e., experimental studies vs observational studies vs cross-national analyses). A brief study-by-study overview typically introduces the synthesis, followed by a narrative summary of the key defining features of relevant studies (e.g., inclusion/exclusion criteria), including their populations (e.g., diagnosis-related), intervention/exposures (e.g., types of omega-3 fatty acid), cointerventions (e.g., psychotropic medication), outcomes, study quality, applicability and results. Whether or not data can be organized according to these subheadings depends on the number of studies addressing a given question and the amount or variety of detail available in the study reports. For example, having identified too few studies per research question that do and do not exhibit significant effects for a given clinical outcome can preclude determining the impact of covariables with the potential to modify or confound study results (e.g., type or dose of omega-3 fatty acids).

Juxtaposing, in turn, all pertinent studies' parameters for a given research question has two key consequences. It allows us to identify the “gaps” in knowledge deemed crucial by content experts to understand the clinical phenomenon (e.g., efficacy of omega-3 fatty acids). That is, data regarding possible confounders may be lacking, making it difficult to interpret study results with unfettered confidence. These gaps point to those variables requiring measurement and experimental or statistical control in future research. Second, it affords an understanding of the definition and extent of the included studies' clinical homogeneity (i.e., population, intervention, cointervention, outcome), which can then inform decisions regarding the appropriateness of meta-analysis. Where strong clinical heterogeneity is observed, it may be important to forego meta-analysis because the “population” to which any point estimate, and measure of precision, might be extrapolated may not exist per se; it, too, is synthetic (e.g., the “average” schizophrenic). Subject to scrutiny in the evaluation of cross-study clinical homogeneity is the ability of each study to control for confounding influences and yield results that can be interpreted without serious question marks. The existence of statistical heterogeneity also plays a role in the decision to do without a quantitative synthesis. Whether or not meta-analysis is considered appropriate, an attempt is made to make sense of the possible influence of covariates and confounders within the context of the qualitative synthesis.

Where eligibility criteria permit, evidence from research designs with a lesser inherent potential to control for biasing influences are used to see whether, collectively, they confirm the picture of efficacy, or association, derived from designs with a greater inherent potential to achieve this goal (see Eligibility Criteria). For the purposes of interpreting results, greater emphasis is placed on the latter, with “greater emphasis” meaning that we assign greater interpretative, not numerical or statistical, weight to these intrinsically stronger designs. Factors other than study design also taken into account in interpreting results include study quality, the number of studies, and whether studies were sufficiently powered.

Quantitative Data Synthesis

Given its greater potential to control for possible confounding factors, only RCT evidence regarding the question of interventions' efficaciousness was considered for inclusion in meta-analysis. All things being equal, it was also assumed that priority in meta-analysis might be given to clinical outcomes pertinent to the present day practice of psychiatry and psychology.

Providing the result would have a clearly defined population to which to generalize a synthetic result, and that sufficient numbers of prospective controlled studies exist (e.g., RCT; cohort study), meta-analysis was considered with respect to data investigating questions of the possible association of the onset, continuation or recurrence of a psychiatric disorder or condition with either the intake of omega-3 fatty acids or the PUFA content of biomarkers. Prospective controlled designs constitute the most appropriate way to establish these risk-relationships among variables. Decisions regarding statistical models are provided where results of meta-analysis are reported. Reasons to forego meta-analysis are likewise described.

Overview of Relevant Study's Characteristics and Results

Likely at one US site, Marangell et al. randomized 36 adult outpatients (18–65 years; racial/ethnic background unreported) meeting DSM-IV criteria for major depressive disorder (duration unreported), without psychotic features, to receive either 2 grams per day (2 g/d) DHA or placebo (source undefined) in a 6-week parallel design (followups at 2 and 6 weeks).⁹⁵ Inclusion criteria were a score of at least 12 on the Montgomery-Asberg Depression Rating Scale (MADRS), a score of at least 17 on the Hamilton Depression Rating Scale (HDRS), no psychotropic medication for at least 2 weeks, and dietary intake of no more than one fish serving per week. Exclusion criteria included any significant comorbid psychiatric or medical conditions, and a lifetime failure of at least two adequate antidepressant trials. Clinical response was the primary outcome, and was defined as a mimimum 50% reduction, from baseline to 6 weeks, on the MADRS. Funding was provided by way of an investigator-initiated grant from Martek Biosciences Corporation.

Response rates were 27.8% and 23.5% in the DHA (n=18 with at least one followup) and placebo groups (n=17), respectively, with the difference failing to reach statistical significance in an intention-to-treat analysis (ITT). This null finding held after an adjustment for baseline HDRS score. The two groups did not vary in terms of age, percent male participants, alcohol intake, or education. The placebo group comprised a significantly greater number of smokers and a lower weight. Both at baseline and at study endpoint the placebo group exhibited a significantly higher HDRS score. Only in the DHA group did the absolute level of RBC DHA content increase in statistically significant fashion from baseline to endpoint, whereas a report of a similar difference in the change in DHA's percent weight of total fatty acids was not accompanied by results of a statistical test of significance. No information was provided regarding the reason one participant in the placebo group did not reach final followup.

A summary matrix is not required for a single study. Study quality assessed via the Jadad total score was low, with insufficient clear information preventing us from concluding that the allocation to study groups had likely been adequately concealed. There were insufficient details reported by Marengell et al. to permit the determination of a level of applicability even though the trial appeared to have been conducted in the US.

This was a single study with a limited sample size and a limited complexity to its design (e.g., no stratification for covariates). Thus, other than the observation that the possible confounding impacts of certain factors (e.g., between-group differences or on-study changes in psychotropic medication type or dose; alcohol intake; education, age, sex) were likely controlled in this primary treatment study, little can be said about the possible impact of additional factors with the potential to influence mental health outcomes. Yet, one factor with the potential to influence these outcomes, current smoker status, was not distributed equally across study groups although the observation that more placebo group members were smokers makes it difficult to see how this may have contributed to a null between-group difference in the primary clinical outcome. This between-group difference could have influenced the observations of a between-group difference in changes in RBC DHA content, however, given the effects of smoking on EFA status.⁶⁰ The restriction on weekly fish intake likely made study groups somewhat more comparable. Meta-analysis was considered unnecessary.

Overview of Relevant Studies

Peet and Horrobin conducted a dose-ranging study of the effects of ethyl eicosapentaenoate (E-EPA: i.e., a pure ethyl ester derivative of EPA) in adult outpatients (n=70; 18–70 years) identified with persistent depressive symptomatology despite ongoing treatment with an adequate dose (undefined) of a standard antidepressant (Summary Table 2).⁵³ Recruited by family physicians, study participants were randomized into a 12-week parallel RCT (followups at 4, 8 and 12 weeks) on a double-blind basis to receive either placebo (liquid paraffin) or total doses of 1 g/d, 2 g/d or 4 g/d E-EPA via 500 mg soft gelatin capsules (taken morning and evening). The primary outcome was HDRS score, with the MADRS and the patient-completed Beck Depression Inventory (BDI) serving as secondary outcome measures.

Nemets et al. randomized 20 Israeli outpatients (mean age: 53.4 [28–73] years), meeting DSM-IV criteria for a current diagnosis of major depressive disorder, to receive either 2 g/d E-EPA derived from 96% pure fish oil (stabilized with 0.2% vitamin E) or matching placebo (undefined) given in 1 g doses twice daily (via 50 0mg soft gelatin capsules) for 4 weeks.⁹⁷ Only one patient did not continue receiving the antidepressant treatment they had been taking for at least 3 months, making this male's trial an evaluation of the impact of E-EPA as a primary treatment. He was exhibiting a 4-month severe depressive disorder that had been resistant to treatment with two different SSRIs. All other study participants had, in the past, suffered relapses when antidepressant doses were reduced or discontinued altogether. The primary outcome measure was the HDRS, with ratings conducted at baseline and weekly thereafter in this double-blind trial.

Su et al. conducted an eight-week, double-blind, placebo-controlled parallel RCT.⁹⁶ They compared the impact, on HDRS scores, of 6.6 g/d of omega-3 fatty acids (i.e., 4.4 g/d EPA and 2.2 g/d DHA from menhaden fish) against placebo (i.e., olive oil ethyl ester) in 28 physically healthy outpatients diagnosed with DSM-IV major depressive disorder. Five identical gelatin capsules containing 440 mg EPA and 220 mg DHA were taken twice daily. Inclusion criteria were an HDRS score of at least 18, and no change in medication or psychotherapy 4 weeks prior to enrolment. Participants could not exhibit comorbid Axis I or Axis II psychiatric disorders, or be receiving antipsychotics or mood stabilizers. Placebo responders (i.e., mimimum 20% decrease in HDRS score) during a pre-randomization, one-week run-in were excluded. One participant in each group was free of medication, indicating that their trials assessed the primary treatment of depression. Followups using the HDRS occurred every 2 weeks. Dietary frequency ratings, recorded food diary data, and blood samples to assess the fatty acid content of RBC membranes were assessed during the run-in and at 8 weeks.

Qualitative Synthesis of Relevant Studies' Key Characteristics

Study characteristics. Three parallel RCTs involving adults addressed the question (Summary Table 2; Evidence Table 1: Appendix E).^{53, 96, 97} Only Su et al.⁹⁶ and Nemets et al.⁹⁷ provided detailed descriptions of both inclusion and exclusion criteria. Only Peet and Horrobin⁵³ employed a design having more than two study groups (i.e., 4). A total of 118 adult outpatients were randomized. The mean sample size for the three studies was 39.3 (range: 20–70) participants, with the Peet and Horrobin trial being much larger than either of the other two. The studies' participants received the intervention for an average of eight (range: 4–12) weeks, with Peet and Horrobin's intervention period lasting the longest. The RCTs were conducted in three countries outside North America: the UK,⁵³ Israel,⁹⁷ and China.⁹⁶ The UK RCT was funded by industry (Laxdale Research Ltd),⁵³ the study from China was funded by government (National Science Council) and industry (China Chemical & Pharmaceutical Company),⁹⁶ and the Israeli trial's funding source was not reported.⁹⁷

Population characteristics. The mean age of study participants across the three trials was impossible to determine given that full sample means were not given for two trials.^{53, 96} Participants' ages ranged from 18–73 years when two studies' data were combined.^{53, 97} Participants in the Su et al. study tended to be younger (mean for omega-3 group=35.2 years; placebo group: 42.3 years), on average, than participants in the Peet et al. (means of four groups: 43–48 years)⁵³ or Nemets et al. studies (omega-3 group mean: 54.2 years; placebo mean: 52.1).⁹⁷ Females were consistently more strongly represented in the three trials (82–85%). Racial/ethnic backgrounds included Asian⁹⁶ and Middle Eastern,⁹⁷ yet no data were provided for a potentially diverse UK population.⁵³

Only the Peet and Horrobin RCT⁵³ did not report having employed any formal diagnostic criteria. The other two studies used DSM-IV to identify populations with major depressive disorder. Rather, the UK study required a score of at least 15 on the HDRS, and thus likely identified their participants as merely experiencing persistent depressive symptomatology. Nemets et al.'s participants had had relapses in the past when medication dosages were reduced or discontinued.⁹⁷ All three studies employed the HDRS score to establish the severity of the psychiatric condition. While Peet and Horrobin required a score over 15, Su et al. required a score above 18.⁹⁶ Nemets et al. only reported mean actual HDRS scores of 22.3 (placebo group) and 24.0 (EPA group).⁹⁷ Based only on completer data subjected to a statistical test, Su et al. noted that HDRS-defined severity was equivalent for study groups.⁹⁶ Peet and Horrobin did not present baseline HDRS severity data for study groups.

Peet and Horrobin⁵³ did not identify or exclude any comorbidity while Nemets et al. required that there be no unstable medical disease, no alcohol or drug abuse, no psychotic features, no history of hypomania or mania, and no comorbid psychiatric diagnosis other than panic disorder (n=2, one per study group), dysthymic disorder (n=2, one per study group), and obsessive compulsive disorder (n=1, E-EPA group).⁹⁷ Su et al. asserted that no one in their sample received any other Axis I or any Axis II psychiatric diagnosis.⁹⁶

Likely because their study participants did not receive a formal diagnosis, Peet and Horrobin did not report data concerning the duration of the current depressive episode, age of onset, the number of previous episodes, or the time since diagnosis.⁵³ Su et al. reported the study groups' mean current episode duration (omega-3 group: 21.5 weeks; placebo group: 22.8 weeks), age of onset (omega-3 group: 30.6 years; placebo group: 35.1 years), number of previous episodes (omega-3 group: 2.5; placebo group: 2.3), but not the time since diagnosis.⁹⁶ Statistical tests of the possible significance of between-group baseline differences exclusively for completers revealed that study groups were comparable on these bases as well as with respect to age, percentage of males, body mass index, HDRS score, and both EPA and DHA levels in RBCs. Nemets et al. reported their sample's mean current episode duration (EPA group: 44.6 days; placebo group: 43.1 days), time since diagnosis (EPA group: 7.6 years; placebo group: 8.0 years), number of previous episodes (EPA group: 2.1; placebo group: 1.9), but not their age of onset.⁹⁷ While statistical tests of significance were not employed, notable between-group differences at baseline were not observed for these variables.

Only Nemets et al. controlled for two of these potential confounding influences by excluding participants if they had had substance abuse or unstable medical problems.⁹⁷ Only Su et al. reported data reflecting the omega-3 fatty acid content of biomarkers at baseline, which by statistical analysis, were comparable between study groups. They did not, however, report the units of measurement for biomarker data (e.g., absolute level; percent of total fatty acids).⁹⁶

Intervention/exposure characteristics. Both Su et al.⁹⁶ and Nemets et al.⁹⁷ identified the source of their intervention as fish oil whereas Peet and Horrobin⁵³ reported no details. Only Su et al. identified the exact type of source: menhaden fish.⁹⁶ Nemets et al. compared 2 g/d E-EPA derived from 96% pure fish oil (stabilized with 0.2% vitamin E) and a matching, albeit undefined placebo.⁹⁷ Peet and Horrobin employed 1 g/d, 2 g/d, 4 g/d or placebo (liquid paraffin) as their intervention.⁵³ Su et al.'s participants received 6.6 g/d of omega-3 fatty acids (i.e., 4.4 g/d EPA and 2.2 g/d DHA) or placebo (i.e., olive oil ethyl ester).⁹⁶ Only Su et al. used DHA in addition to EPA. Each RCT employed a placebo control and used the appropriate numbers of capsule and amounts of placebo content to equalize the total daily “intervention” across their study groups.

Omega-3 fatty acid contents were delivered by capsule in each study. However, there are few clear data to suggest that all three studies were equally able to eliminate the possible confounding influence of having unequal amounts of calories, as energy, provided for their different study groups. Nemets et al.'s placebo was not defined, making it impossible to know whether participants in each study group received the same number of calories. Although it is possible, ultimately it is unclear whether a unit of Peet and Horrobin's liquid paraffin, an inert lubricant laxative, provided the same caloric/energy content as that received from purified EPA. Given that the typical laxative dose is 15–30 g/d, and that Peet and Horrobin's study, as well as others described in this review, have consistently used much smaller daily doses, it is unlikely that its laxative effect would be any worse than that produced by a similar food oil.⁸⁷ Su et al., on the other hand, used olive oil ethyl ester to match their groups for energy/caloric intake.

