Overview

A total of 86 reports, describing 79 unique studies, investigated questions pertinent to this systematic review of the evidence concerning the effects of omega-3 fatty acids on mental health. Not all of the mental health disorders or conditions included in this review had evidence addressing all of the first three basic questions posed in this review—primary or supplemental treatment with omega-3 fatty acids (Question 1), or the association between the onset, continuation or recurrence of the disorder or condition and either the intake of omega-3 fatty acids (Question 2) or the omega-3 or omega-6/omega-3 fatty acid content of biomarkers (Question 3). Schizophrenia (n=28 studies) and depression (n=22 studies) were, by far, the most frequently investigated psychiatric disorders. Many possible explanations likely exist for why these two disorders have received the most attention, including the prevalence of depression and the presumed intractability of schizophrenia.

Of the collections of studies on schizophrenia and depression, 50% (n=14/28) and 36.4% (n=8/22) examined the possible association of schizophrenia and depression outcomes, respectively, with the omega-3 or omega-6/omega-3 fatty acid content of biomarkers. These two clinical areas (i.e., schizophrenia and depression) have produced an abundance of animal studies, as well as both animal and human models concerning the etiology of these disorders.⁵⁵ This may help explain why we found many more human studies examining biomarkers data (Question 3), as well as the possible association between omega-3 intake and clinical outcomes (Question 2), than studies investigating the treatment of these disorders or conditions (Question 1). It is conceivable that the research community has assumed it was necessary to first use biomarkers and epidemiological data to demonstrate the plausibility of treating these clinical entities with omega-3 fatty acids. Only recently have studies been published concerning the primary or supplemental treatment of depression (n=4 RCTs: 2002 or 2003) or schizophrenia (n=5 RCTs: 2001 or 2002), and this may signal a trend towards an increased emphasis on treatment investigations.

While this review was not initiated to test the specific deficiency hypotheses relating to the etiology of depression or schizophrenia, below we have, nonetheless, examined the evidence to address the possible soundness of these positions. The justification for the study of the remaining psychiatric disorders or conditions for which we reviewed evidence ranged from the view that certain types of individual (e.g., bipolar disorder patients) may also suffer from deficiencies in “mood-regulating” omega-3 fatty acids, to little or no justification based on human or animal models or data (e.g., obsessive-compulsive disorder). Nonetheless, a study was included if it met our eligibility criteria.

For each psychiatric topic, in turn, we present a synthesis of the key findings with respect to each of the first three basic questions. This includes a critical appraisal of the individual studies from which the results were drawn. Attention is paid to the numbers, size, quality and applicability (i.e., to relevant North American populations) of studies in trying to ascertain larger patterns of result. The broader implications of these findings, including potential future research, are highlighted. We begin with the cross-cutting issue of safety.

Evidence Synthesis and Appraisal

Only interventional studies employing omega-3 fatty acids as supplementation provided safety data. Some interventional studies, which employed various populations, interventions/exposures and followup durations, did not report either having solicited adverse effects data from study participants or having received such reports. Results from these studies suggest that omega-3 fatty acid exposures were, for the most part, well tolerated. In spite of a small number of discontinuations presumed to have been instigated by an adverse event, it is unlikely that moderate or severe side effects were ever observed in relation to an omega-3 fatty acid exposure. Occasionally, adverse events were linked to the intake of oily substances, rather than to the omega-3 fatty acid contents in the oils.

Reported difficulties tended to be mild and transient, often involving gastrointestinal upset or nausea. Aside from the minor adverse effects associated with Stoll et al.'s very high dose of omega-3 fatty acids (i.e., 3 patients had to decrease the number of capsules swallowed per day, yet with none required to discontinue),¹¹² no other discernible patterns were seen regarding the impact of dose, type (e.g., DHA vs EPA) or source (e.g., marine, plant, nut) of omega-3 fatty acids on safety. In the study by Richardson and Puri, one AD/HD child in the active treatment group had to leave the study due to problems swallowing the capsules.¹¹⁹ Few of the events described in two trials by Hamazaki et al., which enrolled healthy volunteers, suggested that the adverse effects had been directly related to the exposure.^{129, 130} The ability of purified forms of EPA (i.e., E-EPA) to maintain blinding, due in large part to the oil's minimal fishy taste and odor, could not be evaluated because there were too few studies with which to construct meaningful comparisons.

Four RCTs addressed the questions concerning the primary or supplemental treatment of depression. One addressed primary treatment,⁹⁵ whereas three investigated supplemental treatment.^{53, 96, 97} Marangell et al. found no benefit related to 2 g/d DHA employed as primary treatment despite an increase in the absolute RBC levels of DHA in the active treatment group.⁹⁵ Reasons for this null result could include the use of too small a dose, too short an intervention period, the “wrong” omega-3 fatty acid, broken blinding (i.e., unidentified by group, 14 patients experienced a fishy aftertaste) or failure to modify background omega-6 fatty acid intake at the same time. Clearly, more than a single, likely underpowered trial of low quality (i.e., internal validity) and undetermined applicability is required to ascertain the value of omega-3 supplementation as primary treatment for depression. Data from two of the supplemental treatment studies that included a few patients who were not receiving medication could not be used to address the question concerning primary treatment because the study reports did not provide these individuals' results separately.^{96, 97}

Peet et al.'s dose-ranging study of E-EPA as supplemental treatment for depression found that only 1 g/d for 12 weeks had a significant impact on various clinical outcomes.⁵³ Two trials of shorter duration also showed significant benefits associated with 2 g/d E-EPA and 6.6 g/d of EPA+DHA, respectively;^{96, 97} the significant clinical effect reported by Su et al. was associated with a significant increase in RBC EPA exclusively in the active treatment group.⁹⁶ However, it was decided to forego meta-analysis due to: variations in dose both within and between studies; variation in the definition of the omega-3 fatty acid interventions; different followup lengths; and, the use of different sources of placebo material. In addition, unlike the other two supplemental treatment trials, Peet et al.'s did not formally identify patients with a depressive disorder.⁵³ This last observation may account for Peet et al.'s finding that 1 g/d E-EPA had a beneficial effect on depressive symptomatology.⁵³ It is conceivable that this low dose would not have helped the treatment-resistent depressive disorders investigated in the other trials. Yet, this likely cannot explain why Peet et al.'s higher doses (2 g/d, 4 g/d) did not likewise ameliorate depressive symptoms, or why more responders (i.e., 50% improvement) were found in the placebo group than in the 2 g/d E-EPA group. It is possible that the study by Su et al. was confounded by uncontrolled combinations of medication.⁹⁶

There were too few included studies to reliably ascertain the impact of extra-interventional variables with the potential to influence clinical results, or the possible covariation of clinical and biomarker effects. Overall, in spite of the sound internal validity of the three included trials, this study collection is too small to permit us to determine whether omega-3 fatty acid supplementation is efficacious as a supplemental treatment for depressive disorders or symptomatology. Moreover, the three studies exhibited weak applicability to even a predominantly female North American population. Yet, these preliminary findings suggest there may be promise in pursuing investigations into the use of omega-3 fatty acids as a supplemental treatment for depression.

All 12 of the studies addressing the question concerning the possible association between depression outcomes and omega-3 fatty acid intake could be construed as focusing on whether (foods containing) omega-3 fatty acids might protect against the onset of depressive disorders or symptomatology. No study investigated subquestions relating omega-3 fatty acid intake to either the continuation or recurrence of depressive disorders or symptoms.

The types of research design providing evidence relating to onset varied in terms of their inherent ability to meaningfully investigate this question. Best suited to address this question were three controlled prospective studies; yet, these constituted a minority. Of these three RCTs,^98–100 the study by Wardle et al. merely assessed the impact of diets without distinguishing the exact nature of the role played by oil fish intake within the Mediterranean diet;⁹⁹ and, the data generated by Ness et al. confirmed the assumption that advice to eat fish will not guarantee the compliance of study participants.¹⁰⁰ Thus, the results of both “intervention” studies could not meaningfully shed light on the question of onset.¹⁰⁰ The RCT of Llorente et al. examined the use of supplementation to prevent postpartum depression. While well-designed, the study included a narrowly defined population (i.e., breastfeeding women) and did not reveal a significant clinical benefit related to omega-3 fatty acid intake, despite a significant increase in plasma phospholipid DHA. Moreover, most of the women in Llorente et al.'s trial exhibited, at worst, minimal depressive symptomatology. Therefore, in spite of their RCT designs, these three studies did not constitute the best tests of the possibility that omega-3 fatty acids might protect against the onset of depressive disorders or symptomatology.

The observational studies did not contribute much to resolving the question of onset, despite their somewhat consistent picture of a lack of association. The reason is that their designs also did not constitute the best tests of omega-3 fatty acids' protective potential. The 36-month single prospective (uncontrolled) cohort study by Woo et al. found no significant, adjusted association between fish intake and depressive symptoms in an elderly Chinese population.¹¹⁰ Hakkarainen et al.'s prospective cohort study observed no significant, adjusted association between fish consumption or (calculated) omega-3 fatty acid intake and indices of depression.¹¹¹ Edwards et al.'s multiple-group cross-sectional study revealed that none of the dietary omega-3 or omega-6 fatty acid variables were significant predictors of depressive symptomatology.⁴⁸ Two single population cross-sectional surveys completed by Tanskanen and colleagues in Finland each described a significant association between the frequency of fish consumption and depressive symptomatology in females,⁸¹ and in both males and females.⁸⁰ It is unclear why the first significant association was observed only for females.⁸¹ Tanskanen et al. employed a single question to assess the exposure, and perhaps a food frequency questionnaire would have been better.⁸¹ Suzuki et al.'s single population cross-sectional survey revealed a nonsignificant association between depression and the intake of fish or seafood despite observations that ALA and total omega-3 fatty acid intake were inversely related to the likelihood of depressive symptomatology.¹⁰⁷

In cross-sectional designs, the absence of a meaningful temporal separation between the measurement(s) of the exposure (e.g., intake of fish or specific omega-3 fatty acids) and the clinical outcome (e.g., onset of depression) prevents the possible observation of cause and effect, which thereby precludes drawing causal inferences concerning the impact of the exposure on the (likelihood of the) clinical outcome. Cross-sectional surveys are also limited by recall bias. At the same time, five of the six observational studies exhibited the weakest applicability to a North American population.^{80, 81, 107, 110, 111} Each study took place either in Finland or Asia, where dietary fish intake is considerably higher than it is in North America. It is likely that greater fish intake yields a lower omega-6/omega-3 fatty acid intake ratio in the background diet.

The most consistent picture of an inverse relationship between the exposure and clinical outcomes was observed in the type of study providing the weakest evidence: cross-national ecological analyses.^{47, 108, 109} These provide possible evidence of the covariation of exposure (e.g., apparent national seafood consumption) and outcome (e.g., prevalence of depression) from often large samples of data derived invariably from non-overlapping sources, that is, where a given sample of individuals does not provide both exposure and outcome data. Thus, individual- or patient-level inferences cannot be drawn. Exposure data are at best crude indices of intake, failing to reflect the dietary practices of individuals or even population subgroups;¹⁰⁹ and, these types of study are readily confounded by cultural, economic, social and other factors.⁴⁷ An additional barrier to drawing conclusions based on these findings is that their cross-national focus precludes generalization to the North American population of subjects who may be at risk of developing depressive disorders or symptomatology.

Taken together, the inconsistent results, as well as the limitations of both the inherently stronger (i.e., prospective controlled studies) and weaker designs having produced them, suggest that there is currently insufficient evidence to decide whether or not omega-3 fatty acid intake can protect individuals—with or without known predispositions—from developing either depressive disorders or symptomatology. The observation that the risk of depressive symptomatology is inversely related to fish/seafood consumption or omega-3 fatty acid intake was less likely to be produced by research designs that more appropriately permit the drawing of causal inferences regarding the etiologic role of exposure to omega-3 fatty acids in the development of depressive disorders or symptoms. Studies that are prospective, controlled and focused on subject-level data were less likely to demonstrate evidence for a significant protective relationship. Having identified too few of these stronger study designs also made it inappropriate to conduct a quantitative synthesis and impossible to comprehensively assess the possible influence, on clinical outcomes, of extra-exposure variables (i.e., covariates, confounders). As well, too few studies produced results that could be meaningfully extrapolated to North Americans.