If, as it was decided in consultation with the TEP working with us on our review of the evidence regarding the effects of omega-3 fatty acids on asthma,⁷² and recognizing the FDA view that a 3 g/d dose of EPA and DHA is safe,¹⁶⁰ then two of the standardized doses in the three studies met our criterion that 3 g/d is a high dose of omega-3 fatty acid supplementation: Su et al's 6.6 g/d EPA plus DHA,⁹⁶ and Peet and Horrobin's 4 g/d E-EPA.⁵³

None of the RCTs provided omega-6 fatty acids or any other supplement as cointervention, and none attempted to implement a specific on-study ratio of omega-6/omega-3 fatty acid intake through diet and/or supplementation. Nemets et al.⁹⁷ and Peet and Horrobin⁵³ did not report whether their study participants were told to maintain their background diet, to alter their background diet in some uniform fashion (e.g., modify omega-6 fatty acid intake and thereby change their omega-6/omega-3 fatty acid intake), or whether participants routinely complied with any such mandates. Only Su et al. established, for example, that study groups did not differ in terms of their on-study dietary frequency of fish intake as reported via 24-hour recall and three-day dietary records.⁹⁶ Few compliance data, in general, were provided. Peet and Horrobin used capsule counts to report that at least 90% of the dose had been consumed in each of the four study groups.⁵³ Even then, this method for determining compliance may not be overly accurate. Moreover, fatty acid content in biomarkers is likely not a perfect methodology either, given that EFA status can be influenced by various factors in addition to intake (e.g., oxidative degradation). Without hard data it is thus difficult to rule out the possibility in at least two studies that notable changes did not occur in the on-study background diet (i.e., Nemets et al., Peet et al.) or that protocol violations with respect to the number of capsules ingested did not occur, leaving unknown the extent of possible confounding with regards to clinical outcomes (i.e., from unplanned changes in the study groups' equivalence of energy/caloric intake from the “exposure” or related to unplanned changes in the between-group difference in the amount of omega-3 fatty acid received from supplementation).

Of the three trials, only Peet and Horrobin⁵³ failed to report having stabilized their omega-3 fatty acid doses with some form of anti-oxidant. Su et al. attempted to maintain blinding by having all capsules vacuumed to deodorize any odour, and having their contents blended with an orange flavor.⁹⁶ Anti-oxidant tertiary butylhydroquinone (0.2 mg/g) and tocopherols (2 mg/g) were added to all capsules both to maintain blinding, by preventing oxidation and rancidity, and to avoid possible confounding that could occur if these were added only to active treatment capsules and actually produced psychotropic effects,. In spite of no effort to deodorize their intervention, Nemets et al.'s participants were unable to reliably guess which capsules they had taken.⁹⁷

For all three RCTs, the manufacturer of the omega-3 intervention was reported. Purity data were provided for two of the trials' exposures.^{53, 97} In the one study that evaluated the fatty acid content of biomarkers, no notable inappropriate methods to extract, prepare, store or analyze lipids were described.⁹⁶ No study report included details as to whether, or how, the presence of methylmercury was tested or eliminated from the omega-3 fatty acid exposure.

Cointervention characteristics. Given the focus of the present question is supplemental treatment, it could be argued that the omega-3 fatty acids are the cointervention. Nevertheless, to simplify matters, “cointervention” is defined as those other treatments or interventions that are provided concurrently, even if their initiation predated the omega-3 fatty acids intervention.

Peet and Horrobin reported similar distributions of background treatment by type of antidepressant (i.e., tricyclics, serotonin selective reuptake inhibitors [SSRIs], and others) in each study group.⁵³ They did not present data regarding whether or not this antidepressant use remained constant, by type or dose, over the study for any of their study groups. Nemets et al. described their participants as having received their antidepressants for at least three weeks at the current dose.⁹⁷ However, antidepressant medication was not distributed equally by type or dose across study groups. There was similar fluoxetine and mirtazapine use and doses, but five placebo and one E-EPA participant received paroxetine, usually at 20 mg/d. As well, the E-EPA study group included the only three users of fluvoxamine and the only recipient of citalopram. Moclobernide was given to a single participant in the placebo group. Participants on prestudy medication in Su et al.'s trial maintained their dosages on-study, with only oral sedatives/hypnotics (loazepam or zolpidem) permitted as additional therapy for possible anxiety or insomnia.⁹⁶ They did report statistically-tested between-group baseline comparability for completers' duration of antidepressant use prior to enrollment or their (fluoxetine equivalent) dose of antidepressants while being enrolled.

Certain population characteristics have the potential to influence mental health outcomes if, in controlled investigations, study groups diverge significantly at baseline on these bases, or if unplanned on-study changes unrelated to the exposure occur in their status that vary notably across study groups (or within a single study group). Some cointerventional factors may exhibit a similar potential to confound clinical outcomes (e.g., psychological interventions, other licit drug use, use of complementary/alternative medicine/products, other supplement use with psychotropic potential). Not reported in the three included RCTs were data regarding the between-group comparability at baseline, or data regarding the on-study change in the status of these factors, making it difficult to rule out the possibility that these variables influenced clinical outcomes.

Outcome characteristics. All three RCTs employed the validated HDRS as the primary outcome.^{53, 96, 97} While Peet used the validated MADRS and BDI as well, Su et al. assessed the omega-3 fatty acid content of biomarkers.

Study quality and applicability. The three RCTs received a mean Jadad total quality score of 3.6, indicating sound internal validity (Summary Matrix 1). The trials conducted by Peet and Horrobin⁵³ and Nemets et al.⁹⁷ each received a score of 4, while Su et al.'s score was 3.⁹⁶ The latter two studies^{96, 97} each received an applicability rating of III, and a II was assigned to Peet and Horrobin's UK trial.⁵³ Overall, these studies' individual or collective results were not readily generalizable to a North American population.

Qualitative Synthesis of Individual Study Results

Peet and Horrobin's trial conducted both ITT (last observation carried forward) and per-protocol (PP) analyses, the latter assessing study completer data.⁵³ Analyses of variance (ANOVA) compared data reflecting change from baseline to study endpoint for each active treatment group with comparable data from participants receiving placebo. All of the ITT and PP analyses involving each of the 3 scales showed that participants in the 1 g/d group improved significantly more than did those in the placebo group. For the 2 g/d and 4 g/d groups no comparison reached a level of statistical significance, with only the 4 g/d results for the PP population approaching statistical significance. For the 1 g/d versus placebo contrast, the HDRS and MADRS differences were already statistically significant at 4 weeks; only the BDI scores failed to show statistically significant changes at 4 weeks. Eight-week data regarding 1g/d supplementation was only provided for the BDI, and changes in these scores at 8 weeks only approached statistical significance. Analyses of specific items from the rating scales (i.e., the three main components of the HDRS [items 1–3: depression; 4–6: sleep; 9–11: anxiety] and the ten MADRS items) demonstrated that, for the comparison involving the PP population, there were no significantly greater improvements in the 1 g/d group compared with the placebo group. Statistically significant differences in favor of the 1 g/d dose were observed on the BDI-defined items pertaining to sadness, pessimism, inability to work, sleep disturbance and libido. These results suggest improvements defined by both patient and clinician assessments. While results of tests of statistical significance were not reported, the number of participants exhibiting a 50% improvement was always higher, when compared with placebo rates, in the 1 g/d and 4 g/d groups for all three scale scores assessed in both the ITT and PP populations. Yet, response rates for placebo participants consistently exceeded those from participants receiving the 2 g/d dose.

Analyses of covariance (ANCOVA) of Peet and Horrobin's data for each of the rating scales at each of the followups assessed overall differences between study groups.⁵³ They revealed that center (exact number unreported) and background medication, by class (i.e., tricyclic, selective serotonin reuptake inhibitor [SSRI], or either norepinephrine or mixed reuptake inhibitors), had no significant effects on any rating scale scores in the ITT and PP populations. Baseline HDRS score had no effect on the HDRS and MADRS outcomes in either of the ITT or PP populations yet had a significant effect on BDI outcome only in the ITT population. Treatment had a significant overall effect on all three rating scale scores for both the ITT and PP populations yet the p-value for the HDRS comparison in the ITT population barely missed indicating statistical significance.

Nemets et al.'s multivariate analysis of covariance (MANCOVA), with baseline HDRS score as covariate, described a statistically significant treatment-by-time interaction in the ITT population (i.e., last value at week three carried forward, n=1).⁹⁷ This observation was maintained after the week three HDRS score from the sole placebo dropout was excluded. Compared to placebo, E-EPA yielded significantly improved HDRS scores at each of weeks 2, 3 and 4. The mean reduction in HDRS score in the E-EPA group (12.4 points) was greater than that in the placebo group (1.6), and was considered clinically meaningful. Only one of ten patients in the placebo group and six of 10 in the E-EPA group achieved a 50% reduction in HDRS score. Item analysis showed that E-EPA influenced core symptoms such as depressed mood, feelings of guilt, feelings of worthlessness and insomnia. The investigators did not remove from any analyses the one study patient who was receiving E-EPA as monotherapy.

Su et al. observed, by week 4, and likewise for weeks 6 and 8, a statistically greater HDRS-defined improvement in the active treatment group.⁹⁶ By repeated measures ANOVA it was found that the rate of reduction in HDRS scores was also significantly greater in the omega-3 fatty acids group. Pre- and post-intervention RBC fatty acid status data were limited, with a significantly increased level of DHA seen at post-treatment for the EPA group (n=7) but not the placebo group (n=6). No statistically significant increases in EPA levels were observed for either study group.

Nemets et al. reported a single study dropout, from the placebo group, by week 3 because of worsening depressive symptoms.⁹⁷ Six of 28 participants dropped out prior to week 8 in Su et al.'s trial; two had been receiving active treatment (one lost to followup, and one lost due to noncompliance), and four were in the placebo group (three lost to followup, one lost due to noncompliance).⁹⁶ Ten participants left Peet and Horrobin's trial,⁵³ with four from the placebo group (one lost to followup, one withdrew consent during study, one violated protocol, one had an adverse event presumed to be unrelated to treatment) and two from each of the E-EPA groups (no data by group: three withdrew consent, one left due to lack of efficacy, one violated protocol, one had gastrointestinal adverse event).

Quantitative Synthesis

We decided it was reasonable to explore the possibility of conducting meta-analysis for this question. HDRS was chosen as the primary outcome measure. We aimed to extract the mean change from baseline in HDRS, together with the standard deviation of this change, for each study group. The goal was to focus on the ITT population. We requested data to afford this analysis from the lead investigators of the Peet and Horrobin⁵³ and the Su et al. trials.⁹⁶ Only the former replied, passing on our request to the company now holding their data. A representative of the company stated they would consider the request yet no further reply was received.

In order to help decide on the possibility and appropriateness of meta-analysis, we created a forest plot of all possible combinable results. Length of follow-up varied notably between studies, so from each study we considered the longest followup data reported in addition to 4 week and 8 week results where they were provided. Su et al.'s study⁹⁶ used capsules containing EPA together with DHA to yield a very high total dose (6.6 g/d). The other two studies^{53, 97} employed capsules exclusively containing E-EPA. The study by Peet et al.⁵³ reported change in HDRS after 12 weeks of treatment for four different study groups (placebo, 1 g/d, 2 g/d, and 4 g/d). Although the standard deviation of change from baseline was not reported, p-values for change from baseline relative to placebo were reported for each dose so that the standard error for each contrast could be inferred. Su et al.⁹⁶ reported mean HDRS scores at baseline and post-treatment, but not the standard deviation in the change from baseline. We were nevertheless able to extract estimates from one of their graphs.

The Peet et al.⁵³ and Nemets et al.⁹⁷ studies reported ITT analyses (using a last observation carried forward strategy). Yet, it was unclear whether Su et al.⁹⁶had also employed an ITT approach. In addition, their data concerning loss to followup at 4 weeks and 8 weeks were unclear.

After a careful appraisal of the estimates and key study parameters, however, it was decided not to conduct meta-analysis. No pooled estimate was derived because of the variations in dose both within and among studies, and in view of variations in the length of followup. It should also be noted that, in the Peet and Horrobin study⁵³ the estimates for the different doses involving placebo shared the same placebo group. As well, Su et al.'s intervention was the only one including DHA in addition to EPA,⁹⁶ as the other trials employed purified forms of E-EPA.^{53, 97} All three RCTs employed different types of placebo. Finally, unlike the other two studies wherein patients had been formally (DSM-IV) diagnosed with major depression, Peet and Horrobin's use of a HDRS cut-off score to identify study participants yielded, at worst, a population with persistent depressive symptomatology.⁵³

Impact of Covariates and Confounders

Overall, the Su et al. study was the one exhibiting the best control of extra-interventional factors with the potential to influence, and thus confound, study results.⁹⁶ Without repeating all of the details presented in the qualitative synthesis, these investigators indicated that study groups were balanced for key population (e.g., severity of depression, age of onset, absence of other Axis I or Axis II disorders) and cointervention parameters (e.g., patients asked to maintain constant on-study medication, although they did not demonstrate between-group baseline comparability for types and doses; established a maximum weekly frequency of background fish intake). Although Peet et al. provided few population data, their undiagnosed sample did not exhibit comorbid conditions and their study groups' patterns of medication use were similar.⁵³ Nemets et al., on the other hand, did not evaluate whether study groups of depressed patients were similar at baseline in terms of the severity of their depressive symptomatology. They also reported that study groups varied in terms of their antidepressants. Still, their groups did not contain any individuals with unstable medical disease or substance abuse, and few comorbid conditions were observed. None of the studies reported data concerning prestudy/baseline omega-3 or omega-6/omega-3 fatty acid intake via diet or supplementation. Only Su et al. reported data regarding study groups' baseline comparability in their baseline omega-3 fatty acid content of biomarkers.⁹⁶ They did not, however, report the units of measurement for these data (e.g., absolute level; percent of total fatty acids).

Dose, omega-3 fatty acid type, and whether the exposure was purified all failed to reliably predict clinical effects. For example, significant effects were associated with the largest (6.6 g/d EPA+DHA)⁹⁶ and smallest doses (1 g/d E-EPA),⁵³ various types of omega-3 fatty acid (EPA+DHA⁹⁶ vs E-EPA^{53, 97}) and both EPA+DHA⁹⁶ and E-EPA.^{53, 97} Employed as a possible surrogate measure of background diet, or possibly even the background diet's omega-6/omega-3 intake ratio, the country in which a study was conducted did not predict study results. The lowest dose (1 g/d E-EPA), given to a UK population,⁵³ and the highest dose (6.6 g/d EPA+DHA), given to a Chinese population,⁹⁶ each yielded a significant clinical effect. The majority of study participants were female. Overall, though, there were too few studies with which to properly evaluate the impact of extra-interventional variables with the potential to influence study results.

Overview of Relevant Studies

Three RCTs,^98–100 six observational studies^{48, 80, 81, 107, 110, 111} and three cross-national ecological analyses^{47, 108, 109} were found to address this question.

Plasma DHA gradually decreases during the last trimester of pregnancy and remains low for some time during the postpartum period, and particularly in lactating women.¹⁶¹ It has been postulated that brain DHA levels may be low during late pregnancy and the early postpartum period, and that these levels may contribute to the development of postpartum depression.¹⁶² Postpartum depression is defined in DSM-IV as a major depressive, manic, or mixed episode in major depressive disorder, bipolar I or bipolar II disorder, or brief psychotic disorder. Llorente et al. thus attempted to determine the effect of DHA supplementation on the onset of postpartum depression as well as on plasma phospholipid DHA content in breastfeeding women (Summary Table 3).⁹⁸ Mothers who planned to breastfeed their children (n=138; 18–42 years) were randomly assigned, in double-blind fashion, to receive either ~200 mg/d DHA or placebo (undefined) for the first four months after delivery. Clinical outcome was determined via the BDI, and was collected at baseline, 3 weeks, 2 months, and 4 months post-delivery. Depression-related data were obtained through the Structured Clinical Interview, DSM-IV, Axis I Disorders, Clinical Version (SCID-CV). As well, scores on the Edinburgh Postnatal Depression Scale (EPDS) of postpartum depression symptoms were obtained from subgroups of the sample. Plasma phospholipid data were collected just before delivery and at 4 months.

Wardle et al.'s RCT investigated whether cholesterol-lowering diets influence mood, including depression, anxiety, anger/hostility, stress, and general psychological well-being.⁹⁹ Adult volunteers (n=176) with elevated serum cholesterol levels (>5.2 mM [198 mg/dL]) were allocated to a low-fat diet (n=59), a Mediterranean diet (n=61) or a waiting-list control (n=56). Dietary treatments were given in eight sessions over the 12-week period. Waiting-list controls were offered treatment at the end of their waiting period. Participants were exhibiting at least mild hypercholesterolemia by UK standards. Participants completed a 7-day dietary intake diary before the first assessment. Outcomes were assessed at baseline, 6 weeks and 12 weeks. These included the BDI, personal history of depression established through interview, and the following validated instruments: State-Trait Anger Inventory (STAI), the anxiety and anger subscales of the Profile of Mood States (POMS), the General Health Questionnaire (GHQ) to assess general psychological well-being, and the Perceived Stress Scale (PSS). Dietary diaries were filled out at baseline and 12 weeks.

Reflecting one factor of a factorial RCT investigating interventions to reduce mortality in angina (including: advice [not] to eat fruits and vegetables; [no] stress management), 452 males were allocated to receive advice to eat more fatty fish (i.e., mackerel, herring, kipper, pilchard, sardine, salmon, trout) or to receive no such advice. Study participants were supplied with MaxEPA® fish oil capsules if they did not like the taste of fish.¹⁰⁰ Fish intake and mood (depression, anxiety) were assessed at baseline and at 6 months, the latter using the validated Derogatis Stress Profile (DSP).

In a recently published observational study, Hakkarainen et al. investigated the relationship between the dietary intake of omega-3 fatty acids and low mood, major depression, and suicide in males 50 to 69 years of age living in southwestern Finland in 1985 (Summary Table 4).¹¹¹ The study identified a cohort (n=29,133) from a primary prevention RCT (ATBC Cancer Prevention Study). Followup lasted 9 years. The intake of fatty acids and fish consumption were derived from a validated food use questionnaire focused on the “last 12 months.” Self-reported depressed mood, suicides and hospital-based treatments for major depressive disorder were evaluated.

Tanskanen and colleagues undertook two non-overlapping cross-sectional surveys in Finland.^{80, 81} Both were published in 2001. In a random sample of 3,204 Finnish adults (25–64 years), depressive symptomatology was measured using the BDI.⁸¹ A single food-frequency question assessed fish consumption (fish type unspecified) regarding the previous 6 months. For this study, the sample was drawn from two coastal and two lakeside areas in 1992 (n=8,000). After health questionnaires were returned, and medical examinations completed, a random sample of participants was selected based on birthdays between the twelfth and the last day of each month (n=5,105). Following other clinical measurements, this group was given a questionnaire, which included psychosocial variables. The response rate was 67% (n=3,403), while another 199 individuals did not provide complete data sets. In all, 3,204 subjects became the study sample.

In Tanskanen et al.'s other study a sample was selected, in 1999, based on a random population sample (National Population Register) of both sexes (n=3,004; 25–64 years).⁸⁰ The number of respondents was 1,767 (59%), and they resided in Kuopio in the central-eastern part of Finland (lakeside area). Data were gathered on fish consumption, depression (BDI) and suicidality. The latter was measured using a single BDI item.

Three additional studies involved more specific definitions of population in evaluating the possible relationship between omega-3 fatty acid intake and the risk of geriatric depression (Summary Table 5). In 1991-1992, Woo et al. conducted a single cohort, 3-year prospective study examining the possible relationship of physical activity, dietary habits (e.g., fish consumption), smoking and alcohol consumption with three-year mortality as well as other health outcomes.¹¹⁰ Participants included 2,032 elderly Chinese subjects (mean age: 80 years) recruited by stratified (by age: e.g., 80–84 vs 85–89 vs >90 years) proportional random sampling.

In a cross-sectional study examining the possible association of omega-3 fatty acid intake and the prevalence of depression in 902 Japanese individuals newly diagnosed with primary lung cancer, Suzuki et al. employed a food frequency questionnaire and the depression subscale from the validated Hospital Anxiety and Depression Scale (HADS).¹⁰⁷ Data from 771 patients were analyzed after excluding those failing to complete the HADS (n=73) or the food frequency questionnaire (n=62), or those having incorrectly completed the latter (n=24).

Edwards et al. measured the dietary PUFA intake as well as the fatty acid content of RBCs in a cross-sectional study of ten depressed patients and fourteen matched healthy control subjects.⁴⁸ Biomarker results are described in a later section although the key study parameters are presented in relation to the current research question concerning the possible association of omega-3 fatty acid intake and depression. Analyses controlled for stress and smoking status.

Cross-national, ecological analyses can highlight evidence concerning the possible relationship between intake of omega-3 fatty acids and risk of depression in spite of certain inherent limitations of these data (see Discussion). Hibbeln ⁴⁷ utilized cross-national epidemiology data from eight (of the ten) countries in Weissman et al.'s study regarding major depression (and bipolar disorder),⁴⁶ to which they added prevalence data from Japan (Summary Table 6).⁴⁷ Weissman et al.'s study had evaluated 35,000 participants using a random prospective design, repeat sampling techniques, multiple community sampling, and a structured interview process with accepted diagnostic criteria.⁴⁶ Apparent fish consumption was estimated; it is an economic measure of disappearance of seafood from the economy.¹⁰⁸

In a second ecological analysis, Hibbeln assessed the interrelationships among seafood consumption, the DHA content of mothers' milk, and prevalence rates of postpartum depression (n=14,532 subjects in 41 studies).¹⁰⁸ To maximize comparability the investigator identified only published prevalence data for postpartum depression that had used the EPDS, and correlated these data with those indicating the EPA, DHA and AA content in mother's milk as well as published seafood consumption rates from 23 countries.

Peet's cross-national ecological analysis focused on international variations in the prevalence of depression and the outcome of schizophrenia, and their possible prediction by patterns of omega-3 fatty acid intake.¹⁰⁹ Data on food use were taken from the FAOSTAT database, and reflected apparent national food consumption.¹⁶³ Data on depression prevalence were again borrowed from Weissman et al.⁴⁶ and the same Japanese source used by Hibbeln.⁴⁷ Two-year outcome data relating to schizophrenia were drawn from the WHO's International Pilot Study of Schizophrenia (IPSS).¹⁶⁴ A second source of schizophrenia outcome data was the Determinants of Outcome of Severe Mental Disorders (DOSMED) study.¹⁶⁵ Schizophrenia results are presented later in this report.

Qualitative Synthesis of Relevant Studies' Key Characteristics

Study characteristics. One RCT employed a parallel design with two study arms,⁹⁸ a second included three study groups,⁹⁹ and the data from the third study came from one factor of a factorial RCT design (Summary Table 3; Evidence Table 1: Appendix E†††).¹⁰⁰ In one study, the focus was on the possible utility of omega-3 fatty acids to affect the likelihood or intensity of mood changes in a population at risk for postpartum depression.⁹⁸ In the other two studies, the intervention given for a medical disorder conveniently allowed the investigators to examine the possible relationship between omega-3 fatty acid intake and mood.^{99, 100} These two studies' inclusion and exclusion criteria therefore pertained to the primary reasons these narrowly defined populations were studied in the first place: adults with elevated serum cholesterol levels, and for whom one of their cholesterol-lowering treatments was thought to have the potential to influence mood;⁹⁹ and, males with angina, whose “fish advice” intervention was also thought to have the potential to affect mood.¹⁰⁰

The populations from the latter two studies did not include individuals with formal diagnoses of depression.^{99, 100} Given the heterogeneous nature of the populations, it made little sense to synthesize many of the study characteristics (e.g., mean sample size). The interventions lasted an average of 18.4 (range: 12–26) weeks. Two of the studies were conducted in the UK^{99, 100} and a third in the US.⁹⁸ Llorente et al.'s study was supported by industry (Martek Biosciences Corporation),⁹⁸ Ness et al.'s by the UK Medical Research Council,¹⁰⁰ and Wardle et al.'s by government as well (Biotechnology and Biosciences Research Council).⁹⁹

Three of the included observational studies were conducted in Finland.^{80, 81, 111} Inclusion/exclusion criteria were published elsewhere for Hakkarainen et al.'s study,¹¹¹ while eligibility criteria were delineated in each of Tanskanen et al.'s reports.^{80, 81} None of the study reports made reference to a funding source.

Woo et al.'s single prospective cohort was well-defined.¹¹⁰ Clearly delineated eligibility criteria regarding the cancer diagnosis were included in Suzuki et al.'s report.¹⁰⁷ Their survey was filled out both prior to and during hospital admission. Edwards et al. reported well-defined exclusion criteria.⁴⁸ Neither Woo et al.¹¹⁰ nor Edwards et al.⁴⁸ identified their funding source(s). Funding for Suzuki et al.'s study was received as a Grant-in-Aid for Cancer Research and Second-Term Comprehensive Ten-Year Strategy for Cancer Control and Research of the Japanese Ministry of Health, Labour, and Welfare.¹⁰⁷

Eligibility criteria were sparse in Hibbeln's first report of a cross-national ecological analysis⁴⁷ and were explicitly stated in their second one.¹⁰⁸ Peet's descriptions of eligibility criteria were clear.¹⁰⁹ Neither Hibbeln⁴⁷ nor Peet¹⁰⁹ reported their funding source. Hibbeln's second study was funded in part by a Young Investigators Grant from the National Association for Research on Schizophrenia and Depression (NARSAD).¹⁰⁸

Population characteristics. Given the heterogeneous nature of the included studies' populations, again it made little sense to synthesize some of the population characteristics from the three RCTs (e.g., mean age, mean percent males). Racial/ethnic backgrounds included a potentially diverse UK sample,⁹⁹ a predominantly Caucasian (82%) US population,⁹⁸ and a likely Caucasian/European one.¹⁰⁰

In Llorente et al.'s trial, women intending to breast-feed were excluded if they exhibited a chronic medical condition, were current smokers, or had been pregnant more than five times.⁹⁸ No significant differences were observed for baseline BDI scores for depression. As well, there were no mean differences between study groups for mother's age, education, racial/ethnic distibution, and several pregnancy/delivery/infant-related factors (i.e., parity, gravidity, delivery weight, prepregnancy weight, gestational ages of infants, infants' birth weight, sex distibutuion of births, Apgar scores at 1 or 5 minutes).

In Wardle et al.'s study, adults with elevated serum cholesterol levels (>5.2 mM [198 mg/dL]) were excluded from the RCT if they were pregnant, lactating or planning pregnancy.⁹⁹ There were no significant differences among the study groups for any of the baseline mental health (i.e., BDI; depression, anxiety or anger scores on POMS; general psychological well-being; stress; state anger and anger reactions scores on STAI), background diet (i.e., g/d or percent of energy saturated fat; g/d fiber), or other characteristics (i.e., age, marital status, sex, BMI, total, HDL, LDL cholesterol and triglyceride parameters). Seven-day diary data showed that reported energy intakes were reasonable for adults this age.

No statistical differences were observed with respect to the following baseline characteristics in Ness et al.'s RCT of adult males with angina: depression score, anxiety score, eicosapentaenoic acid (mg/d: measurment undefined), social class, past history of cardiovascular disease, smoking, fish intake, and fruit and vegetable intake.¹⁰⁰ Data regarding the baseline between-group comparability, or on-study change, with respect to other factors with the potential to influence mental health were not reported for the three studies (e.g., stressors, social support).^98–100 Likewise, data regarding the baseline comparability, or on-study change, with respect to key dietary characteristics (e.g., omega-3 or omega-6/omega-3 fatty acid content of the baseline or on-study diets) were not provided by study authors. For example, whether study groups' caloric/energy intake was equivalent in any of the studies could not be determined from reports.

The focus in Hakkarainen et al.'s observational study was on the mood experienced in the 4 months prior to their previous study visit.¹¹¹ Mood difficulties ranged from 5 to 8 years in duration (median=6). Alcohol consumption was also assessed via validated food-use questionnaire. Various covariates were entered into data analysis (see below). In each of Tanskanen et al.'s observational studies, the nature of the sampling meant that individuals varied in terms of their levels of depressive symptom, or even their thoughts of harming themselves.^{80, 81} Woo et al.'s population were elderly Chinese potentially experiencing depressive symptomatology¹¹⁰ while Suzuki et al.'s population of newly diagnosed primary lung cancer patients included 436 (of 771: 56.5%) with analyzable data, and who were exhibiting depressive symptomatology.¹⁰⁷ Edwards et al. identified ten individuals with a major depressive episode using DSM-IV criteria. Each was receiving antidepressant medication. Exclusion citeria included physical illness of a severity or nature suggestive of low omega-3 fatty acid levels. The 14 healthy controls showed no history of psychiatric disorder although how this was determined was not reported. Matching was based on age, sex, social class, BMI, number of children, recent life events, smoking habits, and alcohol consumption. An assessment of these data revealed the soundness of the matching strategy. All study participants were evaluated using the BDI.

Hibbeln's cross-national ecological analysis⁴⁷ identified its populations by using Weissman et al.'s cross-national prevalence study, which included a structured clinical interview employing accepted diagnostic criteria (DSM-III).⁴⁶ The core symptoms of major depression, and not merely symptom severity as reflected in rating scale scores, were identified. Hibbeln drew data on the annual prevalence of major depression in Japan from the Ministry of Welfare (n=130,000). However, one limitation of these Japanese data is that they were not produced using a structured diagnostic instrument or random population sampling techniques.

In his second analysis Hibbeln's populations were drawn from countries varying in terms of their background diet.¹⁰⁸ A high score indicated severe depressive symptomatology rather than a major depression. Prevalence data were derived from well-defined populations. Peet's cross-national ecological analysis¹⁰⁹ included data on depression prevalence once again borrowed from Weissman et al.⁴⁶ and the same Japanese source used in Hibbeln's first analysis.⁴⁷ Two-year data concerning the outcome of schizophrenia were drawn from the WHO's International Pilot Study of Schizophrenia (IPSS). A second source was the Determinants of Outcome of Severe Mental Disorders (DOSMED) study.