Eight controlled studies were identified that had the potential to address Question 3 concerning the possible association between biomarkers data and the onset of depressive disorders or symptomatology.^{48, 98, 101–106} However, only one study was prospective by design.⁹⁸ The other seven were multiple-group cross-sectional designs. Therefore, it was impossible to draw causal inferences concerning the onset of depression from the results of the seven weaker designs, or to consider meta-analysis. With respect to the multiple-group cross-sectional studies, we focused solely on comparisons between groups of patient diagnosed with depression and controls, with the latter typically identified as “healthy” and sometimes matched by age and sex. For each study, this contrast established the sharpest differentiation possible between study groups, even though only a few of the study reports described the methods by which they had formally ruled out the presence or risk of depressive disorders or symptomatology in their control subjects. Results are described as they appeared in the literature, with the data from the strongest design (i.e., RCT)⁹⁸ providing, at best, a very limited answer to the research question.

The two earliest publications included in the review—one by Ellis and Sanders¹⁰⁵ and the other by Fehily et al.¹⁰⁶—revealed a pattern of findings that has since been disconfirmed. Each study reported that plasma CPG EPA and DHA levels were higher in those with a diagnosis of endogenous depression than in healthy controls. They also noted that similar between-group differences in RBC CPG levels of EPA, DHA and AA existed although they were less pronounced (no data reported). These results would disconfirm the omega-3 fatty acid deficiency hypothesis introduced in Chapter 1. However, unlike virtually all subsequent studies, their “endogenously depressed” populations exhibited substantial diagnostic heterogeneity to the extent that it would likely be impossible to find the appropriate populations to which the results of these two studies might be meaningfully generalized.

Arguably the two best controlled cross-sectional studies were conducted by Maes and colleagues in Belgium.^{101, 103} A priori they excluded many potential confounders (e.g., background diet, alcohol use, heavy smoking, medications at assessment, comorbid conditions) in addition to matching for age and sex. Also, like few other other studies,¹⁰⁴ they explicitly described having excluded from the control group, those subjects with notable psychopathology. Maes et al.'s results are thus likely more reliable, but not solely because they were less prone to confounding. They also employed formal research diagnostic criteria. To draw one comparison, Fehily et al.¹⁰⁶ combined data from unipolar, bipolar and adjustment disorder subjects, whereas Maes et al. excluded subjects with bipolar diagnoses and distinguished between subjects with minor and major depression. Peet et al.'s study was less tightly-controlled experimentally yet they also attempted to rule out psychopathology in controls while noting the absence of significant between-group differences for smoker status.¹⁰² Most studies did not control for the likely confounding effects of stress, smoker status or diet.¹⁷⁷ Both studies by Maes et al., as well as the study by Tiemeier et al., admitted patients to hospital to establish a highly controlled environment.^{101, 103, 104}

Maes et al.'s first set of results indicated that levels of ALA, total omega-3 fatty acids and EPA in serum cholesteryl esters, as well as EPA in serum phospholipids, were significantly lower in major depressed patients compared with healthy volunteers.¹⁰³ As well, AA/EPA in both cholesteryl esters and phospholipids were significantly higher in the major depressed patient group compared with controls. ALA, EPA and DHA levels collectively discriminated between these two study groups as well as distinguished minor depressed individuals.

Peet et al.¹⁰² reported that the picture of depleted omega-3 fatty acids in serum cholesteryl esters described by Maes et al.¹⁰³ was observed in the RBC total omega-3 fatty acids and DHA of drug-free patients compared with healthy controls.¹⁰² Yet, Peet et al. pointed out that the difference in the number of current smokers across their two study groups (i.e., compared with controls, all but two depressed patients were nonsmokers) constituted a possible source of confounding.¹⁰² They also acknowledged that there may have been prestudy medication in cell membranes, which likewise may have confounded study outcomes. Nevertheless, Edwards et al. reported similar findings of significantly lower RBC total omega-3 fatty acids, DHA and EPA in medicated depressed patients compared with matched healthy controls.⁴⁸ Controlling for stress and smoker status had no effect on RBC values in Edwards et al.'s study.⁴⁸

Maes et al.'s second study revealed significantly lower fractions of EPA, DHA, AA and total omega-3 fatty acids in the serum phospholipids of major depressed patients compared with healthy volunteers.¹⁰¹ As in their first study, they also reported higher AA/EPA fractions in the patient group.¹⁰¹ Significantly lower fractions and concentrations of ALA, EPA and total omega-3 fatty acids were observed in the serum cholesteryl esters of these patients.¹⁰¹ Significantly higher AA/EPA and total omega-6/omega-3 fractions were observed in the patient group as well.¹⁰¹ Tiemeier et al. found that percentages of AA, AA/DHA and total omega-6/omega-3 in plasma phospholipids were significantly higher in depressed patients compared with controls.¹⁰⁴ The percentage of DHA was lower in the depressed patient group.

Correlational data showed significant negative relationships between: HDRS scores and EPA in serum cholesteryl esters¹⁰³ or other PUFAs;¹⁰¹ BDI scores and RBC ALA, DHA and total omega-3 fatty acids, although multiple regression revealed that only ALA levels predicted BDI scores;⁴⁸ and, between plasma phospholipid DHA and results on the BDI, EPDS or SCID.⁹⁸ These last findings were obtained from Llorente et al.'s RCT.⁹⁸ Significant positive relationships defined HDRS scores and both AA/EPA and total omega-6/omega-3 fatty acid levels in the plasma phospholipids of major depressed patients.¹⁰³

Collectively, the between-group differences suggest a possible balance of PUFAs such that significantly decreased levels of omega-3 fatty acid content coexist with increases in some omega-6 fatty acid levels and in some omega-6/omega-3 fatty acid ratios. Medication status did not appear to modify this picture.¹⁰¹ For example, EFA levels were not affected in those of Edwards et al.'s patients who had antidepressants added prior to a second assessment of their EFA status.⁴⁸ Peet et al. reported that, after six weeks of treatment with antidepressants in ten depressed patients, PUFA levels did not change significantly (no data reported).¹⁰²

However, these results were obtained from cross-sectional studies from which causal inferences relating to onset cannot be drawn. Selection bias can also influence study outcomes in these designs. In addition, PUFA status in studies of mental health is likely determined by multiple factors, suggesting that any between-group differences in PUFA content observed in this review may not simply reflect the disease process itself. Other influences include: age; sex; the dietary intake, metabolism and incorporation into cell membranes of various types and amounts of both omega-3 and omega-6 fatty acid content (given their competitive relationship with respect to enzymes, for example); the disease process underlying any possible comorbid conditions; the efficiency of the PUFA metabolic processes, including the availability and effectiveness of enzymes implicated in the processes of desaturation and elongation; the long-lasting effects of psychotropic medication (e.g., mood stabilizers, antipsychotics) on cell membranes; and the ability of protective mechanisms to deal with degradation from oxidation and other sources (e.g., smoking, alcohol consumption).^{48, 101–103, 178–180}

Differences in RBC PUFA content can be attributed to the mechanisms of action of mood stabilizers (i.e., postsynaptic signal transduction processes) or the abnormal psychoimmunology of patients with bipolar disorder.¹¹³ Mood stabilizers can reduce the AA turnover rate;¹⁸¹ and, smoking has been observed to deplete PUFAs from cell membranes (e.g., DHA, [omega-3-]DPA, omega-6 fattty acid series).¹⁸² Given that these variables have been highlighted as influences on EFA status requires that they be controlled for experimentally or statistically in studies assessing the possible association between the fatty acid content of biomarkers and clinical outcomes (e.g., onset of depression).

Thus, with only cross-sectional evidence available to address the question of onset there exists the need for more appropriate tests of the deficiency hypothesis. Ideally, these would employ controlled prospective study designs. The available results, at best, suggest the possible definition of the EFA profile that future research might identify as being responsible for the development of depression. Until then we can only speculate that “it is more likely that changes in fatty acid intake in the population influence depression prevalence than vice versa.”⁵³ The possible role played by omega-3 fatty acid intake or the omega-3 or omega-6/omega-3 fatty acid content of biomarkers in the continuation or recurrence of depression could not be assessed given no studies with these foci were identified. Whether PUFAs' influence on mental health also entails, for example, the activation of the inflammatory response system, including the production of eicosanoids, remains to be determined.

Only Question 2 could be addressed with respect to suicidal ideation or behavior. Hakkarainen et al. reported no significant associations between either intake of fatty acids or fish consumption and successful suicides.¹¹¹ Tanskanen et al. noted that the adjusted risk of suicidal ideation decreased significantly in frequent fish consumers.⁸⁰ The evidence base is thus too small, and the designs less than optimal, to permit us to conclude anything with respect to the possible association between omega-3 fatty acid intake and the onset of suicidal ideation or behavior. Their applicability is limited by the fact that both studies were conducted in Finland.

Two controlled studies investigated the supplemental treatment of bipolar disorder with omega-3 fatty acid supplementation,^{93, 112} although only one report gave us an opportunity to systematically assess its study parameters and results. While the Stoll et al. trial had to be stopped prematurely, their very high dose of 9.6 g/d EPA+DHA produced a significantly longer period of remission in the active treatment group compared with controls.¹¹² Medication status did not alter this finding. Rating scale results, including depressive symptomatology, showed greater improvement in the omega-3 fatty acid intervention group compared with controls. While these pilot observations appear to be promising, there was also evidence that the blind had been broken. Almost 90% of active treatment patients correctly guessed that they had received fish oil capsules, with data from patients indicating that both the clinical response and a fishy aftertaste contributed to their deduction. For the sake of both its promising findings regarding the impact on a subacute course of bipolar disorder, and its limitations (i.e., its loss of power because of its stoppage; broken blind), this study requires replication. It might also be useful to use a lower dose even though the present one did not produce even moderately severe side effects. A lower dose might also better control the fishy aftertaste. At present, the evidence base is too limited to definitively conclude anything about the potential of omega-3 fatty acids as supplemental therapy for bipolar disorder.

The same must be said with respect to the capacity of omega-3 fatty acids to prevent the onset of bipolar disorder (Question 2). Evidence suggesting the possibility that seafood consumption plays a protective role was identified by a single, cross-national ecological analysis.⁹⁰ Yet, while the investigators employed stratifications for both age and sex, they did not control for socioeconomic status, urban/rural ratio, educational level, marital status, alcohol consumption, smoker status or family history. These are likely significant omissions given that these risk factors can predict the onset of bipolar illness.⁹⁰ The authors also recognized that these data cannot shed light on whether the lifetime risk for bipolar disorder was affected by low seafood consumption in adulthood and/or by nutritional insufficiency in early neurological development.⁹⁰ Nutrient deficiencies during the second and third trimester of pregnancy can increase the risk of developmental affective disorders in children.¹⁸³ Noaghiul and Hibbeln's results,⁹⁰ while parallelling observations obtained from the above-noted cross-national ecological analyses regarding depression, likewise exhibit limited applicability to individuals/patients and to the North American population.

The results from two multiple-group cross-sectional studies did not agree on whether a diagnosis of bipolar disorder was associated with a specific biomarker profile when compared with data from controls (Question 3). This divergence may be attributable to the fact that the two studies obtained their PUFA samples from different biomarker sources. Chiu et al. noted significant between-group differences in AA and DHA from RBC membranes.¹¹³ Adding medication did not appreciably change the EFA levels in Chiu et al.'s bipolar patients. They did, however, fail to control for diet. Mahadik et al. assessed AA and DHA compositions of cultured skin fibroblasts, finding no significant between-group differences for small numbers of bipolar patient and controls.¹¹⁴ Although Mahadik et al. controlled for dietary intake as one key influence on RBC and brain PUFA levels,¹¹⁴ PUFA levels from skin fibroblasts may not reflect brain PUFA levels.¹¹³ Moreover, the clinical status of their patients (i.e., duration of illness, mood state [mania, depression, mixed], symptom severity) was poorly defined and controlled for; and, no data were reported indicating patterns of mood stabilizer or antipsychotic medication use, which have been found to influence PUFA levels (see above). The studies were conducted in countries varying in terms of their background diet, and likely their omega-6/omega-3 fatty acid content intake ratio, and this factor may have also influenced the results. In any event, the fact that both efforts employed cross-sectional designs precludes deriving casual inferences regarding the onset of bipolar disorder.