Intervention/exposure characteristics. Given the great divergence of interventions or exposures, it likely makes little sense to synthesize many of the intervention/exposure characteristics (e.g., mean dose or serving size; number of studies utilizing a “high” dose or serving size). Two of the RCTs encouraged specific dietary patterns^{99, 100} while the third provided supplementation capsules.⁹⁸ Regarding the latter investigation, Llorente et al. provided ~200mg/d DHA from algae-derived triglyceride capsules yet did not describe the number of capsules constituting a “dose” or what the placebo capsules contained.⁹⁸ Wardle et al.'s RCT allocated adults with elevated serum cholesterol levels either to a low-fat diet (i.e., reduce energy from [saturated] fats to <20%, and ingest mostly polyunsaturates), a Mediterranean diet (i.e., increase intake of fruits and vegetables, oily fish; reduce fat to 30% of calories; use monounsaturated fats instead of saturated fats) or a waiting-list control (i.e., no advice given yet not discouraged from making dietary changes). This entailed educating participants about recommended dietary changes and included a cognitive-behavioral intervention focused on implementing changes in eating behavior. Participants were also given spreadable fat and oils consistent with their assigned diet. Finally, Ness et al. observed that, at 6 months, of the men allocated to fish advice, 78% were consuming fish weekly or taking fish oil capsules (21%), as compared to 14% of those who did not receive fish advice. These details, in addition to the observation that compliance data were not always available, raise some doubt that participants in the different study groups in each RCT actually received a constant difference in their amount of omega-3 fatty acid intake, or an equivalent intake of calories/energy over the intervention period, sufficient to control for possible confounding stemming from such protocol violations.

None of the studies specifically identified the omega-6/omega-3 fatty acid content of their planned on-study^{99, 100} or background diets.⁹⁸ Llorente et al. did not report an attempt to maintain blinding via deodorizing their omega-3 fatty acid materials and/or by preventing oxidation and inevitable rancidity. Neither general purity data, nor data concerning possible methylmercury contamination, were provided regarding their DHA supplementation.⁹⁸ On the other hand, Llorente et al. appeared to use appropriate methods to handle blood lipid materials.

Hakkarainen et al.'s observational study assessed habitual dietary intake over the previous year (measure undefined).¹¹¹ This included fish consumption as well as the intake of omega-3 fatty acid content from fish and vegetables, and total omega-3 fatty acid intake. Total omega-3 fatty acid intake was calculated as 2.2 g/d or 0.47 g/d from fish, a value that they asserted is considerably higher than what is observed in North American populations.¹¹¹

Tanskanen et al's single food-frequency question assessed fish consumption over the past six months ("How often do you usually eat fish or fish meals? 1, < once a month or never; 2, once or twice a month; 3, once a week; 4, twice a week; 5, almost daily; 6, once a day or more often).⁸¹ Responses of 2 or less constituted infrequent consumption. In their second observational study Tanskanen et al.⁸⁰ estimated fish consumption via a food-frequency questionnaire (undefined) purported to produce comparable results to a 7-day food record.¹⁶⁶ A frequent fish consumer was defined as someone eating fish at least twice a week.

Fish intake in Woo et al.'s study was measured via a food frequency questionnaire administered at the participants' residence.¹¹⁰ Suzuki et al. utilized a validated semiquantitative food frequency questionnaire regarding 138 foods, including 18 fish and seafood items. It had in the past exhibited a significant association with dietary record data.¹⁰⁷ Participants were asked to report the average frequency, and usual serving size, of consumption during the year immediately preceding the onset of cancer symptoms. From this, Suzuki et al. calculated an average daily intake of food and nutrients. They then calculated the daily intake of omega-3 fatty acid content using the Fatty Acid Composition Table of Japanese Foods.¹⁶⁷ For the 771 participants whose data were analyzed, total omega-3 fatty acid intake was primarily ingested from vegetable oils and fats (37% of total intake), followed by 17 types of fish (35%), soybean products (11%), seasonings (5%), and cereals (4%). The daily intake consisted of 62% ALA, 20% DHA, and 11% EPA.¹⁰⁷ Edwards et al. completed a full analysis of the current diet using a 7-day weighted intake method.⁴⁸ Although data are reported later in this report, it should be stated here that no notable inappropriate methods to extract, prepare, store or analyze lipids were described.⁴⁸

Hibbeln's first ecological analysis estimated apparent fish consumption as: fish catch plus imports, minus exports.⁴⁷ This method is not as reliable as direct dietary surveys but at least this analysis included comparable data across countries. In his second analysis apparent fish consumption data were drawn by Hibbeln from the National Marine Fisheries Service and the Food and Agriculture Organization of the United Nations.¹⁰⁸ Data on food use were taken from the FAOSTAT database,¹⁶³ and captured apparent national food consumption in Peet's analysis.¹⁰⁹ Food use was estimated from the total domestic production of food plus imports, minus exports, while taking into account changes in stocks (e.g., stored grain), and subtracted food lost to waste during processing. Fish and seafood data were included, and were expressed as supply in kilograms per capita per year. Annual food consumption was approximated closest to the years in which the clinical studies were conducted (i.e., IPSS=1970; DOSMED=1980; depression=1990).¹⁰⁹

The manufacturer of Llorente et al.'s intervention was Martek Biosciences Corporation.⁹⁸ Purity data concerning its contents were not provided. No study report included details as to whether, or how, the presence of methylmercury was tested or eliminated from their omega-3 fatty acid exposure.

Cointervention characteristics. In Llorente et al.'s study, breast-feeding women were excluded if they used dietary supplements other than vitamins.⁹⁸ Wardle et al. excluded participants if they were currently using, or had used within the last 3 months, lipid-lowering medication.⁹⁹ In Ness et al.'s study, male adults were receiving anti-anginal medication, details of which were not provided in their report.¹⁰⁰ The possible use of other products with psychotropic properties was not reported for these studies.

Of the six observational studies,^{48, 80, 81, 107, 110, 111} only the study by Edwards et al. reported on the status of possible cointerventions. Each individual diagnosed as depressed was receiving antidepressant medication.⁴⁸ Similar data were not reported in any of the cross-national ecological analyses.^{47, 108, 109}

Outcome characteristics. Llorente et al. employed the BDI, EPDS and the SCID-CV, with the latter supporting DSM-IV diagnostic criteria.⁹⁸ Wardle et al. used the BDI, along with the STAI, anxiety and anger subscales of the POMS, GHQ to assess general psychological well-being, and the PSS. Dietary diaries were filled out at baseline and 12 weeks. Ness et al. used the DSP to measure depression and anxiety.

Hakkarainen et al.'s study evaluated depressed mood via self-report (no measure identified).¹¹¹ Assessments were recorded three times annually. Data on hospital-based treatments for major depressive disorder were drawn from the National Hospital Discharge Register, and suicides were identified from death certificates. Cox's proportional hazards regression models estimated the relationships between baseline dietary intake of omega-3 fatty acids (from fish, vegetables, and total intake), calculated from the food-use questionnaire and categorized in tertiles (with the lowest tertile as reference category), and measures of mood level and hospital-based treatments for major depressive disorder. The following potential risk factors for major depressive disorder and suicide were entered, as covariates, into the regression models: age, BMI, energy intake, serum total cholesterol, HDL cholesterol level, alcohol consumption, education, marriage, self-reported depression, self-reported anxiety, and smoking. Dietary factors were adjusted for energy intake.

The following BDI-defined distinctions were made in Tanskanen et al.'s first study: scores below 10 indicated no or minimal depressive symptoms; scores from 10–18 indicated mild symptoms; 19–29, moderate symptoms; and 30–63, severe symptoms.⁸¹ For bivariate analyses, the categories were normal mood, 0–9, and mild to severe symptoms, 10–63.⁸¹ Multiple logistic regression analysis assessed the relationship between BDI-indexed depressive symptomatology and fish consumption. Adjustments were made for these potential confounders: age, marital status, unemployment, current smoker status, irregular physical activity, female, BMI, more than 120g per week of pure alcohol, at least seven cups per day of coffee, low education level, and serum cholesterol level. The other Tanskanen et al. study also employed the BDI, while analyzing separately data for the single item pertaining to suicide ideation.⁸⁰ Analyses adjusted for the following potential confounders: sex, age, marital status, education, employment status, work ability, area of residence, financial status, general health smoker status, alcohol intake, coffee intake, and physical activity.

Woo et al. utilized the GDS while adusting for age and baseline health status at the start of their 3-year study.¹¹⁰ From previous validational work it was reported that a score of at least 8 on the 15-point GDS provides a sensitivity and a specificity of 96.3% and 87.5%, respectively, for a psychiatric diagnosis of depression in the local Chinese population. The HADS depression subscale was employed by Suzuki et al.¹⁰⁷ A cutpoint of 4 out of 5 has previously been observed to reflect good sensitivity and specificity (91.5% and 58%, respectively) for screening depression (e.g., major depression). Analyses adjusted for age, sex, performance status, clinical stage, histology, pain, breathlessness, employment status, smoker status, alcohol consumption, and BMI. Edwards et al.'s analyses controlled for stress and smoking status.⁴⁸

Given the limitation of the Japanese data, in that they were not produced using a structured diagnostic instrument or via random population sampling techniques, data analysis in Hibbeln's first cross-national assessment was completed both including and excluding data from Japan.⁴⁷ Since adverse personal, social and economic conditions can increase the risk of depressive symptomatology in the postpartum period, the following variables were controlled for in Hibbeln's second cross-national ecological analysis: study time postpartum, low socioeconomic status, percentage of young mothers, percentage of mothers without partners, percentage of mothers with secondary education, and the influence of Asian cultures.¹⁰⁸ Data on depression prevalence in Peet's cross-national ecological analysis¹⁰⁹ were captured from Weissman et al.⁴⁶ and the same Japanese source used by Hibbeln.⁴⁷ From the IPSS study, data on mean days out of hospital and percentage of patients with schizophrenia and severe social impairment were used as outcomes.¹⁰⁹ In addition, a total outcome score was derived.¹⁴⁵ It is a composite score taking into account all IPSS outcomes. From the DOSMED study, outcomes selected were percentage of patients never hospitalized and the percentage of patients with little social impairment. Urban data were used exclusively, where available. A “total best outcome” score was derived by adding data from various “best possible” DOSMED outcomes (e.g., remitting course with full remissions; on no antipsychotic medication during followup).

Study quality and applicability. The mean total Jadad quality score was 3,^98–100 with two of the three RCTs adequately concealing their allocation of participants to study groups.^{98, 99} The third RCT received an Unclear allocation concealment rating.¹⁰⁰ The mean quality score for the two single prospective cohort studies was 4.5, with both studies attaining a III applicability rating.^{110, 111} All three cross-sectional surveys received an applicability rating of III, and together they achieved a mean quality score of 4.7.^{80, 81, 107} The single cross-sectional study received a quality score of 6 and an applicability rating of II.⁴⁸ The three cross-national ecological analyses received a mean quality score of 4, with all achieving an applicability rating of III.^{47, 108, 109} Of all the studies, only the Edwards et al. one received an applicability rating other than III (i.e., II),⁴⁸ and only three investigations achieved a study quality grade of A.^{98, 107, 108}

Qualitative Synthesis of Individual Study Results

Llorente et al. reported data only for completers for whom they had baseline and 4-month data.⁹⁸ After 4 months of supplementation, plasma phospholipid DHA content in the DHA group had increased by 8% while the DHA content in the placebo group had decreased by 31%, the former observation indicating a reversal in the typical decline in DHA levels. The DHA content of the DHA group was 50% higher than that of the placebo group 4 months post-delivery. However, there were nonsignificant statistical differences between study groups after 4 months for the BDI, EPDS and the SCID-CV (diagnostic counts). Yet, according to BDI scores, only nine women in the placebo group and 11 women in the DHA group achieved a score of at least 10 at one of their followups, indicating minimal symptoms of depression (>9 may indicate mild symptoms). Two and four women in the placebo and DHA groups, respectively, had a BDI score of at least 20, indicating moderate symptoms. SCID-CV observations confirmed these results. Only seven women (DHA group=4; placebo group=3) met DSM-IV diagnostic criteria for a “current depressive episode” during the 4-month postpartum period.

All three of Wardle et al.'s study groups showed significant within-group improvement on many of the mental health outcomes after 12 weeks (i.e., BDI score and anger reactions in both diet groups; stress and anxiety only in Mediterranean diet group).⁹⁹ Yet, there were no significant between-group differences observed for any of the following clinical outcomes: depression, anxiety, anger/hostility, stress, and general psychological well-being. Thus, no reliable associations between intake of omega-3 fatty acids and any of the examined indices of mental health were observed.

Ness et al. observed that self-reported fish intake was higher in the fish advice group at study's end.¹⁰⁰ No statistical difference was observed in the fish advice group either for depression or anxiety; and, controlling for baseline mood, the between-group difference for each outcome was not statistically different. This last observation did not change following an additional adjustment made for one's status as having been randomized to the stress management arm, nor was there any statistical evidence of interaction between these factors in their effects on mood. Looking exclusively at the upper quartile of baseline depression or anxiety score did not contradict these observations.

In Llorente et al.'s study no subject withdrew due to adverse effects related to the supplement.⁹⁸ However, 37 of 138 women either withdrew or were dropped. Thirteen withdrew because of maternal illness, 14 were dropped due to lactation failure or excessive formula intake by the child (>20% of total intake), one mother moved away, five were dropped due to infant illness, and four discontinued participation. Wardle et al. reported that similar numbers of patient withdrew before 12 weeks in each study group (low-fat=7; Mediterranean=8; control=6), and typically due to attendance problems.⁹⁹ Seven men died within 7 months of randomization into Ness et al.'s trial (fish advice group=3), and for reasons other than the intervention.¹⁰⁰

Hakkarainen et al.'s attempt, in their observational study, to assess the possible relationship between low dietary intake of omega-3 fatty acids and depression revealed that, accounting for the above-noted covariates, there was no significant association of fish consumption or calculated intake of omega-3 fatty acids and self-reported depressed mood, hospital treatment required due to major depressive disorder, or suicide (data not reported).¹¹¹

Tanskanen et al. showed that, using BDI scores, 20% of their sample experienced mild depressive symptoms (n=647), 6.3% had moderate symptoms (n=201), and 1.5% reported severe symptoms (n=48).⁸¹ Sixty-four percent reported eating fish or fish meals once or twice a week (n=2,053), 6.3% ate fish daily (n=201), and 30% ate fish once or twice a month, or less often (n=950). From bivariate analysis, mild to severe depressive symptoms were more prevalent among women who infrequently consumed fish (less than once a week) than those who were frequent fish eaters (more than once per week). A similar trend was observed among men, yet the results were not statistically significant. Compared with frequent fish consumers (bivariate analysis), infrequent consumers were younger, less physically active, less obese, less likely to have a lower serum cholesterol level, unmarried, smoked, and drank a lot of coffee. Yet, higher age, being unmarried, unemployment, smoking, lower levels of physical activity, greater degree of obesity, low level of education, and higher serum cholesterol level were associated with depressive symptoms. Multiple logistic regression analysis, including confounders, revealed that infrequent fish consumption was independently associated with depressive symptoms. The likelihood of exhibiting mild to severe depressive symptoms was 31% higher among infrequent fish consumers than frequent consumers. Depressive symptoms were significantly associated with infrequent fish consumption for females only.

In their second cross-sectional study Tanskanen et al. observed that the risk of being depressed and the risk of suicidal ideation were significantly lower among those who frequently ate lake fish, compared with infrequent fish consumers.⁸⁰ These relationships held after adjusting for the above-noted factors.

Woo et al. found nonsignificant adjusted and unadjusted estimates of association involving fish intake and depressive symptoms in a Chinese elderly population although data were not reported per se.¹¹⁰ They also observed that increasing levels of physical activity and occasional intake of alcohol were associated with a reduced risk of depressive symptoms. After 36 months, 341 participants (17%) had been lost to followup and 519 had died.

For newly diagnosed lung cancer patients Suzuki et al. reported a statistically significant adjusted odds ratio for depression between upper and lower quartiles of ALA and total omega-3 fatty acid intake, indicating a significant inverse association.¹⁰⁷ They also found a statistically nonsignificant adjusted difference for depression between upper and lower quartiles of intake of EPA, DHA and EPA+DHA.¹⁰⁷ Results of tests for trend paralleled these findings. But, no association was observed for depression and intake of fish or seafood, also following adjustments for potential confounders.

There were no significant between-group differences for current intake of omega-3 fatty acids or total energy intake assessed by 7-day dietary intake in Edwards et al.'s study.⁴⁸ Within the depressed patient group there was a significant negative correlation between the BDI-defined severity of depressive symptomatology and dietary intake of total omega-3 fatty acids as well as ALA. When data were pooled from patients and controls and entered into multiple regression, none of the dietary omega-3 or omega-6 fatty acid variables were significant predictors of depression. Data pertaining to smoker status and stress were entered only into analyses involving biomarker data.

In their first cross-national ecological analysis Hibbeln found a significant, inverse correlation between apparent fish consumption (fish pounds per person; 1 pound = 0.4536 kg) and major depression.⁴⁷ When data were excluded from Japan for the above-noted reason, a significant correlation was maintained. Data regarding potential confounders were not consistently available for each of the countries in Hibbeln's second cross-national ecological analysis.¹⁰⁸. Nevertheless, simple regression and a logarithmic equation revealed that higher national seafood consumption predicted lower prevalence rates of postpartum depression, that higher DHA content in mother's milk predicted lower prevalence rates, and that the AA and EPA content of mother's milk were unrelated to prevalence rates of postpartum depression. Only low socioeconomic status, young maternal age, the percentage of women without partners, and percentage of mothers with a secondary education predicted prevalence rates. These relationships appeared to be influenced by data from Brazil and South Africa, suggesting that these results may have confounded the findings in the primary analyses. However, first excluding Asian countries' data because of their stronger intake of omega-3 fatty acids in the background diet, and then data from Brazil or South Africa, yielded findings paralleling those from the primary analyses. This indicated the robustness of the main findings.

Peet's schizophrenia results are presented below.¹⁰⁹ He observed a significant association between high consumption of fish and seafood and a reduced prevalence of depression.¹⁰⁹

Quantitative Synthesis

Very few of the studies that met the eligibility criteria for this question actually demonstrated the inherent capacity to afford the drawing of causal inferences regarding the possible relationship between the intake of omega-3 fatty acids and the onset of depression as disorder or symptom. Only three of 12 studies were eligible for quantitative synthesis in that they were both controlled and prospective by design.^98–100 The observation that these three RCTs employed highly different target populations, interventions, controls and outcomes made it inappropriate to consider conducting meta-analysis. Moreover, only one trial investigated the potential of specific amounts of omega-3 fatty acid content, via DHA supplementation, to protect its population (i.e., breastfeeding women) from developing (postpartum) depression.⁹⁸

Impact of Covariates and Confounders

With such diverse designs, populations, exposures, controls and outcomes it is difficult to cull patterns of notable finding regarding the influence of extra-exposure variables on outcomes pertinent to this review. The designs with the greatest inherent potential to control for confounding influences (i.e., RCTs)^98–100 did not yield a single significant result, although the primary goal in two of them did not entail demonstrating the potential of omega-3 fatty acids as protection against depression.^{99, 100} At the same time, the three RCTs likely confirmed, in part, the success of their randomizations by showing that study groups were equivalent at baseline on certain important bases (e.g., mental health variables).

Likewise, the multiple-group cross-sectional study did not reveal a significant association between omega-3 fatty acid intake and depression while also reporting that study groups were equivalent in their intake of omega-3 fatty acids, for example.⁴⁸ And, while most of the uncontrolled observational studies did a reasonable job of adjusting for confounders in their analyses (e.g., age, smoker status, alcohol consumption),^{80, 81, 107, 111} their results did not consistently show a significant association between the intake of omega-3 fatty acids and the risk of depression. Even the two surveys conducted in Finland, where fish intake is considerably higher than in North America, for example, failed to produce a consistent result for both sexes.^{80, 81}

The ability to control for confounders in the three cross-national ecological analyses depended on the initial data collection strategies, which produced the databases from which the three studies' data were obtained. Without all of the details, many of which were not published in the included reports, it is difficult to draw conclusions about these three analyses' successes or failures in controlling for key influences on outcomes.

Overview of Relevant Studies

Seven multiple-group cross-sectional studies^{48, 101–106} and one RCT⁹⁸ provided data pertaining to this question. Two studies have already had their key study parameters introduced with respect to the question of the possible association of omega-3 fatty acid intake and the onset, continuation or recurrence of depression. Yet, the Llorente et al. RCT⁹⁸ and Edwards et al.'s multiple-group cross-sectional study⁴⁸ data were nevertheless placed in summary tables.

Ellis and Sanders assessed the fatty acid content of plasma choline phosphoglycerides (CPG) and RBC ethanolamine phosphoglycerides (EPG) in patients diagnosed with endogenous depression (n=6), patients on the same ward yet with non-depressive psychiatric disorders (n=4; types undefined), and age- and sex-matched controls drawn from hospital staff (n=6) (Summary Table 7).¹⁰⁵ Fehily et al. compared the fatty acid content of plasma CPG and RBC phospholipids in patients with: endogenous depression (n=26; mean age: 52 [21–74] years; 7 bipolar and 16 unipolar diagnoses; 54% drug-free for at least 2 weeks before study), those with reactive depression (n=23; mean age: 38 [22–65] years; 65% drug-free for at least 2 weeks), other psychiatric disorders (n=11; mean age=35 [19–59] years; 6 schizophrenia and 5 personality disorder diagnoses; 46% drug-free for at least 2 weeks) and age- and sex-matched healthy controls (n=undefined; age undefined).¹⁰⁶

Maes et al. examined the fatty acid composition of serum cholesterol esters and phospholipids in 36 patients with major depressive disorder (with [n=11] or without melancholia [n=25]), 14 with minor depression (i.e., adjustment disorder with depressed mood and dysthymia) and 24 healthy volunteer subjects (staff or their family members) (Summary Table 8).¹⁰³

Peet et al. investigated fatty acid composition in the RBC membranes of 15 drug-free patients with major depressive disorder, unipolar variety, and 15 age- and sex-matched healthy controls (Summary Table 9).¹⁰² All medication was stopped for 8 to 91 days prior to blood sampling.

Edwards et al. measured the dietary PUFA intake as well as the fatty acid content of RBCs in a cross-sectional study of ten depressed patients diagnosed with major depression using DSM-IV criteria, and 14 matched healthy control subjects.⁴⁸ Each depressed patient was receiving antidepressant medication. Analyses controlled for stress and smoking status. Additional details regarding exclusion criteria or matching requirements were presented in relation to the question concerning the possible association between omega-3 fatty acid intake and the onset, continuation or recurrence of depression.

Maes et al.'s second study investigated 34 major depressed inpatients and 14 healthy volunteers in an attempt to establish whether major depression was associated with a decrease in omega-3 fatty acids or an increase in omega-6 fatty acids in serum phospholipids and cholesteryl esters (Summary Table 10).¹⁰¹ They also assessed the relationship between these PUFAs and levels of serum zinc (with a low level being a marker of the inflammatory response system's activation), as well as the effects of 5 weeks of subchronic treatment with antidepressants (i.e., fluoxetine 20 mg/d alone or with trazodone 100 mg/d or pindolol 7.5 mg/d) on fatty acid levels in 20 patients. Patients underwent a 10-day drug-free period upon hospital admission.

Tiemeier and colleagues investigated whether community-dwelling elderly with depression have a fatty acid composition different from those who are not depressed (Summary Table 11).¹⁰⁴ As part of the Rotterdam population-based cohort study (n=7,983), 3,884 adults of at least 60 years of age were screened for depressive symptoms. Those that screened positive had a psychiatric interview to diagnose depressive disorders. After excluding individuals with other disorders (n=29), and following the loss of 14 subjects, study groups became: those with depressive disorder (n=106; 61–97 years), those with subclinical depressive symptoms (n=115; 61–93 years) and randomly selected controls who had screened negative for depression in the Rotterdam study (n=461; 61–101 years). The analysis included an assessment of the possible roles played by atherosclerosis and the inflammatory response, the latter measured by C-reactive protein. Given that certain factors such as chronic diseases, smoking and cholesterol concentrations have been related in community-dwelling populations to depression and fatty acid composition,¹⁶⁸ they were investigated for their possible roles as confounders. Other confounders included: age, sex, level of education, history of stroke, cognitive function (Mini Mental State examination), functional status, and blood pressure.

Llorente et al. assessed the effect of DHA supplementation on the onset of postpartum depression as well as on plasma phospholipid DHA content in women who breast-feed.⁹⁸ Mothers who planned to breast-feed their children (n=138; 18–42 years) were randomly assigned, in double-blind fashion, to receive either ~200 mg/d DHA or placebo (undefined) for the first 4 months after delivery. Plasma phospholipid data were collected just before delivery and at 4 months. Additional data regarding study, population and intervention parameters are presented with reference to evidence concerning the possible association of omega-3 fatty acid intake and the onset, continuation or recurrence of depression.

Qualitative Synthesis of Relevant Studies' Key Characteristics

Study characteristics. With the exception of Llorente et al.'s RCT,⁹⁸ studies were cross-sectional studies involving at least two groups.^{48, 101–106} Some study reports provided very detailed inclusion/exclusion criteria establishing strong experimental controls,^{101, 103} whereas others provided very little information.¹⁰⁵ Other reports provided sufficient detail to allow for an appreciation of the rigor associated with these studies.^{48, 98, 102, 104, 106} Study sizes varied between 16¹⁰⁵ and 682 participants.¹⁰⁴ Countries where the studies were conducted included England,^{48, 102, 105, 106} Belgium,^{101, 103} Holland¹⁰⁴ and the US.⁹⁸ Fehily et al's study was supported by a grant from the South Thames Regional Health Authority.¹⁰⁶ The first Maes et al. study was funded by numerous sources, including the National Funds for Scientific Research (Belgium), the IUAP program (Belgium), as well as grants from the US Preventive Health Services (USPHS) and the Elisabeth Severance Prentiss and John Pascal Sawyer Foundations.¹⁰³ One author was the receipient of a USPHS Research Center Career Scientist Award. The second Maes and colleagues study was funded in part by the National Funds for Scientific Research (Belgium), the Clinical Research Center for Mental Health (Belgium), in addition to an Staglin Investigator Award given to the lead investigator.¹⁰¹ Tiemeier et al.'s work was supported by the Research Institute for Diseases in the Elderly, which is funded by the Ministry of Education and Science, and the Ministry of Health, Welfare, and Sports through the Netherlands Organization for Scientific Research, and a grant from Numico Research.¹⁰⁴ Llorente et al.'s study was supported by industry (Martek Biosciences Corporation).⁹⁸ Three groups did not report a funding source.^{48, 102, 105}

Population characteristics. Complete population age data were not reported for both the full sample and/or the different study groups in some studies, this despite the avowal of the authors that study groups were matched by age and sex.^{48, 102, 105} In the study by Fehily et al., the group of subjects with endogenous depression was older than the control group.¹⁰⁶ The pregnant women examined in Llorente et al.'s study were between 18 and 42 years of age; there was no significant difference in ages of the participants between study groups.⁹⁸ In each of Maes et al.'s studies, neither the between-group age of participants nor the between-group female/male ratio were significantly different.^{101, 103} Neither age nor sex significantly predicted any omega-3 fatty acid fractions or any omega-6/omega-3 ratios in these two studies.^{101, 103} Yet, both variables were entered as covariates in subsequent analyses due to their possible relationship with fatty acid levels.^{101, 103} Peet et al.'s study population was between 18 and 65 years of age, and the authors confirmed successful matching for this possible confounder.¹⁰² Tiemeier et al.'s study groups were age-matched, with ages ranging from 61 to 101 years of age; subjects with depressive disorder were more likely to be female.¹⁰⁴ Six of the study populations were not described in terms of ethnic/racial background, while Maes et al.'s participants were explicitly identified as Caucasian of Flemish origin.^{101, 103}

The studies conducted by Ellis and Sanders¹⁰⁵ and Fehily et al.¹⁰⁶ each included heterogeneous subtypes of endogenous depression, with subtypes undefined in the former¹⁰⁵ and with neither report presenting outcome data broken down by any of these subtypes. The remaining studies identified reasonably well-defined groups for which to compare biomarkers data. The Maes et al. studies likely serve as the best examples of a well-conceived and operationalized separation of study groups.^{101, 103} In this regard, their depressed patients were identified using DSM-III-R diagnostic criteria applied via the SCID, patient version.^{101, 103} Peet et al. employed DSM-IV criteria to identify depressed patients.¹⁰² Neither Ellis and Sanders nor Fehily et al. described their diagnostic criteria.^{105, 106} Depressive disorders were identified by Tiemeier et al. via a score of at least 16 (i.e., clinically significant depressive symptoms) on the validated Dutch version of the Center for Epidemiologic Studies Depression scale (CES-D) during a home interview, followed by a psychiatric workup using the Dutch version of the Present State Examination (i.e., a semistructured interview from the validated Schedules for Clinical Assessment in Neuropsychiatry). DSM-IV criteria were used to guide the diagnosis, with categories including major depression and dsythymia in addition to minor depression.¹⁰⁴

Few studies adequately ruled out the presence of possible psychopathology, or risks thereto, in subjects typically identified as “healthy volunteers” or “healthy controls.”^{102, 105} Maes et al.'s investigations carefully provided the basis for separating their study groups to achieve control of this confounder. Healthy volunteers were excluded for present, past and family (first degree) history of Axis I or Axis II disorders using the SCID, Lifetime version.^{101, 103} All participants had low scores on the Zung Depression and Anxiety Scales (<32) and the BDI (<9).¹⁰³ Controls were medication-free for at least 1 month prior to blood sampling.^{101, 103} None had ever taken psychotropic drugs¹⁰³ or was a regular drinker.^{101, 103}

A few studies established the baseline severity of symptomatology. For example, Fehily et al. used the BDI,¹⁰⁶ Maes et al. employed the HDRS, and Peet et al. used the MADRS (no data reported).^{102, 103} In their first study Maes et al. observed that those with major depression had significantly higher baseline HDRS severity scores than did those with minor depression.¹⁰³ Ellis and Sanders, for example, did not measure severity.¹⁰⁵ Baseline data concerning the duration of the current episode, age of onset, number of previous episodes, and time since diagnosis were rare.