Two RCTs yielded data investigating the possible protective influence (Question 2) of omega-3 fatty acid intake and the onset of symptoms but not disorders of anxiety.^{99, 100} Both the Wardle et al. and Ness et al. studies failed to find a significant association. As noted with regards to the subject of depression, neither RCT constituted an appropriate assessment of this question.

Fux et al.'s results indicated that E-EPA was ineffective as a supplemental treatment for obsessive-compulsive disorder.¹¹⁵ However, nothing definitive can be concluded from a single, underpowered crossover study, which failed to describe a washout period.

Two cross-sectional studies investigating the possible association between the onset of anorexia nervosa and the fatty acid content of biomarkers analyzed plasma phospholipid data.^{116, 117} Their observations concurred that both ALA and total omega-6 fatty acid levels were significantly lower in anorexic patients than in controls. However, their findings differed in that Holman et al.¹¹⁶ noted a similar reduction in DHA in anorexic patients while Langan and Farrell found that DHA levels were significantly reduced in controls.¹¹⁷ Holman et al.¹¹⁶ noted a significantly lower level of EPA in patients, and Langan and Farrell¹¹⁷ reported a reduction of LA in these patients compared with controls. Only Holman et al. evaluated the contents of plasma cholesteryl esters, with respect to which they observed no significant between-group differences. However, in plasma triglyceride fractions they did find significantly reduced total omega-3 fatty acid content in their patients. Irrespective of these results, these small studies utilized a design preventing the drawing of causal inferences regarding etiology.

Notwithstanding the noncomparability of interventions, comparators and populations (i.e., with^{118, 120, 121} or without a formal diagnosis of AD/HD;¹¹⁹ with¹²⁰ or without significant comorbidity^{119, 121}), the results of the three RCTs^118–120 and the comparative before-after study¹²¹ addressing the question about the primary treatment of AD/HD were inconsistent at best. They did not show uniform improvement in clinical outcomes, and in some cases, significant improvements were observed only for control children.¹²⁰ The studies by Hirayama et al.¹²⁰ and Harding et al.¹²¹ failed to report any significant between-group clinical differences. These are the only two studies which clearly distinguished omega-3 fatty acids as the “intervention.” Moreover, Harding et al.'s results are likely unreliable given the selection bias that results from having parents chose which intervention their child will receive.

Each of the two studies exhibiting a few significant clinical effects had used a “cocktail,” which included much more than omega-3 fatty acids; and the nature of the synergies involving the components comprising the respective “cocktails” was not evaluated.^{118, 119} In one of these two studies, the research design did not allow the researchers to tease out the possible specific benefit of omega-3 fatty acids.¹¹⁹ None of the studies employing DSM-IV to identify AD/HD actually distinguished their populations by AD/HD subtype (e.g., Inattentive vs Hyperactive/Impulsive vs Combined), which is an important source of clinical heterogeneity. The different subtypes entail dissimilar clinical pictures given the various clusters of symptom or behavior required to identify their presence.¹³

With respect to the supplemental treatment of AD/HD, Voigt et al. observed only nonsignificant between-group clinical differences.¹²² These observations were associated with increased plasma phospholipid DHA levels observed exclusively in the DHA study group. Stevens et al. found almost no evidence of clinical benefit for their “cocktail” exposure compared with a very high dose of olive oil as placebo.¹²³ This was accompanied by observations of no significant between-group differences for fatty acid content in plasma phospholipids. Participants in the Stevens et al. trial were also entered into the study based merely on parental, not professional, confirmation of an AD/HD diagnosis. The clinical features of AD/HD can exist as isolated clusters of symptom insufficient to merit a formal diagnosis of AD/HD and so, there is no guarantee that all children would have received a DSM-IV diagnosis of AD/HD. Brue et al. reported a benefit for problems of inattentiveness yet not for hyperactivity and impulsivity.¹¹⁸

Overall, these supplemental treatment RCTs may have employed intervention lengths that were too short. Primary treatment trials lasted longer. It is also conceivable that weight-adjusting doses of omega-3 fatty acids would have produced a different picture of the efficacy of these primary or supplemental interventions, although all elements of the sometimes complex interventions would likely have required similar adjustments.^{118, 119, 123}

While the results of the supplemental treatment studies are more uniformly generalizable to the North American population than those generated by primary treatment studies of AD/HD, there were too few studies whereby the specific effects of omega-3 fatty acids could be isolated, thereby preventing us from concluding one way or the other about the specific efficacy of omega-3 fatty acids as a primary or supplemental treatment.^{118, 119, 123} The only consistent observation is that, contrary to the situation in the trials of depression, where the majority of subjects were female, most of the participants in the two collections of AD/HD study were male. This is not surprising given what has often been observed in clinical practice.

Yang et al.'s multiple-group cross-sectional design was not concerned with trying to establish a link between omega-3 fatty acid intake and the onset of AD/HD.⁹⁴ At best, the results from this study might hint at the possible conditions maintaining AD/HD, although controlled prospective designs are required to determine causality. Nevertheless, Yang et al. found that, relative to healthy controls, AD/HD children consumed significantly lesser amounts of LA and ALA. These observations, while requiring replication, could be suggestive if it turns out that lower LA and ALA content in biomarkers also distinguishes those individuals with AD/HD and healthy controls.

In their first study Mitchell et al. failed to observe any univariate between-group differences for RBC fatty acid content, although multivariate analysis revealed that levels of ALA and AA, along with a few other fatty acids, distinguished hyperactive and control children.¹²⁶ Stevens et al. also found significantly lower AA levels in hyperactive boys.¹²⁴ In their second study, Mitchell et al. reported that levels of DHA, AA and DGLA in serum phospholipids were significantly reduced in formally diagnosed hyperactive children.¹²⁵ Stevens et al. also found significantly reduced AA, EPA, DHA and total omega-3 fatty acids in the plasma phospholipids of hyperactive boys.¹²⁴ However, Stevens et al.'s study did not confirm this observation.¹²⁴ They noted higher PUFA intake in the diet of hyperactive boys. More work is needed to resolve this divergence of findings.

Only the second Mitchell et al. study employed formal diagnostic criteria (i.e., DSM-III) to identify their hyperactive subjects.¹²⁵ However, none of these biomarker studies formally ruled out the presence of psychopathology in the control subjects. The use of cross-sectional designs by so few studies necessitates additional empirical work.

Based on a single observational study, which controlled for age, income, smoking, alcohol consumption and eating patterns, mental health status was observed to be lower in those consuming no fish.¹²⁷ However, this cross-sectional design precludes inferring that the onset of mental health diffulties is related to fish consumption.

Seven studies, including three RCTs enrolling healthy volunteers, investigated the relationship between omega-3 fatty acid intake and tendencies or behaviors with the potential to harm others. All but Gesch et al.'s study were designed to address the relationship of intake and the onset of these tendencies or behaviors.¹³¹ Gesch et al.'s trial investigated the possibility of using an exposure to prevent the recurrence of antisocial behavior (i.e., secondary prevention). It is difficult to discern any reliable, significant patterns, or lack thereof, across the various outcomes, populations and designs, however.

Hamazaki et al.'s work with university students showed that, when a stressor was applied, DHA supplementation provided some protection against aggression directed at the external world;¹³⁰ however, a subsequent study, involving no stressor component, showed that control oil capsules had a similar beneficial impact on aggression in control subjects.¹²⁹ The first observation was associated with no between-group differences for DHA, EPA or AA content in serum phospholipds. The behavioral finding in their second study, in favor of the control subjects, was coupled with significant increases in RBC EPA and DHA content in the DHA group, and a significant increase in RBC LA content in the control group. For Hamazaki et al.'s elderly Thai population, some benefit related to the prevention of extraaggression was observed for university employees yet not for villagers.¹²⁸ Appropriate between-group analyses of RBC content data were not performed. No reliable patterns relating clinical and biomarker effects could be discerned across Hamazaki et al.'s trials.

Enrolling very different populations, yet focused on trying to see if omega-3 fatty acid exposures prevent the onset of tendencies or behavior with the potential to harm others, Wardle et al.'s RCT observed no significant benefits for anger/hostility associated with special diets,⁹⁹ Iribarren et al.'s cross-sectional survey found that high intake of DHA and the consumption of fish rich in omega-3 fatty acids may be related to a lesser likelihood of high levels of hostility in young adults,¹³² and Hibbeln's cross-national ecological analysis revealed that lower apparent seafood consumption was associated with higher rates of death due to homicide.¹³³ Gesch et al.'s “cocktail” supplementation provided young adult prisoners with some (secondary) protection against committing new offences.¹³¹

Overall, these findings are sufficiently inconsistent and involve too few research designs permitting the drawing of causal inferences (i.e., cross-sectional survey,¹³² cross-national ecological analysis¹³³) and too many different definitions of the exposure, population and outcome^{99, 128–133} for us to be able to derive an individual/patient-level conclusion regarding the protective benefits of omega-3 fatty acid intake when it comes to tendencies or behavior with the potential to harm others. Moreover, as a whole, the generalizability of their findings to North Americans is limited.

Very few significant between-group differences were observed in the three included studies addressing the biomarkers question with respect to the onset of tendencies or behavior with the potential to harm others;^134–136 and, given the differences in the investigated populations, generalizations cannot be made. That said, Hibbeln et al. reported no significant between-group differences for PUFA content when violent and non-violent subjects were compared.¹³⁴ The only observation identified in more than one study entailed lower DHA levels in the plasma phospholipids of patients with antisocial personality compared with healthy controls,¹³⁵ and in the plasma phospholipids of aggressive cocaine addicts compared with nonaggressive cocaine addicts.¹³⁶ However, only the Hibbeln et al. study did not include a small number of participants.¹³⁴ The exclusive use of cross-sectional designs precludes drawing any inferences regarding etiology.

The conflicting results regarding reduced PUFA content in alcoholic patients reported by Alling et al.¹³⁸ and Hibbeln et al.¹³⁷ may not simply be attributable to the different biomarker sources that were investigated. The lower levels of LA, DHA, DGLA and AA found by Alling et al.¹³⁷ in male chronic alcoholics, compared with healthy male controls, could have been caused by the consumption of alcohol itself.⁶⁰ Yet, the fact that Hibbeln et al.'s abstinent alcoholics exhibited higher PUFA concentrations, while also having smoked many more cigarettes per annum than did healthy controls, is not easily explained. Whatever the correct explanation, findings linked to cross-sectional designs again preclude drawing any inferences regarding the etiology of alcoholism.

Zanarini et al.'s RCT examined E-EPA as a primary treatment for borderline personality disorder and found that there were significant clinical effects over the course of the study, as the E-EPA group had, at study end, significantly lower mean scores on both the MADRS and MOAS compared with the placebo group.¹³⁹ Notwithstanding its strong applicability to the North American population, this was a small study requiring replication.

While the results of the Peet et al. trial⁵⁸ indicated placebo-controlled benefits accruing to omega-3 fatty acid supplementation as primary treatment for schizophrenia, this was a small and methodologically adequate pilot trial with little applicability to the North American population. More work is required before we can decide anything about omega-3 fatty acids' promise in this context. Considerably more can be said about their role as supplemental treatment for schizophrenia.

Four recently published RCTs, exhibiting sound internal validity, examined omega-3 fatty acids as supplemental treatment for schizophrenia.^{58, 87, 89, 140} Three of them reported significant clinical effects in favor of EPA using total PANSS scores,^{58, 87, 140} although Peet et al.'s study observed this effect only for those receiving clozapine as primary treatment.⁸⁷ Emsley et al.'s RCT also found that the reduction in PANSS total scores associated with E-EPA supplementation was greater in patients taking conventional antipsychotic medications when compared with those taking clozapine.¹⁴⁰ EPA did not significantly ameliorate negative (PANSS) symptoms in any study, and improvements were rarely seen for positive (PANSS) symptoms.⁵⁸ General psychopathology (PANSS) scores were seldom improved significantly (i.e., by E-EPA¹⁴⁰). In Emsley et al.'s trial, tardive dyskinesia was ameliorated using 3g/d E-EPA.¹⁴⁰ The only study employing DHA as an intervention showed nonsignificant benefits when compared with placebo or EPA.⁵⁸

Results of our meta-analysis of PANSS total data revealed that dose influenced outcome. A significant placebo-controlled effect was identified for 2g/d EPA yet not for doses of at least 3g/d EPA. However, the significant result demonstrated somewhat limited applicability to the North American population although the inclusion of two UK studies meant that the potentially confounding influence of background diet was controlled for. While these findings are suggestive, they are not definitive given that the results subjected to meta-analysis were derived from a small number of trials involving a small number of patients with schizophrenia. Moreover, the effect might have been more pronounced had the data entered into meta-analysis come exclusively from patients taking clozapine as primary treatment. Peet et al. did not distinguish their results by type of primary treatment,⁵⁸ and we did not enter data exclusively from patients receiving clozapine in Peet et al.'s second trial.⁸⁷

Peet et al.'s patients who received clozapine were typically switched to this medication because existing pharmacotherapies had failed.⁸⁷ This suggests that these patients, for whom a placebo effect was far less likely than for patients receiving other antipsychotic medication, were more impaired than those patients receiving the other pharmacotherapies. Moreover, patients taking clozapine were at best partial responders to this agent given that they still exhibited PANSS total scores of at least 50 at the start of the RCT. Thus, patients on clozapine likely exhibited more “room for improvement” than did patients receiving the other drugs. That said, the positive response to E-EPA in this exploratory trial is quite interesting, and suggests the need to replicate this finding in an adequately-powered trial, which at minimum would need to enrol patients stratified by type of antipsychotic medication. Yet, the Emsley et al. study found a nonsignificant trend towards greater reduction in total PANSS scores in participants taking typical antipsychotic medication, compared with those receiving clozapine.¹⁴⁰

The overall outcome—a significant impact of low-dose EPA and a nonsignificant effect of high-dose EPA—may have been different if both of Peet et al.'s studies had used E-EPA. Compared with unpurified EPA,⁵⁸ E-EPA's processing minimizes its odour and flavor,⁸⁷ and this, in turn, should better preserve blinding. Another possible influence on the results relates to Peet et al.'s use of “uncontrolled dosing,” which involved pourable oils.⁵⁸ That is, the exposure was not delivered via capsules containing controllable amounts of exposure, but rather via prescribed amounts of oil poured from bottles onto or into foods on a daily basis. Uncontrolled dosing might have produced variability both in the daily and the full study intake of omega-3 fatty acids in the active treatment group and/or in the daily and the full study intake of corn oil in the placebo group. This could lead to confounding stemming from changes in the planned, constant between-group difference in omega-3 fatty acid intake and in the planned, constant between-group equivalence for energy/caloric intake. Controlled dosing likely would have substantially improved the experimental control in Peet et al.'s RCT.⁵⁸ Other potential influences on study results were the short intervention periods, small sample sizes and the use of different placebo sources (i.e., liquid paraffin⁸⁷ vs corn oil⁵⁸). More evidence is required to replicate these findings.

Having more studies to systematically review might eventually facilitate comprehensive assessments of the possible role of key covariates or confounders (e.g., current smoker status). Biomarker data were not meta-analyzed given the exploratory purpose underlying the inclusion of these observations from treatment studies. Baseline RBC EPA level predicted clinical improvement in response to EPA supplementation, for example.⁵⁸

A completed Cochrane review of PUFA supplementation for schizophrenia did not conduct meta-analysis in the way that we undertook ours, despite the fact that they identified the same placebo-controlled trials investigating the impact of omega-3 fatty acids.⁶¹ They did not evaluate the impact of dose on total PANSS scores in the same fashion; and, they combined data obtained from patients in placebo-controlled RCTs investigating omega-3 fatty acid supplementation as either primary or supplemental treatment. In our view, their approach compromised the meaningful interpretation of their observed effect in favor of omega-3 fatty acids even though this finding parallels what we observed exclusively with respect to a 2g/d dose of omega-3 fatty acids. No other completed systematic reviews investigating the benefits of omega-3 fatty acids in mental health were identified by our review.

As an aside, uncontrolled studies of the effect of omega-3 fatty acid supplementation have shown that 10 g/d of concentrated fish oil (MaxEPA®), including 1.7 g/d EPA and 1.1 g/d DHA, over 6 weeks improved schizophrenic symptoms and tardive dyskinesia in schizophrenic patients (n=20) taking their regular antipyschotic medication.⁹¹ However, Rudin et al. failed to identify a clinical benefit when linseed oil was given as a source of ALA (50% ALA) to a handful of patients with schizophrenia (n=5).¹⁸⁴

Research designs, which because of their prospective and controlled nature, are most appropriate for addressing the question of the possible intake of omega-3 fatty acids and the onset of schizophrenia were not found. Thus, there is little that can be said with confidence with regards to this subject. The only prospective study was not controlled, and its followup was very short.⁹¹ This, along with the observation that the diagnosis of schizophrenia had already been assigned in this study, indicates that its attempt to correlate dietary intake data with schizophrenia symptom scores could not be used to illumine the question of etiology. As well, data indicating a significant inverse association of EPA intake over 1 week with total psychopathology, or a similar, inverse relationship involving both ALA and total omega-3 fatty acid intake with positive symptom scores, do not allow us to respond meaningfully to the question of the exposure's possible impact on the disorder's continuation.⁹¹

The results from five case-control studies do not permit us to conclude that there is a reliable association between omega-3 fatty acid intake and the onset, course or outcome of schizophrenia. While Peet et al. noted that schizophrenic patients were significantly less likely to have been breastfed,⁹² findings from three other studies did not support this observation;^142–144 and, Amore et al.'s only statistically significant association indicated that the longer infants were breastfed, the later was the onset of schizophrenia.¹⁴¹

Differences in national or regional feeding patterns might account for differences among studies. At the same time, Sasaki et al. did not adjust or match for key confounders such as sex, maternal age or socioeconomic status.¹⁴⁴ McCreadie¹⁴³did not have access to stratified sampling data whereas Leask et al. did,¹⁴² perhaps leading to differences in observed patterns of breastfeeding. Moreover, less bias may have been associated with Leask's study¹⁴² since their cases and controls came from the same population at risk, with equal baseline risk of inclusion. Another factor potentially distinguishing the Leask et al. and McCreadie et al. studies is that the former's outcomes¹⁴² were incident cases while the latter obtained prevalence data,¹⁴³ which can be biased towards chronic illness. Leask et al.'s finding was likely more reliable for these reasons, although their analyses may have lacked statistical power. In any case, the studies suggested the absence of a significant association.

As well, likely only one case-control study adequately ruled out the possible impact of recall bias. Leask et al. analyzed breastfeeding data from mothers when their children were either two or seven years of age.¹⁴² Mothers in the other studies had to recall events 20–50 years in the past. Some studies have shown that while long-term recall of whether an infant was breastfed is good, the duration of breastfeeding or the timing when other milk products were initiated are recalled less well.^185–187

While cross-ecological analyses do not highlight data indicating individual/patient-level covariations of exposure and outcome, they nevertheless failed to demonstrate either a significant association of seafood consumption and lifetime prevalence rates of schizophrenia⁹⁰ or a significant relationship between fish consumption,¹⁰⁹ or UFA intake,¹⁴⁵ and the course or outcome of schizophrenia. That said, none of these studies attempted to rule out the possibility that (the nature of) early mother-infant contact might just as easily explain any possible association between breastfeeding and schizophrenia.

While medication status may have had somewhat of an influence on between-group differences in RBC or plasma phospholipid fatty acid content when the comparison group was healthy controls, because these data were obtained from cross-sectional studies, no meaningful possibility exists to permit drawing causal inferences regarding patterns of PUFA content and the onset of schizophrenia. The same criticism applies to the single study examining biomarkers data with respect to autism.

Clinical Implications

Omega-3 fatty acids in the present review's collection of interventional studies were not associated with moderate or severe adverse events. Supplementation was well-tolerated, with some mild, mostly gastrointestinal events occurring occasionally. Even the highest doses of omega-3 fatty acids did not produce significant side effects requiring patients to withdraw. The lack of variety in the types of omega-3 fatty acid employed in these studies means that this safety profile refers almost exclusively to the intake of either purified (i.e., E-EPA) or unpurified EPA.

The picture pertaining to the remaining evidence is essentially just as unequivocal. For each psychiatric disorder or condition whose evidence we evaluated, it is impossible to definitively conclude anything with respect to omega-3 fatty acids' efficacy as a therapy or prevention. The existing evidence is therefore insufficient to support clinical recommendations regarding the use of omega-3 fatty acids for the treatment or prevention of any specific mental health condition. Although some individual studies have reported some favorable results, trials have tended to be small, results have often been inconsistent, and study quality has been limited. It is likewise impossible to take existing biomarkers data as constituting reliable predictors of the onset, continuation or recurrence of any psychiatric disorder or condition. Too few large, well-controlled prospective studies employing research designs with the greatest inherent potential to address each of the first three basic research questions were identified in this systematic review. Yet, given their reasonable safety profile, it is likely that the use of (foods containing) omega-3 fatty acids to influence mental health is unlikely to produce notable adverse effects.

While little is known about the primary treatment of schizophrenia, more can be said about its supplemental treatment. A low dose of 2g/d EPA may, in the shortterm, ameliorate symptoms of schizophrenia. High dose (at least 3g/d) EPA did not provide a similar benefit. However, these observations require replication from much larger, longer term studies that also exercise specific experimental and statistical controls. These refinements are described in the next section. Until these studies are conducted we will not feel confident that 2g/d EPA, or any other dose or type of omega-3 fatty acid, can or cannot produce even reliable shortterm symptom improvement in schizophrenia. Similarly, until more appropriate research designs are employed, the existing evidence does not allow us to conclude that either specific patterns of omega-3 fatty acid intake or particular PUFA levels in biomarkers reliably predict the onset, continuation or recurrence of schizophrenia. It is therefore doubtful that the latter observations can be used to unequivocally confirm the PUFA deficiency facet of the membrane phospholipid hypothesis concerning the etiology of schizophrenia.⁵⁵

It has also been suggested that smoker status alone may account for the results indicating between-group differences in PUFA content,⁶⁰ although other factors can influence PUFA status as well (e.g., medication use, alcohol consumption: see above). Hibbeln et al.'s additional analysis⁶⁰ of RCT data from their investigation of the effect of omega-3 fatty acids as supplemental treatment for schizophrenia⁸⁹ revealed some important observations that may raise doubts about the validity of some of the data presumed to support the membrane phospholipid hypothesis regarding the etiology of schizophrenia.⁵⁵ They have suggested that failing to account for current smoker status in studies examining the possible depletion of PUFA content may have confounded numerous, if not most, of the results of cross-sectional studies thought to support the hypothesis.⁶⁰ Their observations are summarized below.

Peet et al.'s demonstration of the superiority of EPA over DHA as supplemental treatment for schizophrenia was not expected.^{58, 188, 189} The investigators had assumed that DHA's prominent role in neuronal membrane phospholipids, via their capacity to affect the configuration and function of neurotransmitters (e.g., dopamine), might contribute to the amelioration of symptoms. (Others have suggested that DHA has mood stabilizing effects because of its action on serotenergic neurotransmission, altered membrane fluidity and suppressed phosphatidylinositol and protein kinase C signal transduction.^{113, 188–190}) Peet et al. also expected that EPA's lesser representation in neuronal membranes would mean that it would play a less important role. They argued that any positive clinical effect of EPA would have to occur independent of its direct incorporation into neuronal membrane phospholipids.⁵⁸

That said, they found that schizophrenic patients with the lowest RBC EPA levels exhibited the weakest response to treatment, an observation, they argued, that would not be predicted if EPA treatment merely entailed correcting a membrane deficiency.⁵⁸ Given their earlier findings of a bimodal distribution of PUFA levels in schizophrenic patients (i.e., very low vs moderately low EPA reductions when compared with healthy controls¹⁴⁹), the group with very low EPA levels in their treatment study may have had a more serious metabolic problem that was less amenable to modification by EPA supplementation. On the other hand, further analyses of biomarker data from Fenton et al.'s supplemental treatment RCT⁸⁹ found no evidence of baseline bimodal distributions of RBC EPA, DHA or AA compositions in schizophrenic patients.⁶⁰

Mechanisms potentially leading to EPA being more effective than DHA in depression have been reviewed briefly by Peet et al.⁵³ In depression, the production of PGs from AA by the cyclooxygenase system appears to be elevated; and, EPA but not DHA has been observed to be an effective substrate for cyclooxygenase, and can compete with AA at this point in the metabolic pathway. Also, in some phospholipase A₂ assays, EPA but not DHA has been seen to be an effective inhibitor. Work investigating EPA's possible mode of action has also suggested the possible role of increased phospholipase A₂ enzyme in the etiology of schizophrenia.⁵⁵ Although the modulation of background drug pharmacokinetics cannot be ruled out as the mechanism of action of E-EPA, Peet and Horrobin have suggested that it is more likely that its action is on cell membranes and signal transduction systems.^{53, 190} These different effects of EPA and DHA, which may be characterized both by synergism and antagonism, suggest that the biological effects of fish oils, which contain both EPA and DHA in highly variable proportions, may be difficult to predict.⁵³

Overall, we agree with Peet et al. that the biomarkers (Question 3) and intake-outcome association data (Question 2) are likely suggestive enough to justify the conduct of more intervention studies pertaining to schizophrenia and depression.⁵⁸ We provide a few details in the next section.