In attempts to control for possible confounding from variability due to comorbid conditions, some studies applied strict exclusion criteria. For example, in both of the studies by Maes et al., patients were excluded if they had Axis I diagnoses other than unipolar depression, including psychotic disorders, organic mental disorders, impulse control disorders, substance use disorder or substance abuse (within the last 6 months), or borderline and antisocial personality (Axis II) disorders.^{101, 103} Also excluded were individuals with abnormal X-rays of heart and lungs, electrocardiogram or electroencephalogram.^{101, 103} All study participants had normal chemical and hematologic tests relating to, for example, liver function and renal function,^{101, 103} as well as electrolyte, thyroid hormone and thyroid stimulating hormone levels.¹⁰¹ All were free of medical illness (e.g., immune and endocrine disorders such as diabetes, inflammatory bowel syndrome, autoimmune disorders, essential hypertension and arteriosclerosis).^{101, 103} None exhibited evidence of allergic, inflammatory or immune responses for at least 2 weeks prior to blood sampling.^{101, 103} BMI was within normal limits.^{101, 103} Heavy smokers (>15 cigarettes per day) were excluded.¹⁰¹

Peet et al. excluded those individuals with a physical illness of a severity or nature associating it with abnormal levels of omega-3 fatty acid levels.¹⁰² Controls were medication-free, and without a history of psychiatric illness, personality disorder, substance abuse or medical illness (method undefined). Yet, Tiemeier et al. noted differences in their study groups, with elderly individuals with depressive disorders more likely to have had a stroke and to exhibit significantly lower activities of daily living scores and cognitive scores compared with those without depressive symptoms.¹⁰⁴ Some studies did not identify possible psychiatric comorbidity or control for it via the application of clearly stated exclusion criteria.^{105, 106} Yet, Peet et al. did note the absence of significant between-group differences regarding smoker status or in the relationship between smoker status and PUFA content.¹⁰²

Six of eight studies did not involve an intervention or exposure. Only Llorente et al. employed supplementation,⁹⁸ as possible prophylaxis, and Edwards et al. assessed dietary intake of omega-3 fatty acids.⁴⁸ In both Maes et al. studies all participants were consuming a normal Belgian diet (PS ratio=0.54±0.43); and, those on a low fat diet were excluded.^{101, 103} No other studies controlled statistically for background diet in their analyses. No study reported inappropriate methods by which lipids were extracted, prepared, stored or analyzed.

Ellis and Sanders did not describe the medication status of their participants (i.e., medication-naïve, medication-free or medicated, with type and dose).¹⁰⁵ Fehily et al. reported that different percentages of individual within study groups were drug-free, indicating heterogeneity within diagnostic groups.¹⁰⁶ This situation could confound the results. Those receiving medication were receiving a hypnotic and/or a tranquillizer. Two schizophrenic patients that had been admitted to the study, yet whose data were not analyzed separately, were taking a neuroleptic. The investigators reported that the fatty acid content of those taking these drugs and those who were medication-free was not different (no data or p-value reported).¹⁰⁶ Peet et al.'s patients were drug-free at first assessment.¹⁰² Seven patients then received dothiepin, three took paroxetine, and one each received trazodone and lofepramine.¹⁰²

Maes et al. excluded those individuals receiving treatment with MAOIs, antipsychotic doses of neuroleptic, anticonvulsants, lithium or ECT in the previous year.^{101, 103} Maes et al. also specified fluoxetine and trazadone in their second study.¹⁰¹ No cholesterol-lowering drugs were permitted.¹⁰¹ Use of any medication known to influence fatty acid metabolism or endocrine and immune function was prohibited as well.¹⁰¹ No significant between-group differences were observed for the prestudy use of antidepressants, benzodiazepines or antipsychotics in Maes et al.'s first study.¹⁰³ Prestudy use of the different drug classes did not significantly predict any of the omega-3 fatty acid fractions or any omega-6/omega-3 ratios.¹⁰³ All antidepressant, benzodiazepine or low dose antipyschotics were discontinued the month prior to an 8–10 day washout period.¹⁰³ The second Maes et al. study mandated the discontinuation of antidepressants upon hospital admission.¹⁰¹ Twenty-six depressed patients had been treated with antidepressants during the depressive episode.¹⁰¹

Twenty-seven patients with depression in Maes et al.'s first study,¹⁰³ and 18 patients in their second study,¹⁰¹ used a low dose of benzodiazepines for severe agitation, anxiety sleep disorders or suicidal ideation during the study period. There was no significant between-group difference in the use of these on-study medications.^{101, 103} As well, there were no significant differences in EFA status data for those depressed patients who did or did not use on-study benzodiazepines.¹⁰¹

Outcome characteristics. Outcomes included all types of fatty acid, from various sources, and were expressed either as percentages, or fractions (i.e., composition), or concentrations.

Study quality and applicability. The seven cross-sectional studies received a mean quality score of 4.6, with four achieving an applicability rating of II,^{48, 102, 105, 106} and three attaining an applicability rating of III.^{101, 103, 104} The single RCT was assigned an Jadad total quality score of 5, an Adequate allocation concealment rating, and an applicability score of I.⁹⁸

Qualitative Synthesis of Individual Study Results

The proportions of plasma CPG EPA and DHA were each significantly greater in the endogenous depression group as compared to healthy controls in the Ellis and Sanders study.¹⁰⁵ On the other hand, AA did not differ between these two groups. These differences were less pronounced for RBC EPG data (no data or p-values reported). There were no significant differences in fatty acid status between non-depressed patients and healthy controls.

Concentrations of DHA and EPA in plasma CPG were each significantly higher, but LA was significantly lower, in Fehily et al.'s endogenous depression group compared with matched controls.¹⁰⁶ There was no significant difference between these two groups in terms of AA levels. The plasma CPG status of patients with reactive depression or other psychiatric disorders did not differ from the controls. Eighty percent of those with endogenous depression had DHA levels of more than 54 mg/g total fatty acid esters detected, as compared to 19% of matched controls. DHA concentrations were correlated with BDI severity score for those identified as endogenously depressed, but not for those with reactive depression. Similar, but smaller, between-group differences were observed for the fatty acid content of RBCs of patients with endogenous depression compared with matched controls (i.e., higher DHA in EPGs; higher EPA in serine phosphoglycerides).

By ANCOVA, with age and sex as covariates, major depressed patients exhibited a significantly higher AA/EPA ratio in both serum cholesteryl esters and phospholipids in addition to a significantly increased total omega-6/omega-3 ratio in cholesteryl ester fractions than did healthy volunteers or minor depressed subjects in Maes et al.'s first study.¹⁰³ Significant between-group differences were not observed for total omega-3 or total omega-6 fatty acid content, or their ratio, in phospholipids. The only significant correlations involved the HDRS score with the AA/EPA or total omega-6/omega-3 fatty acid ratios in phospholipids. Major depressed patients had significantly lower ALA in cholesteryl esters compared with healthy controls. Major depressed patients showed significantly lower total omega-3 fatty acids in cholesteryl esters and significantly lower EPA in serum cholesteryl esters and phospholipids compared with minor depressed subjects or healthy controls. ALA, EPA and DHA cholesteryl ester fractions successfully discriminated the three study groups. MANOVA using ALA, EPA and DHA cholesteryl ester fractions as dependent variables showed highly significant differences among the three study groups. There was a significant negative relationship between EPA in cholesteryl esters and HDRS scores.

Peet et al. observed a significant reduction in RBC membrane total omega-3 fatty acids and DHA content in drug-free depressive patients.¹⁰² They also observed a significant reduction in LA, DGLA and total omega-6 fatty acids. No significant between-group differences were found for AA/EPA, AA/DHA or total omega-6/omega-3 fatty acid ratios. Subsequent intervention with anti-depressants failed to have a significant effect on the RBC omega-3 fatty acid status. Yet, this study failed to fully control for possible confounding influences such as stress, smoking or diet.

Edwards et al. reported that RBC membrane EPA, DHA and total omega-3 fatty acid levels were significantly lower in the depressed patient group.⁴⁸ There were no significant differences for any omega-6 fatty acid levels. There were no significant between-group differences for current dietary intake of omega-3 fatty acids or total energy intake (via 7-day weighted intake). The only significant, and negative, correlations involved omega-3 fatty acids and BDI-defined severity score: for ALA, DHA and total omega-3 fatty acid content. Multiple regression revealed that only RBC membrane ALA significantly predicted BDI score. When dietary and RBC membrane data were entered in stepwise fashion, DHA and LA emerged as the only predictors of BDI severity score. Neither current smoker status nor recent stress had an effect on RBC membrane values.

Maes et al. found in serum phospholipids that major depression was associated with (all significant): higher MUFA fractions, lower adrenic acid (omega-6) yet higher (omega-6-)DPA, lower EPA, lower (omega-3-)DPA, higher AA/EPA ratio, higher (omega-6-)DPA/DHA fractions (i.e., composition: weight as percent of total).¹⁰¹ In addition, lower concentrations (mg/dL) of SFAs, MUFAs, LA, AA, adrenic acid (omega-6), total omega-6 fatty acids, ALA, EPA, (omega-3-)DPA, DHA, and total omega-3 fatty acids (all significant) were found in the serum phospholipids of patients with major depression.¹⁰¹ For serum cholesteryl esters, major depression was associated with (all significant): lower ALA, EPA, and total omega-3 fatty acids; and, higher total omega-6/omega-3 fatty acids and AA/EPA fractions. Additionally, major depression was associated with lower total saturated fatty acids, MUFAs, LA, DGLA, total omega-6 fatty acids, ALA, EPA, and total omega-3 fatty acid concentrations (all significant), in serum cholesteryl esters.¹⁰¹ All analyses included age and sex as covariates. There were no significant correlations between HDRS score for depressive patients and any of the fatty acid variables.

In the phospholipids of major depressed patients, serum zinc was significantly and positively correlated with the percentages and concentrations of EPA, DHA and total omega-3 fatty acids.¹⁰¹ Significant negative correlations were observed for percentage of total omega-6 fatty acids, AA/EPA, (omega-6-)DPA/DHA, and total omega-6/omega-3 fatty acids. In their cholesteryl esters, only the total omega-3 fatty acid percentage, and EPA, were significantly and positively correlated with major depression. There was no significant effect of antidepressant treatment on fatty acid levels. With a decrease of at least 50% in HDRS score defining a good clinical response to antidepressants after 5 weeks, depressed patients were divided into responders and non-responders.¹⁰¹ There were no significant differences in fatty acid percentages between responders and non-responders.

Tiemeier et al. found no significant differences in the percentages or ratios of plasma phospholipid fatty acids between controls and those exhibiting subclinical depressive symptoms.¹⁰⁴ When the comparisons involved depressed subjects and controls, only a few, marginal differences were observed. By ANCOVA, with the above-noted covariates, percentages of AA and DHA were higher and lower, respectively, in the depressed subjects compared with controls. The ratios of total omega-6/omega-3 fatty acids and AA/DHA were higher in the depressed subject group when compared to reference subjects. Neither the inflammation marker C-reactive protein nor atherosclerosis affected these results. A test of interaction showed that the relation between the omega-6/omega-3 ratios and depressive disorders depended on the C-reactive concentration. That is, the difference in the omega-6/omega-3 ratio between depressed and reference subjects increased with lower concentrations of C-reactive protein. Stratification of the analysis at the median of C-reactive protein concentrations (1.5 mg/L), and involving subjects above this cutpoint, revealed no significant difference in fatty acid composition between the depressed and reference groups. Yet, when data were analyzed from those falling below the cutpoint, it was observed that depressed persons had significantly lower percentages of certain omega-3 fatty acids than did reference subjects. By ANCOVA, subjects with depressive disorder showed lower levels of EPA, DHA and total omega-3 fatty acids. As well, depressed subjects showed higher values for total omega-6/omega-3 fatty acids, AA/EPA and AA/DHA.

In their RCT, Llorente et al. observed no statistically significant correlations between plasma phospholipid DHA content, either at baseline or at 4 months, and self-rating (BDI, EPDS) or syndromal measures of depression (SCID-CV).⁹⁸

Quantitative Synthesis

Although all of the included studies were controlled, only one was prospective by design. Thus, meta-analysis was considered inappropriate. The exact nature of the inappropriateness of cross-sectional study data to address the question of onset is described in the Discussion.

Impact of Covariates and Confounders

Numerous factors have the capacity to influence EFA levels, including dietary intake, smoking and alcohol consumption.^101–103 Most of the studies did not control for smoking, for example, which alone could produce a picture of omega-3 fatty acid deficiency.⁶⁰ Only a minority of studies adequately controlled for the possible influence of this or any other variable.

The study by Peet et al.,¹⁰² and especially the two studies by Maes et al.,^{101, 103} employed strict controls, and results suggested an omega-3 fatty acid deficiency in depressed patients. Edwards et al. controlled for stress and smoker status.⁴⁸ However, less well-controlled studies—for example, failing to formally rule out psychopathology in the controls—also produced a similar picture of an omega-3 fatty acid deficiency in depressed patients. It is possible that the between-group differences might have been more pronounced in the latter studies had the possible influence of this and other potential confounding factors been minimized.

Failure to include even minimally homogeneous groups of depressed individual may have produced the only two study results suggesting that, compared with depressed patients, healthy controls exhibited an omega-3 fatty acid deficiency.^{105, 106} As well, in studies where patients were either already receiving antidepressant medication,⁴⁸ or received medication sometime after the initial blood sampling and were subsequently retested,^{101, 102} analyses revealed that antidepressant medication did not substantially modify the between-group difference in omega-3 fatty acid levels in biomarkers. The country in which the study was conducted could not readily be used as a surrogate for background diet in assessing the impact of the latter on study results since there was insufficient variability in study results.

Overview of Relevant Studies' Characteristics and Results

Each observational study has already had its key characteristics described with reference to the question of the possible association of omega-3 fatty acid intake and the onset, continuation or recurrence of depression. Hakkarainen et al. investigated the relationship between dietary intake of omega-3 fatty acids and low mood, major depression, and suicide in males 50 to 69 years of age living in southwestern Finland in 1985.¹¹¹ The study utilized a cohort (n=29,133) from a primary prevention RCT (ATBC Cancer Prevention Study). Followup lasted nine years. The intake of fatty acids and fish consumption were derived from a validated food use questionnaire focused on the “last 12 months.” Suicides were determined from Central Population Register data.

Tanskanen et al.'s sample was selected based on a random population sample (National Population Register) of both sexes (n=3,004; 25–64 years).⁸⁰ Data were gathered on fish consumption, depression (BDI) and suicidality. Suicidality was measured using a single BDI item. Given that the studies varied on the basis of their focus, that is, one on “successful” suicidal behavior and the other on suicidal ideation, and that the key study and population parameters have already been contrasted in an earlier part of the report, only the results are now presented. Quantitative analysis was considered inappropriate.

Adjusting for numerous factors (i.e., age, sex, marital status, education, employement status, work ability, area of residence, financial status, general health, smoking, alcohol intake, coffee drinking, and physical activity), Tanskanen et al. found that the risk of suicidal ideation was significantly lower among frequent consumers of lakefish.⁸⁰ Adjusting for many factors as well (i.e., age, BMI, energy intake, serum cholesterol level, HDL level, alcohol use, education, marriage, self-reported depression and anxiety, and smoking), Hakkarainen et al. observed no significant association between fish consumption or intake of omega-3 fatty acids and suicide.¹¹¹ Both Hakkarainen et al. and Tanskanen et al.'s results, while indicating good statistical control for important key confounders, are insufficient to allow us to infer the role of any covariates or confounders. Meta-analysis was not considered since outcomes were not comparable.