Until data are obtained from more appropriate research designs (i.e., well-controlled prospective designs collecting individual/patient-level data), it is likely impossible to conclude with great confidence that an omega-3 fatty acid deficiency is responsible for the onset of depression (Question 3); or, that these findings, together with data presumed to reflect the “protective potential” of omega-3 fatty acid intake (Question 2), can readily be taken to justify the use of omega-3 fatty acids as either prevention or therapy. Testing the omega-3 fatty acid deficiency hypothesis also requires control of variables with the potential to influence PUFA status. Suggestive results from the small number of typically underpowered RCTs likely cannot be used to confirm or disconfirm the value of using omega-3 fatty acids as a primary or supplemental therapeutic for depressive disorders or symptomatology (Question 1). More evidence is required.

Even less, or nothing, can be concluded about the value of omega-3 fatty acids as (primary or supplemental) treatment or (primary or secondary) prevention for bipolar disorder, suicidal ideation or behavior, symptoms of anxiety, obsessive-compulsive disorder, anorexia nervosa or other eating disorders, AD/HD, tendencies or behavior with the potential to harm others, alcoholism, borderline personality disorder, autism and mental health difficulties in general. The same may be said about the value of PUFA biomarker profiles as reliable predictors of the onset, continuation or recurrence of these disorders. Even if apparently consistent findings were noted, for example when a greater intake of (foods containing) omega-3 fatty acids was associated with lower prevalence rates of both depression and bipolar disorder, these observations came from designs exhibiting the weakest ability to illumine individual/patient-level associations (i.e., cross-national ecological analyses). Furthermore, much of the included research evidence lacked strong applicability to North Americans. Recommendations for further research stem from the identication of limitations characterizing existing studies and are highlighted in the next section.

Research Implications and Directions

One overarching finding revealed by our review is that not all psychiatric disorders have been investigated for their clinical response to primary or supplemental treatment with omega-3 fatty acids (Question 1) or for their possible association (e.g., prevention) with either omega-3 fatty acid intake (Question 2) or the omega-3 or omega-6/omega-3 fatty acid content of biomarkers (Question 3). Some studies have also focused exclusively on psychiatric conditions (e.g., symptoms of anxiety) that are necessary yet insufficient to merit a formal clinical diagnosis.

The primary targets of mostly recent research endeavors have been schizophrenia and depression. Studies examining the association between these psychiatric disorders or conditions and the PUFA content in biomarkers (Question 3) have outnumbered those investigations evaluating their association with the intake of omega-3 fatty acids (Question 2), and have far outnumbered studies assessing treatment of these disorders or conditions with omega-3 fatty acids (Question 1). In intervention studies, the emphasis has been almost exclusively on the supplemental treatment of these disorders or conditions.

The lack of studies pertaining to some psychiatric disorders or conditions (e.g., eating disorders other than anorexia nervosa) means that nothing can be concluded other than the need for multiple research investigations, employing appropriately-controlled designs of sufficient size (i.e., to afford detection of a meaningful effect/association) and incorporating sound methodologies (e.g., reliable and valid outcome measurements). For those psychiatric disorders or conditions for which fewer than all of the first three basic questions were found to have been examined with empirical evidence (e.g., borderline personality disorder), the unstudied questions likewise require research embodying multiple, appropriately-controlled designs of sufficient size, and implementing sound methodologies.

Yet, even for those questions investigated by numerous studies (i.e., associations between the PUFA content of biomarkers or the intake of omega-3 fatty acids and schizophrenia or depression), limited sample sizes, designs (e.g., cross-sectional studies examining associations between biomarkers and the onset of schizophrenia or depression) and methodologies (e.g., using apparent seafood consumption to measure intake of sources containing omega-3 fatty acids in cross-national ecological analyses) highlight the need for studies incorporating modifications to each of these study parameters. Finally, for those few topic areas where studies implementing appropriately-controlled research designs and sound methodologies yielded somewhat suggestive (i.e., supplemental treatment of schizophrenia) or potentially promising results (i.e., supplemental treatment of depression), more, similarly well-designed studies need to be completed, which enroll/allocate larger sample populations and implement additional or refined research design or methodologic characteristics (e.g., account more extensively for covariates and confounders). Given that only minor safety issues were noted in the included studies—despite their likely under-reporting—treatment and prevention trials are justifiable. We now highlight some of the possible directions these investigations might take. While we focus considerable attention on schizophrenia and depression, given their prominence in our review, many of the basic issues apply equally to other psychiatric disorders or conditions.

One possible approach to developing future research avenues is to encourage the collection of data (e.g., animal or human) and the construction of models (e.g., mechanisms of action) suggesting the (e.g., biological) plausibility of clinical treatment effects associated with omega-3 fatty acid supplementation before needlessly embarking upon the expense of studies examining the utility of these treatments for mental health problems (Question 1). From this vantage point, it could be argued that, for those psychiatric disorders or conditions for which few or no empirical treatment data have yet been obtained, the next step would be to establish some degree of plausibility regarding treatment based on empirical evidence addressing other, purportedly more “basic” research questions.

Two such questions, albeit exclusively focused on human data, were addressed in our review: the association between the the onset, continuation or recurrence of psychiatric disorders or conditions and the intake of omega-3 fatty acids (Question 2), or their association with the omega-3 or omega-6/omega-3 fatty acid content of biomarkers (Question 3). Empirical answers to these two questions could suggest important roles (e.g., risk factors) for omega-3 fatty acid contents in mental health, observed in terms of their dietary intake and/or their levels (e.g., composition, concentration) present within blood lipid biomarkers. In turn, these data could serve to justify the use of omega-3 fatty acids as a treatment.

In the present collection of studies, answers to Questions 2 and 3 were too limited, either design-wise or methodologically, to provide support for the deficiency hypotheses relating to either depression or schizophrenia. For the reasons described earlier, both cross-sectional studies and cross-national ecological analyses cannot produce answers that meaningfully identify those omega-3 fatty acid intake or biomarker profiles that may influence the onset of depression or schizophrenia. As well, studies employing other designs did not provide unequivocal support for these hypotheses. Thus, investigators in the domain of inquiry of interest to the present systematic review have at least two options: either wait for these “basic” data to be collected before conducting treatment studies, or find some other rationale supporting the design of new treatment trials. One such raison d'être might be to improve upon the designs and methodologies of studies that have already been completed. We suggest that this option may be especially relevant with regards to the foci of schizophrenia and depression. Moreover, it has also been said that data from epidemiological studies observing the association between fish consumption and psychiatric disorders (e.g., depression), or data from cross-sectional studies observing the association between the PUFA content of biomarkers and clinical outcomes (e.g., presence/absence of a diagnosis of major depressive disorder), each constitute, at best, indirect lines of evidence supporting a role for omega-3 fatty acids in the etiology, pathogenesis and treatment of such disorders.¹⁹¹

What, then, are the directions that these treatment studies might take? After we focus on this question, we highlight briefly some avenues that studies addressing the other two basic questions (i.e., intake, biomarkers) might follow. Many of the issues we raise are pertinent to all questions, given that they reflect the need to exercise tighter control of variables (e.g., population) with the potential to confound results. At present, despite the suggestive picture of efficacy of a 2g/d dose as supplemental treatment for schizophrenia, nothing conclusive can be said about the nature of the impact of any of the confounders (e.g., smoking; alcohol use; omega-3 fatty acid dose) on clinical or biomarker outcomes with respect to any of the questions or disorders/conditions whose studies we systematically reviewed.

With respect to the question of the efficacy of omega-3 fatty acids as primary or supplemental treatment for any psychiatric disorder or condition, there have been too few well-controlled RCTs of sufficient size and intervention length to permit drawing any meaningful conclusions about shortterm symptom relief. Moreover, only one study each examined the primary treatment of depression or schizophrenia. More, larger and adequately powered studies are required.⁵³ As well, three months of omega-3 fatty acid supplementation in the life of a patient with schizophrenia or with treatment-resistant depression likely cannot be considered, in clinical terms, to be a shortterm intervention. Intervention lengths in the studies investigating the supplemental treatment of depression or schizophrenia lasted no longer than 13 weeks; often, patients seen in clinical practice with either of these disorders will receive medication for years or even decades.

Therefore, especially because of the somewhat suggestive evidence pertaining to 2g/d EPA as supplemental treatment for schizophrenia, additional studies need to replicate this shortterm finding over longer investigative periods. These studies should resolve whether or not omega-3 fatty acid supplementation, when provided as a supplemental intervention, can provide (e.g., additional) shortterm and longer term symptom relief. Then, if ever reliable longterm symptom relief is demonstrated, future studies could be conducted to see whether omega-3 fatty acid supplementation can alter the progression of psychiatric disorders or conditions. The effects of certain conditions associated with schizophrenia and/or the medications given to schizophrenic patients might be lessened, or even prevented. One focus could be tardive dyskinesia, whose incidence, severity or progression could be studied as potentially modifiable outcomes. Likewise, for some patients diagnosed with major depression, suicidal ideation or behavior may become less likely or less intense. Dysphoric feelings might also be “prevented” from becoming full-fledged disorders. In addition, patients with “rapid cycling” forms of bipolar disorder could experience a lengthening of the time between major shifts in mood.

To return to the topic of schizophrenia, if ever its symptoms or clinical course are demonstrated to be improved by omega-3 fatty acid supplementation, especially in the longterm, then “medication-sparing” research designs could be used to test whether adding a specific dose of omega-3 fatty acids—which typically exhibits a relatively benign safety profile—to a lower-than-usual dose of antipsychotic medication can maintain at least the same level of clinical improvement (e.g., symptom control) typically associated with a traditional dose of this antipsychotic medication. This is potentially clinically significant since, at full-dose, antipsychotic medications often exhibit a notable safety profile (i.e., moderate-to-severe adverse effects). In this way, patients might be “spared” the likelihood, or a particular intensity, of side effects associated with their regular antipsychotic medication (e.g., extrapyramidal symptoms).¹⁴⁰ This type of study design is often employed in studies examining health problems (e.g., asthma) where it may be best to reduce doses of medication (e.g., oral corticosteroids), which especially in the longterm, can have negative health consequences.

In such a dose-sparing study, the exact amount of the (absolute or percent) dose reduction could be defined either before the RCT begins or established during the course of the study. In either design, patients selected would include those schizophrenic individuals whose symptoms are well-controlled by their regular antipsychotic medication. If these participants vary on the basis of the type of prestudy medication, stratification by medication type could be undertaken. The types and severities of prestudy adverse effect related to their antipsychotic medication would be noted prior to the commencement of the trial.

In the type of RCT where the dose reduction is defined before the study begins, patients would be randomized to one of two conditions: a) an (absolute or percent) reduction in the dose of their regular antipsychotic medication in addition to receiving a daily dose of omega-3 fatty acid supplementation; or b) continuing to receive their regular antipsychotic medication in addition to a placebo to control for the other group's receipt of supplementation. While other design or analytic controls would be required (e.g., adding placebo material to the first group's antipsychotic medication in order to mask the dose reduction; determining whether the degree of symptom control at study baseline for both groups was indeed the same), the outcomes of interest would be whether or not: a) the same degree of symptom control (e.g., PANSS total) was observed in both groups; and b) adverse effects related to the antipsychotic medication occurred less often, or were less severe, in the dose reduction group.

In the second type of RCT, dose reduction in one of the study groups would be conducted on a patient-by-patient basis, and in a predefined and uniformly stepwise fashion, until evidence for a loss of symptom control would signal a halt to the reductions. While implementing all of the aforementioned controls, the outcomes of interest would be: a) the mean (or maximal) percent dose reduction that permits symptom control; and, b) the pattern (i.e., frequency or severity) of adverse effects.

These studies might identify, in empirical fashion, the specific types(s) of patient or medication for which dose reductions are beneficial on the basis of both outcomes. The first type of RCT would typically precede conduct of the second type. A final followup would need to take place no earlier than at six months, to be able to establish the stability of any benefits.

For each disorder or condition, including schizophrenia and depression, much more work needs to be done to identify the exact sources (e.g., marine), types and doses of omega-3 fatty acids, and combinations thereof, which reliably produce clinical effects. Both DHA and ALA were underrepresented in the present evidence base. Whether or not specific doses of EPA and DHA should be combined, and how, and for which disorders or conditions, remains to be determined. Whether or not different types and doses of omega-3 fatty acid are required to treat disorders (e.g., major depressive disorder), compared with psychiatric conditions (e.g., feelings of dysphoria), is unknown. Likewise, whether or not different types and doses of omega-3 fatty acid are required to treat disorders of varying degress of severity, or associated with various types (and severities) of comorbid condition, are unresolved questions. Additionally, with respect to each of these questions, there remains the issue of which combinations of omega-3 fatty acid types and doses are both efficacious and safe, that is, where they minimize the likelihood of even mild, transient adverse events.

There are also a number of questions that need to be addressed further regarding dosage. For example, is one RCT enough to determine that 1 g/d E-EPA yields significant clinical improvement in populations experiencing depressive symptomatology?⁵³ Would such a low dose produce the same kind of effect in those formally diagnosed with a depressive disorder? It is our view that we need more research evidence before we can conclude anything about the utility of this dose for any psychiatric disorder or condition, not just for those individuals exhibiting depressive symptomatology.

At the same time, is 9.6 g/d EPA+DHA too high a dose for patients with bipolar disorder¹¹² or any other disorder? Stoll et al. did not report even moderate adverse events associated with this dose.¹¹² Again, we likely need more than a single study (which was stopped prematurely) before we can conclude anything about this dose's clinical utility. Additional research might reveal that the definition of an “effective dose” is disorder-specific, that doses should be weight-adjusted especially in studies with children, or that doses should be adjusted to fit individuals, and not vice versa.^{53, 190}

While dose-ranging studies may be helpful in determining answers to some of these unresolved issues, it may be wise, however, to avoid situations such as those encountered in the two supplemental treatment RCTs conducted by Peet and colleagues. By having four levels define their intervention (i.e., 4g/d vs 2g/d vs 1g/d vs placebo) in studies examining depression⁵³ and schizophrenia,⁸⁷ these investigators made it difficult to power each study sufficiently to afford detection of significant clinical effects. It may be better to design less complicated studies with respect to levels of the intervention, while instead instituting greater experimental control of variables with the potential to confound clinical outcomes. More about this topic is discussed below.

That said, in designing future trials, we will likely need to select doses which allow us to make sense of why only patients receiving a low dose (1g/d) E-EPA in Peet et al.'s investigation of the supplemental treatment for depression benefited clinically.⁵³ The investigators themselves offered no cogent explanation, yet we suggest that the effect might eventually be found to have been produced by hormesis, or to have been influenced by certain changes—which were unrelated to the intervention—in the clinical status or background diet (e.g., omega-6/omega-3 fatty acid intake ratio) of patients receiving the 1g/d dose of E-EPA.

If the goal is to be able to readily interpret study results aimed at determining the clinical utility of omega-3 fatty acids as an intervention, researchers likely need to satisfy a number of requirements. Studies should likely avoid using uncontrolled dosing methods (e.g., oils poured from bottles), since this approach makes it difficult for studies employing controlled research designs to achieve two key controls typically preferred in supplementation studies; each control is intended to minimize the influence of confounding. For example, requests to pour specific amounts (or ranges) of oil from bottles on a per-meal or a per-day basis, can make it difficult to assure that study subjects consistently pour the prespecified amounts of oil and thereby maintain the planned on-study between-group difference in the intake of omega-3 fatty acids (e.g., 3 g/d EPA vs 0 g/d EPA) as well as the planned on-study between-group equivalence of energy/caloric intake (e.g., 3g/d of oil for each study group).⁵⁸ Failure to maintain these two between-group constants would confound study results.⁷² Then, at the end of the study, when clinical results following uncontrolled dosing require interpretation, it may be impossible to specify, with much precision or confidence, the “daily dose” to which a significant or nonsignificant between-group difference might be attributed.

Uncontrolled interventional studies can also be plagued by this consequence of uncontrolled dosing. It is likely easier to control “doses” when the exposure is delivered via prespecified numbers of swallowable capsules containing finite amounts of omega-3 fatty acid content. Compliance data may someday help to evaluate the present hypothesis regarding the benefits of controlled dosing. For now, the nature of the impact of Peet et al.'s uncontrolled dosing scheme⁵⁸ on the significant clinical effect identified by our meta-analysis, assessing the value of low-dose EPA as a supplemental treatment for schizophrenia, remains unknown.

If the goal is to be able to readily interpret study results aimed at determining the clinical utility of omega-3 fatty acids as an intervention, it is also likely wise to avoid using complex interventions, or “cocktails,” which contain omega-3 fatty acids combined with many other active ingredients. Otherwise, it will be impossible to account for the exact contribution of omega-3 fatty acids to any clinical effects. The issue of complex exposures is discussed further below.

One type of ingredient in omega-3 fatty acid exposures that is likely useful is one which can maintain the freshness of the exposure and thereby prevent the type of rancidity that would allow patients to determine, from the increasingly strong taste or odour of especially fish oils especially, which exposure they are receiving.^{190, 191} Failure to maintain freshness can jeopardize blinding. Several interventional studies added, for example, the antioxidants tertiary butylhydroquinone and tocopherals, to what all study groups received as their exposure so as to maintain the exposure's freshness, but also to avoid any possible confounding were these ingredients ever found to have a psychoactive effect.^{96, 192} Other studies added flavoring to what all study groups received, or even vacuum-deodorized the exposures in order to maintain blinding.^{96, 191} Still others employed purified EPA, or E-EPA, which purportedly eliminates much of an oil's original taste and odour.^{53, 87, 115, 139}

Future research probably needs to carefully examine the impact of using E-EPA, compared with EPA, both to maintain blinding and to influence study outcomes. Processing EPA may change its therapeutic (or preventive) potential, but research assessing its role in mental health is needed to ascertain this possibility. At the same time, the actual purity of all exposures should likely be established.¹⁹¹ Otherwise, unknown elements already contained within fish oil, for example, might somehow mitigate the effects of the oil itself. Only studies that employed E-EPA even broached this subject, meaning that the exposures in the other interventional studies could have included agents that potentially affected their therapeutic (or preventive) value. No study report addressing any of the basic questions described having assessed the possible presence of, or having eliminated, methylmercury from marine sources of omega-3 fatty acids.

The choice of placebo may also affect study results. While the need for a standard placebo format in research on the therapeutic or preventive role of omega-3 fatty acids in mental health has yet to be established, there is preliminary evidence suggesting that investigators might want to consider steering away from the use of olive oil as a placebo in interventional studies of mental health.^{119, 120} Olive oil is a source of oleic acid, from which the psychoactive lipid oleamide can be biosynthesized in mammals;¹⁹³ and, oleamide has psychoactive properties, including the induction of sleep and the modulation of serotonin receptor-mediated signaling.¹⁹⁴ Thus, olive oil may actually affect mood disorders, which might diminish between-group differences in certain clinical outcomes.¹⁹⁵ Stevens et al. has recommended that liquid paraffin oil be used as placebo in supplementation studies.¹²³ Liquid paraffin oil was the choice of various intervention studies reviewed in our report.^{53, 87, 115, 140} Whatever the true influence of placebo contents on clinical outcomes turns out to be, it has been recommended that one way to minimize the placebo response seen on a few occasions in our review may be to include a 2-week placebo run-in period in trials.⁸⁸

To revisit the subject of complex interventions, given the primarily competitive inter-relationships between omega-3 and omega-6 fatty acids, and their respective metabolites—both within the metabolic pathway and within membranes—future research could potentially end up identifying that specific types and quantities of both omega-3 and omega-6 fatty acids require simultaneous modification to reliably produce clinical benefits for some or all psychiatric disorders or conditions. This research might also point out that these significant clinical effects are brought about by changes in levels of specific types of PUFA in specific biomarker sources (e.g., RBCs). That is, significant clinical benefits could result from the subtraction, from the background diet, of specific types and amounts of omega-6 fatty acids, concomitant with the addition of omega-3 fatty acid content. This strategy could essentially lower the omega-6/omega-3 fatty acid intake ratio. It might also decrease the omega-6/omega-3 fatty acid content ratio in certain biomarkers, although, as pointed out earlier, PUFA status has multiple determinants. However, no study identified by our review employed this dual approach.

Given that a high omega-6/omega-3 fatty acid intake ratio has been thought to be associated with patterns of disease,^33–45 the possible success of a strategy to reduce the omega-6/omega-3 fatty acid ratio might not be unexpected. Variables such as the magnitude of the change in each PUFA's content, the intake ratio's actual value, the intervention length, or the timing of these changes in (or prior to) a disease process might determine the success of such a therapeutic (or preventive) strategy.

There is some suggestion that the omega-6/omega-3 intake ratio in the background diet may predict the likelihood of observing significant clinical effects. It comes not from the present review, since there were too few studies per psychiatric disorder or condition with which to assess the impact of this possible confounder. Rather, we shared these observations in a recent report examining the impact of omega-3 fatty acids in asthma.⁷² While we did not feel it was appropriate to perform a meta-analysis of these results concerning asthma, an impressionistic analysis suggested that studies examining the effects of omega-3 fatty acid supplementation conducted within Asian countries—where the omega-6/omega-3 fatty acid intake ratio in the background diet is considerably reduced compared with the omega-6/omega-3 fatty acid intake ratio in the background diet of populations selected from non-Asian countries—were more likely to produce significant clinical improvements in respiratory outcomes.⁷² With less competition for enzymes in the metabolic pathway, and for positions in cell membranes, it is conceivable that in populations eating considerable amounts of fish or seafood, their lower levels of omega-6 fatty acid intake and higher levels of omega-3 fatty acid intake in the prestudy and on-study background diets may make it “easier” for additional omega-3 fatty acid supplementation to make a clinical difference. This speculation may pertain especially (or exclusively) to DHA, given its likely function(s) in cell membranes. Yet, an alternative hypothesis could suggest that significant clinical benefits are less likely when the omega-6/omega-3 fatty acid intake ratio is already reduced prior to a study because cell membranes already contain “enough” omega-3 fatty acid content, and adding typically small amounts of omega-3 fatty acid content via supplementation may not make an appreciable difference.

Whichever hypothesis is confirmed by future research, both perspectives rest on the assumption that clinical effects are brought about by changes in the PUFA levels observed within blood lipid biomarkers. More research could indicate that this is not the case. While PUFA status is influenced by more than just the intake of omega-3 fatty acids, the mechanism promoting clinical changes could actually be even more complicated, implicating the availability of enzymes to, for example, desaturate or elongate PUFA metabolites, or entailing the production or activities of eicosanoids or cytokines. The LC PUFAs especially may be found to directly influence synaptic function through effects on membrane structure and/or indirectly through the production of eicosanoids (PGs, LTs, TXs) or via immune system/cytokine interactions.^{122, 190} It is therefore likely appropriate to continue examining PUFA content levels in biomarkers within studies evaluating the impact of omega-3 fatty acids on mental health.