Study quality and applicability. Although they employed different research designs, both studies were assigned a level III for applicability, and together they received a mean quality score of 4.5.^{80, 111}

Overview of Relevant Studies' Characteristics and Results

Stoll et al. randomized 44 patients with bipolar disorder (18–65 years) to receive either 9.6 g/d of omega-3 fatty acids (6.2 g/d EPA, 3.4 g/d DHA) from menhaden fish body oil, via seven capsules twice daily, or identical gelatin placebo capsules containing olive oil ethyl ester (Summary Table 13).¹¹² Capsules were vacuum deodorized, and both tertiary butylhydroquinone (0.2 mg/g) and tocopherols (2 mg/g) were added as antioxidants. Randomization was stratified by sex (n=9/14 completers in omega-3 fatty acid group; n=11/16 completers in placebo group), concurrent lithium use (n=6/14 completers in omega-3 fatty acid group; n=6/16 completers in placebo group), and rapid cycling status (n=7/14 completers in omega-3 fatty acid group; n=5/16 completers in placebo group). Diagnosis was established using the SCID and DSM-IV criteria for Types I or II bipolar disorder (n=2/14 completers with Type II in omega-3 fatty acid group; n=3/16 completers with Type II in placebo group). Eight patients entered the study without receiving psychotropic medication, and a post hoc analysis of their data constituted an evaluation of omega-3 fatty acids as primary treatment (n=4 per study group). Mood states varied at study entry both across study groups and within each study group.

Patients were free of notable medical and psychiatric comorbidity. They were required to have experienced at least one manic or hypomanic episode within the past year.¹¹² The investigators argued that this inclusion criterion would enhance the study's power to detect a difference since such episodes were likely to recur. Forty percent of participants had exhibited rapid-cycling symptoms (i.e., at least four mood episodes in the past year). While participants were permitted to continue with existing psychotherapies (data unreported), no new regimens were permitted. Those receiving psychotropic medication continued to do so on-study, at constant dosages, and irrespective of whether they were in the therapeutic range (n=4/14 and n=3/16 patients in the active and placebo arms, respectively, received no medication). However, there was considerable heterogeneity both between- and within-study groups in terms of which types of on-study medication (doses unreported) were taken. Clinical assessments took place every second week for four months. The investigators defined the followup required to observe an effect as 30 days, thereby establishing the criterion for data that could be entered into analysis.

While planned as a 9-month trial with 60 patients required based on a sample size calculation, a stoppage in the production of the fish oil material and a significant between-group difference observed via a preplanned interim analysis, conducted when 20 patients had either failed treatment or completed 4 months, together led to ending accrual and reanalyzing data from 30 patients. P-values were adjusted accordingly. While it was reported that the exposure was produced by the National Marine Fisheries Fish Oil Program (US), no data were provided regarding its purity or whether the presence of methylmercury was tested or eliminated.

There were no significant differences in the baseline characteristics of the study groups (i.e., age, sex, rapid cycling in past year, Clinical Global Impression [CGI] scale, Global Assessment Scale [GAS], Young Mania Rating Scale [YMRS], HDRS).¹¹² Results of analyses involving 30 evaluable patients were reported (n=14 in active treatment group).¹¹² Kaplan-Meier survival analysis revealed that the active treatment group exhibited a significantly longer period of remission (i.e., duration of time remaining in the study) than did the placebo group. When data were analyzed exclusively from those subjects who entered the study without receiving psychotropic medication, the same difference was observed. For the full sample, the time to a 50% rate of ending the trial prematurely (“nonresponse”) was 65 days for the placebo group. The investigators interpreted this result as being consistent with the study population. Significant differences in improvement on the CGI, GAS, and the HDRS were observed in favor of the active treatment group. The latter result suggests that depression was also positively affected by supplementation. Post hoc analyses showed that sex, rapid cycling status and bipolar disorder type did not predict response (no data reported).

There was some evidence that the blind had been broken. A “fishy” aftertaste was more often reported in the active treatment group, such that 86% of participants guessed that they had received fish oil capsules. Only 63% in the placebo group guessed correctly. However, debriefing revealed that clinical response, or lack thereof, also played a role in tipping-off subjects to which study group they had been allocated.

Akkerhuis and Nolen reported the spontaneous reduction of psoriasis in a double blind trial wherein patients with bipolar disorder were allocated either to a maximum of 6g/d EPA ethyl ester or placebo (undefined).⁹³ Neither results relating to clinical outcomes nor other study details were provided.

Study quality and applicability. The Stoll et al. trial received an internal validity grade of A (Jadad total score=4), exhibited Adequate concealment of allocation to study groups, and was rated as being applicable to a North American population.¹¹² The Akkerhuis and Nolen report did not provide sufficient data to permit an evaluation of its study's internal validity or applicability.⁹³

Overview of Relevant Study's Characteristics and Results

Noaghiul and Hibbeln conducted a cross-national ecological analysis assessing the possible association of seafood consumption and published lifetime prevalence rates (ages 18–64 years) of bipolar disorder and schizophrenia.⁹⁰ Bipolar I disorder prevalence data from six countries (US, Canada, Puerto Rico, Taiwan, Korea and New Zealand) were obtained from the Cross-National Collaborative Group epidemiological study of 10 countries. To these were added data from Germany, Italy, Israel, Iceland and Switzerland. All studies used structured diagnostic instruments and appropriate sampling methods to obtain clearly defined community samples. For example, with the exception of Switzerland and Israel, all studies used the NIMH Diagnostic Interview Schedule (DIS). The former used the SPIKE and Schedule for Affective Disorders and Schizophrenia, respectively. A Hungarian study of bipolar II disorder met eligibility criteria, as did a study from Norway. The latter used the DIS yet did not provide data subdivided by diagnostic subcategory. Data from Norway were compared with those from other countries after data from different diagnostic subcategories were first combined. All rates were reported as cases per 100,000 persons. Prevalence rates drawn from the Cross-National Collaborative Group epidemiological study were standardized at each site, with a weight calculated per subject, and stratified for age and sex. Data from other sources could not be weighted in this manner since primary data were unavailable. Socioeconomic status and educational level were not taken into consideration. The female-to-male ratio was roughly equal at all sites, with slightly higher rates seen for Canada, Puerto Rico, Korea and New Zealand. Sources of lifetime prevalence data for schizophrenia are described later in our report.

National seafood consumption data were obtained from the National Marine Fisheries Service and the Food and Agriculture Organization of the WHO. The rates of consumption appeared to be stable across the period in which the data were collected. As a measure of the disappearance of seafood from the economy per year, apparent seafood consumption (lb/person/year) was calculated as total catch plus imports minus exports.

Results indicated variability in the rates of bipolar disorder across the countries.⁹⁰ By simple linear regression, greater national seafood consumption predicted lower prevalence rates of bipolar spectrum disorder and bipolar II disorder, but not bipolar I disorder, for which a nonsignificant association was observed. An investigation of the residual plots of these findings suggested that nonlinear models would better express the association. By logarithmic regression, greater seafood consumption predicted lower prevalence rates of bipolar I disorder, bipolar II disorder and bipolar spectrum disorder. The best curve fitting entailed a simple exponential decay regression whereby greater seafood consumption again predicted lower rates of bipolar I disorder, bipolar II disorder and bipolar spectrum disorder. When all subcategories were combined, both linear regression and exponential decay regression remained significant. When outlier data from Iceland (by far the highest seafood consumption, very low rates of bipolar I and bipolar spectrum disorder) were excluded, the association strengthened involving bipolar II disorder but did not change the results for bipolar I or bipolar spectrum disorder.

Study quality and applicability. Given its multiple national entries of data, Noaghiul and Hibbeln's study received an applicability rating of III.⁹⁰ Its total quality score was 4.

Overview of Relevant Studies' Characteristics and Results

Both studies employed a cross-sectional design.^{113, 114} Only Chiu et al. reported their funding source: three National Science Council grants, and the China Chemical and Pharmaceutical Company.

Mahadik et al. investigated AA and DHA compositions of cultured skin fibroblasts of schizophrenic patients (n=12; eight drug-naïve and in a first episode of nonaffective psychosis, four drug-free although presently admitted for recurrence), bipolar patients (n=6; two in first manic episode) and normal controls (n=8).¹¹⁴ Bipolar patients were selected because they do not tend to manifest prominent negative symptoms. Mahadik et al. reported no significant differences between bipolar patients and normal controls for AA or DHA although schizophrenic patients exhibited significantly lower DHA compositions compared with either bipolar patients or normal controls.¹¹⁴

Chiu et al. examined whether there was a depletion of PUFAs in RBC membranes of patients admitted to hospital with DSM-IV diagnosed bipolar I disorder and whose most recent episode manic (n=20; 18–65 years), compared with healthy volunteer controls (n=20; 18–65 years).¹¹³ Excluded were bipolar patients with mixed symptom episodes or comorbid Axis I psychiatric disorders (i.e., due to a medical condition or induced by substance use). The mean age of onset of the bipolar patients was 26.5 (SD=9.9) years with an average duration of 11.1 (SD=9.6) years. The mean number of mood (i.e., manic or depressive) episodes was 5.2 (SD=4.5) and the mean number of hospitalizations was 3.8 (SD-3.2). The mean YMRS score was 32.1 (range: 14–42). During index hospitalization, all bipolar patients continued to receive their mood stabilizers, benzodiazepine or antipsychotic drugs. Fifteen patients were receiving mood stabilizers, including lithium (n=9), valproate (n=5), and valproate with carbamazepine (n=1). Of these, ten were taking antipsychotics. At the time of blood sampling, five patients had been free of psychotropic medication for at least one week. Healthy controls did not have a positive family history of psychiatric disorder or take psychotropic medication although no method to rule out psychiatric disturbance was described.¹¹³ All study participants were of Han background, were free of medical illness (e.g., immune or endocrine disorders) and were exluded if they were on a low fat or vegetarian diet. There were no significant between-group baseline differences for age, sex or BMI.

Chiu et al. found significantly reduced AA and DHA compositions in RBC membranes in bipolar manic patients relative to healthy volunteers.¹¹³ There were no significant differences in either total omega-3 or total omega-6 fatty acid compositions. No significant differences were observed for either the AA/EPA or total omega-6/omega-3 fatty acid ratio. An assessment of the impact of medication on PUFAs in bipolar patients revealed no significant differences for AA and DHA levels. AA and DHA levels were not significantly correlated with age, age of onset, number of episodes or length of illness. There were no significant differences in AA or DHA levels for bipolar patients varying on the basis of their smoker status. No inappropriate methods to extract, prepare, store or analyze lipids were described in either report.^{113, 114}

Although both included studies were controlled, neither was prospective by design. Thus, meta-analysis was not considered. That said, the studies collected fatty acid status data using two very different methodologies, and from different sources. The small numbers of study precluded any meaningful evaluation of the possible impact of covariates or confounders.

Study quality and applicability. Mahadik et al. and Chiu et al.'s studies received applicability ratings of I and III, respectively. Each study received a quality score of 5.

Overview of Relevant Studies' Characteristics and Results

Wardle et al.'s RCT investigated whether cholesterol-lowering diets influence mood, including depression, anxiety, anger/hostility, stress, and general psychological well-being.⁹⁹ Adult volunteers (n=176) with elevated serum cholesterol levels (>5.2 mM [198 mg/dL]) were allocated to a low-fat diet (n=59), a Mediterranean diet (n=61), or a waiting-list control (n=56). Dietary treatments were given in eight sessions over the 12-week period. Participants completed a seven-day dietary intake diary before the first assessment. The outcome measure was the anxiety subscale of the POMS. Dietary diaries were filled out at baseline and 12 weeks. There were no significant between-group differences observed for anxiety. There was no reliable association between intake of omega-3 fatty acids and anxiety.

Reflecting one factor within a factorial RCT investigating interventions to reduce mortality in angina (including: advice [not] to eat fruits and vegetables; [no] stress management), 452 males were allocated to receive advice to eat more fatty fish (i.e., mackerel, herring, kipper, pilchard, sardine, salmon, trout) or to receive no such advice. Ness et al. supplied MaxEPA® fish oil capsules to study participants if they did not like the taste of fish.¹⁰⁰ Fish intake and mood (depression, anxiety) were assessed at baseline and at six months, the latter using the validated Derogatis Stress Profile (DSP). Ness et al. observed that self-reported fish intake was higher in the fish advice group at study's end.¹⁰⁰ No statistical difference was observed in the fish advice group for anxiety; controlling for baseline mood, the between-group difference was not statistically different. This last observation did not change following an additional adjustment made for randomization to the stress management arm, nor was there any statistical evidence of interaction between these factors in their effects on mood. These observations were not contradicted when they looked exclusively at the upper quartile of baseline anxiety scores.

The very different interventions and outcomes precluded quantitative synthesis. The dearth of data concerning covariates and confounders did not permit a meaningful assessment of their possible influence. That said, neither study demonstrated a significant clinical effect.

Study quality and applicability. Both RCTs received a Jadad total quality score of 2, indicating low quality, and level II applicability ratings.^{99, 100} Wardle et al.'s trial⁹⁹ described adequate allocation concealment while Ness et al.'s report was unclear.¹⁰⁰

Overview of Relevant Study's Characteristics and Results

At one Israeli site, Fux et al. selected eleven patients from an anxiety disorders clinic (18–75 years; racial/ethnic background unreported) meeting DSM-IV criteria for obsessive-compulsive disorder (duration: 14.1±8 years).¹¹⁵ Participants began either with 2 g/d E-EPA (96% pure semi-synthetic E-EPA; plus stabilized with 0.2% vitamin E) or matched 2 g/d placebo (liquid paraffin) gelatin capsules in a six-week, two-phase crossover RCT. Selection criteria included having been on a stable dose of SSRIs (paroxetine: n=8; fluvoxamine: n=1; fluoxetine: n=1) for at least 2 months, and having demonstrated some response to treatment yet without further improvement over the last 2 months. Exclusion criteria included no unstable medical disease, alcohol or drug abuse, or comorbid Axis II psychiatric diagnosis. Patients maintained their SSRI dose over the study. None received psychotherapy aside from basic clinical management or support. The primary outcome measures were scores on the Yale-Brown Obsessive-Compulsive Scale (YBOCS), HDRS, and the Hamilton Anxiety Rating Scale (HAM-A). The intervention was prepared by Laxdale, Ltd. No data described its purity or whether methylmercury was tested for and eliminated.

Overall, 91% of the sample completed the full 12 weeks (n=10/11); however, data were analyzed based on an ITT basis, with the last value carried forward for the participant who dropped out at week 10 of the study (i.e., moved out of city). Results indicated that there were no effects of order of treatment on HDRS or HAM-A. Time had a main effect on YBOCS scores, with significant decreases by week 6 for both placebo and E-EPA phases. There was neither a treatment effect for any clinical outcome, or a significant drug-by-time interaction. No assessment of the impact of covariates or confounders was possible. This RCT received a Jadad total quality score of 3 and an applicability rating of III.