In the present review, we could not investigate either directly or indirectly (e.g., using the country in which a study was conducted as a surrogate measure of the omega-6/omega-3 fatty acid intake ratio) the impact on clinical outcomes of the omega-6/omega-3 fatty acid intake ratio of: a) the prestudy/baseline background diet, b) the on-study background diet (i.e., excluding the supplementation), or c) the complete on-study diet (i.e., background diet plus supplementation). Moreover, few investigators conducting interventional studies controlled for this possible confounder either by mandating that patients maintain their prestudy/baseline background diet during the study or by performing a covariate analysis.

Researchers in future interventional studies (i.e., treatment, prevention) will likely need to account analytically (e.g., covariate analysis), if not experimentally (e.g., subject selection criteria; stratification), for prestudy and on-study background definitions of diet, and their inherent omega-6/omega-3 intake ratios, if only because they may influence/predict clinical outcomes. At the same time, it may be premature to assert that given the likely inter-relationships between omega-3 and omega-6 fatty acid contents both in the diet and in blood lipid biomarkers, the questions examined in this review could benefit in the future from an expanded scope, that is, to include a co-focus on the influence of omega-6 fatty acids in mental health.

Future research also needs to assure full knowledge of the details defining study populations so that this source of clinical heterogeneity can be taken into account when analyzing and interpreting the results of interventional or observational studies. Many study reports included in our review failed to specify many of the details pertaining to population variables with the potential to confound study outcomes. These include the possible between-group differences observed at study baseline in controlled studies, which relate to the severity and historical course of the primary disorder (e.g., age of onset, number of episodes, timing of intervention relative to the disease process [e.g., first-episode vs chronic schizophrenia]) or the presence and nature (e.g., severity, age of onset) of comorbid conditions (see Chapter 2). Occasionally, full sample descriptions of these variables were not provided. Failing to have these details made it impossible for us to informally assess their possible impact on study results.

However, this was not always the case in individual studies. In their RCT examining the supplemental treatment of schizophrenia, Hibbeln et al. were concerned that the long duration of illness in their patient population, reflected in notable symptoms despite treatment with newer neuroleptics (e.g., clozapine), may have contributed to the failure to find a significant clinical effect for their full sample.^{60, 88} Patients in other supplemental treatment studies had been younger and exhibited a shorter illness duration. However, additional analyses revealed that duration of illness was not associated with changes in clinical outcomes or in changes in EPA, DHA, AA, or AA/EPA fatty acid contents following EPA supplementation.⁶⁰ Analytic “control” for this possible confounder was achieved, and afforded a clearer interpretation of study outcomes.

Other possible, population sources of confounding may be observed in circumstances where on-study life events unrelated to the exposure (e.g., job loss) can influence subjects' psychiatric status and, in turn, their response to treatment. But, these events need to be measured in order to to statistically control for them. Seldom did the present collection of interventional studies identify the occurrence, or noted absence, of important life events other than those few presumed to be the reason for a discontinuation.

Successful control for population sources of confounding can also be achieved through experimental means. For example, an interventional or an observational study might only include patients exhibiting a single diagnostic subtype, or a minimal or maximal severity level for a particular disorder. They might also exclude patients exhibiting certain types or severities of a particular comorbid condition. While such restrictive conditions limit the possible “breadth” of the population to which study results can be generalized, these experimental controls maximize the specificity of populations to which the evidence can be extrapolated. Ultimately, this could benefit the practice of mental health care.

One final population source of confounding was highlighted in a recent meta-analysis investigating the impact of short-acting Ritalin® in the treatment of AD/HD.¹³ The basic premise is that, while patients or populations may share a given diagnostic label (e.g., AD/HD) assigned using stringent clinical approaches, even sophisticated diagnostic classification approaches (e.g., DSM) can lead to an obfuscation of individual differences when it comes to understanding what each of these individuals is experiencing clinically and, in turn, selecting an appropriate treatment strategy. We focus here on the first observation.

To use AD/HD as an example, three major subtypes of AD/HD are identified by DSM-IV (i.e., predominantly Inattentive subtye vs predominantly Hyperactive subtype vs Combined subtype). Thus, the “AD/HD” label can refer to three different clinical scenarios. In the AD/HD studies included in our review, seldom were the exact subtypes specified or was this source of clinical heterogeneity controlled for analytically. To compound matters, the method employed to assign any one of these subtype diagnoses allows for important variability in the numbers, and combinations of symptom that can be taken to indicate the presence of a single diagnostic subtype. For example, for problems with inattention, DSM-IV asks clinicians to select 6 of 9 possible items, and the same request for 6 items is made with respect to 9 possibilities concerning problems with hyperactivity/impulsivity. Thus, there are several different combinations of symptom referred to by a single diagnostic label; this might be thought of as a homogeneous population in an interventional or observational study, when in fact, it is not. Individuals could vary widely on the basis of their clinical pictures of symptoms. Furthermore, this heterogeneity could influence responses to treatment even in RCTs, where different distributions of clinical picture could characterize different study groups. Such uncontrolled population variability is likely undesirable, and it is further complicated when and if controls are not put in place to deal with similar problems relating to variability in comorbid conditions.

Finally, as introduced in Chapter 1, one other population source of clinical heterogeneity—most important when different studies are compared, as is the case in systematic reviews—stems from relevant studies having used different diagnostic systems, or even different versions of a constantly evolving system (e.g., DSM-III published in 1980, DSM-III-R in 1987 and DSM-IV in 1994), to identify their study populations. Since the diagnostic criteria of these systems can vary, even slightly, then the study populations, or subpopulations, they identify can also vary.¹³ This additional definition of “diagnosis heterogeneity” could account for differences in outcomes observed in different studies. Systematic reviews relating to mental health should therefore consider evaluating the impact of diagnostic systems on study outcomes. However, in our review, having too few studies included per psychiatric disorder prevented us from achieving this task.

Other types of control are likewise required to maximize the interpretability of results of interventional or observational studies involving omega-3 fatty acids. Here, we distinguish between three types of variable based on their possible influence on outcomes. They include: those that have the potential to impact clinical (mental health) outcomes; those that can influence the fatty acid content of biomarkers (and which may turn out to be responsible for specific clinical effects); and, those that appear to affect both types of outcome.

There were too few studies per psychiatric disorder or condition to permit the identification of the nature or extent of the influences of effect modifiers on clinical outcomes relating to any of the basic research questions we investigated. Examples of influences on mental health observed in clinical practice include: illicit drug use, general health status, stressors, social support, exercise, quality of sleep, marital status, education, income and employment status. In both controlled and uncontrolled studies, these factors can independently, or in combination, influence mental health outcomes and thereby confound study results. These influences can be observed, for example, where their on-study status changes in ways unrelated to the intervention/exposure (e.g., an unexpected death in the family). In controlled studies, these variables (e.g., disease severity; comorbid conditions) can also affect study outcomes when study groups differ in their prestudy/baseline status. Either scenario has the potential to mask or artificially inflate the actual benefits of an omega-3 fatty acid intervention/exposure. As a result, future studies relating omega-3 fatty acids and mental health should consider controlling, either experimentally or analytically, for the possible impact of these variables.

Influences on PUFA status include: the disease process itself; dietary intake, metabolism and incorporation into cell membranes of various types and amounts of both omega-3 and omega-6 fatty acid content; efficiency of the PUFA metabolic processes, including the availability and effectiveness of enzymes implicated in the processes of desaturation and elongation; and the ability of protective mechanisms to deal with degradation from oxidation and other sources.^{48, 101–103, 178–180} While the impact of these variables could not be ascertained in our review, future studies which assume that beneficial effects on clinical outcomes might be mediated by changes in the PUFA status of biomarkers likely need to account for these factors.

Variables with the potential to influence both clinical (e.g., control of psychiatric symptomatology) and PUFA (e.g., omega-6/omega-3 fatty acid content in RBCs) status/outcomes include: age; sex; the disease process underlying any possible comorbid conditions; psychotropic medication, including type, dose and duration of use; (e.g., prestudy/baseline and on-study) background diet; alcohol consumption; and current smoker status.^{60, 180} But, other than those data indicating the positive impact of omega-3 fatty acid supplementation on symptoms of schizophrenia in patients taking clozapine,⁸⁷ little can be said about the actual roles of these variables within our evidence base. While the impact of medication status on the PUFA levels of biomarkers could be informally assessed with respect to schizophrenia, little that is meaningful can be concluded since the preponderance of cross-sectional designs prevents drawing inferences about etiology. As stated earlier, the impact of background diet, or of the country in which the study was conducted as a possible surrogate measure of the omega-6/omega-3 fatty acid content thereof, could not be evaluated.

Overall, there were too few studies per psychiatric disorder or condition to permit the identification of the nature or extent of the influences of these or other effect modifiers on outcomes relating to any of the basic research questions we investigated. It is therefore difficult to definitively rule out the possible impacts that these variables may have had on study outcomes. New research should consider routinely accounting for these factors. As was presented above, one interventional study did pursue this ideal regarding the variable of “illness duration.”

Hibbeln et al. also argued, based on additional analyses of interventional data,⁸⁹ that sex and current smoking status are important confounders in studies examining the interventional or observational relationships between omega-3 fatty acids and schizophrenia outcomes.⁶⁰ Each variable was significantly related to fatty acid compositions. DHA was reduced in smokers compared with nonsmokers, and males had lower DHA and EPA fractions compared with females. MANOVA revealed that, of all subgroups, nonsmoking women had the highest EPA and DHA levels while AA did not vary by smoker status or sex.⁶⁰ For females, nonsmokers exhibited a greater RBC EPA and DHA percentages compared with smokers. For males, no significant differences for EPA, DHA or AA were noted when smoker status was evaluated. Both EPA and DHA compositions were higher in female nonsmokers compared with male nonsmokers. Neither fatty acid compositions nor the number of cigarettes smoked per day differed significantly for male and female smokers. Thus, the sex specificity of the smoker status effect is likely not attributable to differences in smoking intensity (i.e., daily number of cigarettes smoked). Smoking intensity was not significantly associated with either absolute or relative amounts of EPA, DHA or AA in RBCs.

Hibbeln et al. then reported that, when they assessed the effects of sex and smoker status on the dietary intake of omega-3 (ALA, EPA, DHA) and omega-6 fatty acids (LA, AA), dietary intakes did not differ by sex or current smoker status when data were expressed as absolute amounts of daily intake.⁶⁰ When intake data were observed as percentages of total fat intake, nonsmokers consumed more ALA, although no other significant differences were observed. The consumption of EPA and DHA, expressed as percentages of total fat intake, were greater in females compared with male patients. Differences in AA were not found when either smoker status or sex were taken into consideration. Dietary EPA intake (as percentage of total fat intake) predicted the RBC EPA composition for all patients, for male patients alone, and for nonsmoking males.⁶⁰ Dietary DHA intake predicted the RBC DHA composition for males and for nonsmoking male patients. All correlations were significant and positive. Hibbeln et al. concluded that sex and current smoking status should be accounted for in research since they are strongly related to the intake of omega-3 fatty acids and to fatty acid compositions.⁶⁰ Finally, while our review could not determine whether PUFA levels indeed reflect the mechanism responsible for clinical effects, in no small part because we remain uncertain that reliable or robust clinical benefits actually exist for any psychiatric disorder or condition, we believe that their possible role in producing clinical benefits behooves researchers to assess their possible influences in new studies.

Turning our attention to the issue of outcomes, investigators conducting future research likely need to identify all of the most clinically pertinent outcomes for a given disorder or condition. The reason is that significant clinical benefits observed in mental health research can be outcome-dependent. A recent meta-analysis demonstrated this with respect to treatment studies examining the impact of short-acting Ritalin® on AD/HD.¹³ Significant clinical benefits were seen only for a subset of the central problems defining the presence of various forms of AD/HD. This observation suggests that each of the key symptoms defining a psychiatric disorder should likely be measured in new research.

To adequately investigate questions concerning the association between omega-3 fatty acid intake or the PUFA content of biomarkers and the onset, continuation or recurrence of psychiatric disorders or conditions, many of the controls discussed thus far are indicated. The ideal design to permit drawing causal inferences about etiology is a prospective and controlled one.