Overview of Relevant Studies' Characteristics and Results

Both studies had a cross-sectional design and were conducted in the US.^{116, 117} Langan and Farrell's study was funded by the NIH¹¹⁷ while Holman et al.'s work was supported by the Carle Foundation, NIH, Harmel Foundation and by Scotia Pharmaceuticals.¹¹⁶

Langan and Farrell investigated the plasma fatty acid composition in a group of females with anorexia nervosa admitted to a hospital (n=17; mean age: 16.8 years; duration of anorexia: 17.2 months) compared to healthy females serving as controls (n=11; mean age: 20.7 years).¹¹⁷ The anorexic patients were admitted because of an electrolyte imbalance, a greater than 25 percent loss of ideal body weight (mean: 28.5 pounds) and severe psychosocial problems. Patients varied in their degree of malnutrition. The control group was slightly older than the patient group. Compared with the control group, the weight-to-height ratio (lb/in) was significantly reduced in the patient group.

Holman et al. compared the plama phospholipid fatty acid composition in young females with anorexia nervosa (n=8; mean age: 18.4 [15–24] years) admitted to a treatment program in an urban clinic, with that of healthy female adults (n=19; mean age: 23.5 years).¹¹⁶ All patients had lost at least 15% of their usual body weight. No inappropriate methods to handle lipids were described in either study.

Langan and Farrell showed that there were no significant between-group differences in the fatty acid composition of total plasma lipids.¹¹⁷ Only plasma phospholipid LA and ALA were significantly reduced in the group with anorexia compared with controls, while DHA was significantly higher in the females with anorexia. The total amount of omega-6 fatty acids was significantly lower in those with anorexia compared with normal controls, yet the AA/LA ratio was significantly higher among patients with anorexia compared with controls.

Holman et al. observed that the phospholipid content of total omega-6 fatty acids was significantly reduced in the patients with anorexia compared with controls.¹¹⁶ The same observation was made with respect to EPA, DHA, ALA and total omega-3 fatty acids. When the analysis was performed on plasma cholesteryl esters, there were no significant between-group differences for the omega-3 fatty acids, while DGLA was significantly lower in patients than in controls. For the plasma triglycerides fraction, total omega-3 fatty acid content was significantly reduced in patients compared to healthy subjects. The only two omega-6 fatty acids exhibiting a significant reduction in the patient group were DPA and GLA.

Study quality and applicability. Both studies received an applicability rating of I. Their mean quality score was 1.5.

Although all of the included studies were controlled, none were prospective by design. Thus, meta-analysis was not considered. Insufficient data precluded an assessment of the possible impact of covariates and confounders.

Overview of Relevant Studies

Richardson and Puri's RCT evaluated the effects of supplementation with highly unsaturated (HUFA) fatty acids in children (n=41; 8–12 years) with both AD/HD-related symptoms and specific learning difficulties (mainly dyslexia).¹¹⁹ Children were not formally assigned a diagnosis of AD/HD. Teacher-identified children were allocated to receive for 12 weeks either olive oil placebo or a “cocktail” including 186mg/d EPA, 480mg/d DHA, 96mg/d GLA, 60 IU/d vitamin E (as antioxidant), 864mg/d cis-linolenic acid, 42mg/d AA and 8mg/d thyme. Behavioral and learning problems associated with AD/HD were assessed using Conners Parent Rating Scale (CPRS). Analyses of teacher ratings were not conducted given that the children were new to their school.

Hirayama et al. investigated primarily the effects of DHA on symptoms of AD/HD.¹²⁰ They conducted an RCT of children (6–12 years) recruited by psychiatrists. Children were assigned to receive, for 2 months, either 3.6 g/wk DHA plus 700mg/wk EPA from fish oil contained in active foods (fermented soybean milk, bread rolls and steamed bread) or these same foods without fish oil. Most of the children were not receiving medication (n=34/40). AD/HD related symptoms, aggression, visual perception, visual and auditory short-term memory, development of visual-motor integration, continuous performance and impatience were assessed in this study.

Brue et al. conducted two 12-week trials to evaluate the efficacy of a dietary supplement combination and flaxseed for the treatment of inattentiveness and hyperactivity in children with AD/HD (mean age: 8.4 years; 4–12 years).¹¹⁸ Each child was supposed to participate in both studies. However, 51 of 60 children enrolled in the first study completed the second RCT as well. To initiate the first RCT, 30 children were chosen randomly from a group not taking any stimulant medication and 30 were randomly chosen from those taking methylphenidate. Each RCT included two experimental and two control groups. Here, we are interested only in the second trial because the first one did not include an omega-3 fatty acid exposure.

The second trial consisted of unmedicated patients randomly allocated to receive either 2 g/d flaxseed plus a dietary supplement combination (40 mg/d Ginkgo biloba [proposed effect: mental clarity/alertness], 800 mg Melissa officinalis [proposed effect: relaxing effect], 120 mg Grapine [proposed effect: attention, memory], 140 mg dimethyaminoethanol [proposed effect: memory, learning], 400 mg L-glutamine [proposed effect: mental clarity/alertness]) or the dietary supplement combination paired with a slippery elm supplement as placebo (amount not reported).¹¹⁸ As will be described below, children taking methylphenidate were likewise randomized to these study groups. Participants were instructed to take their intervention twice daily, once with breakfast and then with an afternoon snack or dinner. Only the results with respect to unmedicated children are presented here. Data from children receiving the intervention supplemental to methylphenidate are described below. The CPRS and CTRS were used to measure study outcomes.

Harding et al. conducted a study in children (7–12 years) with AD/HD. They were recruited by a clinical child psychologist. They were then divided, by parental choice, into two groups. They received, for 4 weeks, either Ritalin at a dose of 5–15 mg two to three times daily (n=10), or dietary supplementation containing a mix of essential fatty acids (e.g., 180 mg/d EPA and 120 mg/d DHA from 1g salmon oil), a multiple vitamin (e.g., thiamine, niacin), multiple minerals (e.g., magnesium, calcium), phytonutrients, phospholipids (soy lecithin), probiotics (n=10) and amino acids (e.g., glutamine).

Qualitative Synthesis of Relevant Studies' Key Characteristics

Study characteristics. Three parallel RCTs^118–120 and one comparative before-after study,¹²¹ each involving children, were conducted to evaluate the efficacy of omega-3 fatty acids as a primary treatment for AD/HD. Inclusion and exclusion conditions were well-defined in three studies.^{118, 119, 121} Hirayama et al. did not specify any exclusion criteria.¹²⁰ Only Brue et al.¹¹⁸ employed a design having more than two groups (i.e., 4). However, only one of their study arms addressed the present question. A total of 161 children were randomized. The mean sample size for the four studies was 40.25 (range: 20–60) participants, with the Brue et al. trial being the largest (n=60) and the Harding et al. study being the smallest (n=20). Study participants received the intervention for an average of 9 weeks, with the Harding et al. intervention being the shortest (i.e., 4 weeks).¹²¹ The RCTs were conducted in the US,¹¹⁸ the UK ¹¹⁹ and Japan.¹²⁰ The UK RCT was funded by the Dyslexia Research Trust Funding,¹¹⁹ and the study from Japan by Japan Fisheries Association and the Foundation for Total Health Promotion.¹²⁰ The funding sources for the two US trials were not reported.^{118, 121}

Population characteristics. The mean age of study participants across the four trials was impossible to determine given that full sample means were not provided in two trials.^{120, 121} The age of the participants ranged from 4 to 12 years when all studies were combined. The sex ratio was provided in three studies.^118–120 Males were consistently more strongly represented in these studies (80%-86%). With respect to racial/ethnic backgrounds, Hirayama et al.'s study likely included an Asian population¹²⁰ while similar data were not reported for the UK, or for the US, sample populations.^{118, 121, 121}

All studies used DSM-IV criteria to identify AD/HD.^118–121 Hirayama et al. reported that eight of 40 children might not have been identifiable as AD/HD according to DSM-IV criteria but this diagnosis was nevertheless “strongly suspected” by two psychiatrists.¹²⁰ Even though there were no significant differences between the two study groups on a number of bases, it should be noted that, in the control group there were more patients taking medication/polymedication (4 vs 2) than in the DHA group. As well, there were more patients in the control group with a comorbid condition (15 vs 12), including Asperger's syndrome (7 vs 2), conduct disorder (3 vs 0) or mood disorder (5 vs 1). Conversely there were more patients with learning disorders in the DHA group than in the controls (10 vs 5).¹²⁰ Overall, almost three-quarters (n=27/40) of the children exhibited comorbidity. At baseline, no significant between-group differences were observed on outcome measures.

In the study conducted by Richardson et al. the participants had, in addition to AD/HD-related symptoms, specific learning difficulties assessed by the Similarities and Matrices subtests from the British Ability Scales (BAS).¹¹⁹ Patients with a history of any other neurological or major psychiatric disorder or significant medical problems were excluded. No patients were receiving any medication. At baseline, the two groups did not differ significantly for age, sex, ethnicity or on any of the Conners scales.

The Brue et al. report indicated that participants taking a stimulant medication other than methylphenidate were excluded, as were those with serious and preexisting medical or psychological conditions such as asthma or depression.¹¹⁸ These authors did not report any baseline data. Harding et al. excluded patients with co-existing conduct disorder or oppositional defiant disorder, medication use, street drugs, or use of other nutritional or botanical supplements.¹²¹

Intervention/exposure characteristics. In the study conducted by Richardson and Puri, children in the treatment group received a supplement containing both omega-3 and omega-6 fatty acids. The sources of these agents were not identified. Vitamin E was added as an antioxidant. The placebo group received an unspecified dose of olive oil in identical capsules.¹¹⁹ In the study by Hirayama et al., the treatment group received active foods containing fish oil (fermented soybean milk; bread rolls and steamed bread) that provided 3.6 g/wk DHA and 700 mg/wk EPA.¹²⁰ Fermented soybean milk was given three times per week and provided 600 mg DHA per 125 mL. Bread rolls and steamed milk were given twice a week, providing 300 mg DHA per 45g and 600 mg DHA per 60g, respectively. The placebo group received the same foods but containing olive oil.¹²⁰ The authors masked the fishy taste in the milk product using special flavors (no method reported). For the other active foods, the fish oil was emulsified with fruit juices. No mention was made as to whether these same procedures were applied to placebo-containing foods. Parents were asked to maintain their child's habitual diet other than reducing bread consumption to accommodate the inclusion of breads containing the exposure.¹²⁰ Brue et al.'s dietary supplement “cocktail” is well described above. Harding et al.'s active ingredients are too numerous to mention them all here.¹²¹

Not one of the trial reports described the manufacturers of the sources of their interventions, the purity of their exposures, or whether, or how, the presence of methylmercury was tested for, or eliminated from, the sources.^118–121

Cointervention characteristics. Omega-3 fatty acids were often given concurrently with other agents, including omega-6 fatty acids, vitamins, minerals, polynutrients, probiotics and amino acids.^{119, 121} Hirayama et al. stated that DHA, from fish oil, was added to foods (fermented soybean milk, bread rolls and steamed bread).¹²⁰ However, the nutritional content of these foods was not reported. In these investigators' control group, four patients were receiving medication, including methylphenidate (n=1), methylphenidate plus risperidone (n=1), carbamazepine plus fluvoxamine (n=1) or carbamazepine plus sulpiride (n=1).¹²⁰ Two patients in the DHA group were exclusively taking methylphenidate.

Outcome characteristics. Richardson and Puri defined changes in CPRS scores (AD/HD subscales, AD/HD global scales) as the primary outcome.¹¹⁹ Hirayama et al. powered their study to assess changes in aggression using a questionnaire developed by the authors.¹²⁰ Other assessments included AD/HD-related symptom criteria based on DSM-IV, visual perception, visual and auditory shortterm memory, visual-motor integration, a continuous performance test and an impatience test. Brue et al. employed the DSM-IV Inattentive and Hyperactive-Impulsive subscales.¹¹⁸ The primary outcome in the Harding et al. study was the Intermediate Visual and Auditory/Continuous Performance Test (IVA/CPT) although CPRS data were obtained as well.¹²¹ Two major quotients are derived from the six primary IVA/CPT scales: the Full Scale Response Control Quotient (FSRCQ: prudence, consistency, stamina) and the Full Scale Attention Control Quotient (FASCQ: vigilance, focus, speed).

Study quality and applicability. The three RCTs received a mean Jadad total quality score of 3.3, indicating sound internal validity.^118–120 Their three applicability ratings ranged from I to III. The applicability rating for the comparative before-after study was I, and it received a total quality score of 4.

Qualitative Synthesis of Individual Study Results

Richardson and Puri compared the changes in CPRS subscale scores after 12 weeks.¹¹⁹ From the 41 patients enrolled, 15 in the active group and 14 in the placebo group completed the study. Analyses at endpoint revealed that the active treatment group had significantly lower scores on DSM Inattention, Conners ADHD Index and psychosomatic symptoms.

Hirayama et al. reported that all subjects completed the study. Data analyses did not show any improvement in AD/HD symptoms (e.g., problems of inattention, hyperactivity/impulsivity) in the DHA group compared to the placebo group. The number of errors of commission on the continuous performance test decreased significantly in the control group. Visual short-term memory was significantly improved in the control group. Excluding data from those receiving medication (i.e., supplemental treatment patients) or from those only suspected of being AD/HD did not change these results (no data reported). Food consumption was estimated to be close to 100% (no data reported).

Brue et al.'s results indicated no significant between-group differences on parent and teacher ratings of inattentiveness. Teacher-reported hyperactivity/impulsivity was significantly lower in the flaxseed plus supplement combination group, compared with the supplement combination plus placebo group. However, the opposite was observed for parent ratings.

Significant improvements were observed on both the FSRCQ and FSACQ in each of the study groups of Harding et al.¹²¹ There were no significant between-group differences on either the FSRCQ or the FSACQ. No significant between-group differences were observed for the following four subquotients although both study groups' improvements were statistically significant: Auditory Response Control Quotient, Visual Response Control Quotient, Auditory Attention Quotient and the Visual Attention Quotient.

Quantitative Synthesis

Meta-analysis was not attempted for several reasons. The two studies employing DHA and EPA as active treatment employed different research designs (i.e., Harding et al.'s noncomparative before-after study¹²¹ vs Hirayama et al.'s RCT¹²⁰). More importantly, though, in the only two studies using a common comparator (i.e., olive oil pacebo), their active treatments were completely different (i.e., Richardson and Puri's “cocktail”¹¹⁹ vs Hirayama et al.'s DHA+EPA exposure¹²⁰).

Impact of Covariates and Confounders

A few studies attempted to control for possible confounding, including the study of Harding et al.,¹²¹ which excluded children with externalizing disorders commonly associated with AD/HD (i.e., conduct disorder, oppositional defiant disorder), as well as the study of Hirayama et al., where the children maintained their background diets.¹²⁰ On the other hand, the latter study also included subjects with a wide range of comorbid conditions;¹²⁰ and, Richardson and Puri included children with a variety of learning difficulties.¹¹⁹ Yet, the inconsistent findings, including a small number of significant clinical effects, and the variability in both the types of intervention and comparator made it impossible to begin to reliably identify key covariables affecting clinical outcomes.