Primary prevention RCTs may be the best way to examine the possible protective role of omega-3 fatty acid intake, although the length of followup required to establish meaningful effects may necessitate studies that are exceptionally long and too expensive to conduct. Prospective cohort studies investigating the possible association between clinical outcomes and omega-3 fatty aid intake or PUFA status of biomarkers are likely a good choice yet they, too, might be too costly given the exposure period required to permit the detection of incident cases. When it comes to the question relating the intake of omega-3 fatty acids and review-pertinent clinical outcomes (e.g., onset), while prevention RCTs could implement precisely defined interventions, involving capsule-based supplementation, prospective cohort studies are more likely to assess patterns of consumption of foods containing omega-3 fatty acid content. The latter studies, while perhaps demonstrating greater ecological validity, nevertheless suffer from difficulties in precisely delineating the types and amounts of omega-3 fatty acid content associated with clinical outcomes.

A question that would need to be resolved likely prior to undertaking any of these studies relates to the timing of the intervention or exposure period. That is, when should a study with either a treatment or prevention focus begin, and end? The answer is likely disorder-specific, and should be based on what is known from clinical practice and research regarding the age of onset for the disorder. It probably makes little sense to state, without distinction, that these studies should all begin shortly after birth, with followups to occur every 2 years, for 20 years or more.

What may be more useful is to think somewhat outside the box. For example, it may make some sense to “piggyback” the assessment of mental health outcomes in well-controlled and adequately-powered studies that initially aim to assess the impact of omega-3 fatty acid supplementation, via formula feeding, on growth, visual, neurodevelopmental or cognitive outcomes. With time, and and further assessments of these outcomes, what began as an RCT could eventually become a less-expensive, single group observational study designed to identify the development of possible problems in mental health. Five-year telephone followups could be conducted to assess recent intake of (foods or supplements containing) omega-3 fatty acids and mental health status, such that if there is evidence that psychiatric symptoms or disorders may have emerged since the last followup, the individual could be seen and assessed formally. What this approach could offer is an opportunity to relate the intake of omega-3 fatty acids both early and later in life with the development of psychiatric disorders or conditions. Moreover, the inter-relationships among the various outcomes assessed in the study could reveal the broadest developmental context possible within which to understand, at the very least, the etiology of psychiatric disorders or conditions. As in most early intervention studies of formula supplementation involving PUFAs, a reference group of breastfed offspring could be followed in parallel. Assessments of the fatty acid content of biomarkers could be included in such an endeavor, including during pregnancy and at various times post-delivery.

That said, where the questions regarding the association between the onset, continuation or recurrence of psychiatric disorders or conditions and the intake of omega-3 fatty acids or the fatty acid content of biomarkers are concerned, it may be sufficient to avoid both cross-sectional studies and cross-national ecological analyses. Neither design generates data that allow us to resolve either of these questions. Case-control studies constitute an option,¹⁰⁹ although this design likely better suits the question examining the relationship between omega-3 fatty acid intake and the onset, continuation or recurrence of psychiatric disorder or conditions.

What also needs to be determined via new research is whether patterns of omega-3 fatty acid intake or patterns in the fatty acid content of biomarkers can prevent a psychiatric symptom (e.g., feelings of dysphoria) from developing into a full-fledged disorder. As well, what remains to be seen is whether patterns of omega-3 fatty acid intake or patterns in the fatty acid content of biomarkers can predict and perhaps prevent the continuation or recurrence of psychiatric disorders or conditions. It is conceivable that additional research could someday illumine the secondary preventive value of the intake of omega-3 fatty acids, for example.

In general, if researchers ever hoped to establish or reinforce the plausibility of employing omega-3 fatty acids as a treatment for any psychiatric disorder or condition using evidence from studies examining the associations between clinical outcomes and the intake of omega-3 fatty acids as well as the fatty acid content of biomarkers, it is our view that this has not been achieved. Therefore, the recent publication of treatment studies directed at depression, and especially schizophrenia, may suggest either that researchers have interpreted these data in ways that diverge greatly from our interpretation or, that they have perceived some other, inherent value in undertaking treatment trials.

Nonetheless, if future research is going to produce data that are unequivocally applicable to North Americans, it will likely need to enroll either North American populations or populations exhibiting a high omega-6/omega-3 fatty acid intake ratio similar to what has been observed in the diet of North Americans. It is our view that the dietary omega-6/omega-3 fatty acid intake ratio may eventually be seen to play an important role in the prevention and treatment of psychiatric disorders or conditions.

Limitations of the Review

While there are some limitations characterizing the present systematic review, almost none could likely be considered a serious impediment to the interpretation of the evidence we identified and synthesized. Overall, we found too few studies investigating a given question, and employing an appropriate research design of sufficient size and sound methodology, to have these limitations alter the view that, at present, we cannot conclude anything definitive about the disorder/condition-specific or overarching roles of omega-3 fatty acids in mental health. As well, the possible roles played by likely covariates and confounders could hardly be evaluated at all. We were limited in what we could observe because of the paucity of relevant studies per question and because many studies did not specifically investigate the influence of these variables. This is unfortunate since alcohol consumption, for example, like current smoker status, appears to influence both mental health and PUFA status.

The only limitations of possible significance concern the meta-analysis conducted with data obtained from RCTs investigating the efficacious use of omega-3 fatty acids as supplemental treatment for schizophrenia. It was less than ideal to use post-treatment means from Peet et al.'s study⁵⁸ when data indicating changes from baseline in outcomes are preferred. Another study report by Peet and colleagues, also used in the meta-analysis, did provide these data.⁸⁷ Unfortunately, our request for change from baseline data from the Peet et al. study⁵⁸ did not yield a successful response. Furthermore, what remains unknown at this time are the independent or combined impacts of combining data from studies failing to distinguish outcome data from patients receiving different primary medications,^{58, 87} and using different interventions (i.e., purified⁸⁷ versus unpurified EPA;⁵⁸ controlled⁸⁷ versus uncontrolled dosing⁵⁸), on the meta-analytic estimate or its precision. It should be recalled that there is, for example, limited empirical evidence indicating that low-dose (2 g/d) EPA yielded a clinical benefit solely for those patients receiving clozapine.⁸⁷ Replication efforts are required, however, before we can feel confident in the reliability of this observation.

Knowing, in advance, that data from cross-sectional designs could not possibly permit drawing causal inferences about the etiology of psychiatric disorders or conditions, from data reflecting either the dietary intake of omega-3 fatty acids or the PUFA content of biomarkers, might have afforded a decision to a priori exclude these studies from the review. However, this would have left us with few studies to review; and, while the purpose of our review did not include testing the deficiency hypotheses regarding the onset of depression or schizophrenia, reviewing these studies (or cross-national ecological analyses) nevertheless allowed us to highlight the evidence typically used to support these etiologic explanations.

At the same time, it is unlikely that expanding our focus with respect to PUFA metabolism beyond the compositions or concentrations of PUFA metabolites (e.g., to include data regarding the possible presence or activity of enzymes involved in the processes of desaturation or elongation within the metabolic pathway) would have produced results revealing a less unclear picture concerning the association between the PUFA content of biomarkers and the onset, continuation or recurrence of a psychiatric disorder or condition. These relationships were typically investigated in similarly limited, cross-sectional designs.

As stated in Chapter 2, in light of the relatively limited details often provided in reports about the ways in which lipid samples were extracted, stored and analyzed, we could only readily identify situations where investigators described inappropriate methods. It is unclear how this state of affairs might have influenced the observations we gleaned from the evidence base concerning the role of omega-3 fatty acids in mental health.

Time constraints made it impossible to complete dual-assessor appraisals of the quality (i.e., internal validity) of studies employing designs other than an RCT, or the applicability of all the included studies. One experienced quality assessor conducted these evaluations. At the same time, we conducted these quality assessments of designs other than RCTs using items we either modified from existing instruments or which we had to develop outright because no similar tools existed (e.g., cross-national ecological analyses). A design-specific, total quality score was then generated for each study, from which a single summary value was derived (i.e., A, B, C). This simplification permitted the entry of these values into summary matrices. However, the design-specific cutpoints used to assign these values were established without any validational basis, and so their value is likely extremely limited. The applicability indices, while continuing the work we did when we systematically reviewed the evidence for the health effects of omega-3 fatty acids on asthma,⁷² also did not receive any validational support. Nevertheless, given the limited number of studies addressing a specific question, and using a design whose data could meaningfully elucidate it, it is unlikely that these shortcomings could have had a meaningful impact on the “take home messages” highlighted by our review. Formal statistical assessments of the impacts of study quality or applicability on study outcomes could not be conducted. Finally, with a very limited number of studies entered into meta-analysis, an examination of the possible presence and impact of publiation bias could not be conducted.

Conclusion

Studies investigating omega-3 fatty acids employed as an intervention revealed the absence of a notable safety profile associated with any type or dose of omega-3 fatty acids represented in the review. Only with respect to the supplemental treatment of schizophrenia is the evidence even somewhat suggestive of omega-3 fatty acids' potential as a shortterm intervention. Even then, these results pertaining to 2g/d, or low-dose, EPA^{58, 87} need to be replicated using larger sample populations, longer investigative periods and instituting various methodologic (i.e., experimental or analytic) controls. The observed failure of high-dose (i.e., ≥3 g/d) EPA supplementation to produce a clinical benefit likewise requires replication with similar design modifications.

Data regarding the supplemental treatment of depression suggest a focus where considerable additional, clarifying research might eventually reveal the shortterm or longterm therapeutic value of omega-3 fatty acid supplementation. One study demonstrating a significant placebo-controlled clinical effect related to 1 g/d E-EPA given, over 12 weeks, to 17 patients with depressive symptoms—rather than depressive disorders—cannot be taken to support the view of the utility of this exposure as a supplemental treatment for depressive symptomatology or disorders.⁵³ Equally inadequate to establish the efficacy of omega-3 fatty acids as a supplemental treatment for depression are two other trials,^{96, 97} which lasted 4 or 8 weeks and employed active exposures and exposure-placebo contrasts that distinguish them from the study which highlighted the efficacy of 1g/d E-EPA.⁵³

Nothing can yet be concluded concerning the clinical utility of omega-3 fatty acids provided as a supplemental treatment for any other psychiatric disorder or condition, or as a primary treatment for all psychiatric disorders or conditions, examined in our review. Primary treatment studies were rare. Much more research is needed before we can begin to ascertain the possible utility of (foods or supplements containing) omega-3 fatty acids as a primary prevention for psychiatric disorders or conditions. From both an economic and scientific point of view, it might be worthwhile to “piggyback” studies of the primary protective potential of omega-3 fatty acids in mental health onto controlled, longitudinal studies of their impact on general health and developmental outcomes (e.g., growth; neurodevelopment; visual and cognitive development). Requisite modifications for treatment or prevention studies include well-defined and appropriately sampled populations, followup periods of suitable lengths, key experimental or analytic controls (e.g., for confounders) and individual/patient-level data. Overall, almost nothing is known about the therapeutic or preventive potential of each source, type, dose or combination of omega-3 fatty acids.

Because of limited study designs, little is known about the relationship between PUFA biomarker profiles and the onset of any psychiatric disorder or condition. Studies examining the possible association between the intake of omega-3 fatty acids, or the PUFA content of biomarkers, and the continuation or recurrence of psychiatric disorders or conditions were virtually nonexistent.

If future research is going to produce data that are unequivocally applicable to North Americans, it will likely need to enroll either North American populations or populations exhibiting a high omega-6/omega-3 fatty acid intake ratio similar to what has been observed in the diet of North Americans. Furthermore, if a reasonable view is that omega-3 fatty acids may play a role in mental health, then given the observed or proposed inter-relationships between omega-3 and omega-6 fatty acid contents both in the regular diet and in the human biosystem, it may behoove researchers to investigate the possible therapeutic or preventive value of the dietary omega-6/omega-3 fatty acid intake ratio.

To this end, interventional studies could concurrently modify the intake of omega-3 and omega-6 fatty acids, and thereby manipulate experimentally the omega-6/omega-3 fatty acid intake ratio. Prospective observational studies could trace the possible links between the omega-6/omega-3 fatty acid intake ratio and the development, course or outcome of psychiatric disorders or conditions. Finally, any notable causal or correlational relationships observed between the omega-6/omega-3 fatty acid intake ratio and the development, course or outcome of psychiatric disorders or conditions might then be “explained” by observed patterns of omega-6/omega-3 fatty acid content in peripheral, or even brain, biomarkers.