|Home | About | Journals | Submit | Contact Us | Français|
Randomized trials of omega-3 polyunsaturated fatty acid (PUFA) treatment for depression have differed in outcome. Recent meta-analyses ascribe discrepancies to differential effects of eicosapentaenoic acid (EPA) vs. docosahexaenoic acid (DHA) and to diagnostic heterogeneity. This meta-analysis tests the hypothesis that EPA is the effective component in PUFA treatment of major depressive episodes.
PubMed was searched (1960 through June 2010) using terms “Fish Oils”[Mesh] AND (“Depressive Disorder”[Mesh] OR “Bipolar Depression”) AND “Randomized Controlled Trial”[Publication Type], for placebo-controlled trials of PUFA supplementation, a depressive episode as primary disorder, published in English, supplemented by manual bibliography review.
The search yielded 15 trials involving 916 participants.
Sample sizes; PUFA doses; mean ages, baseline and endpoint depression ratings and standard deviations; and p values were extracted.
In a mixed-effect model, percentage of EPA in the supplements was the fixed-effect predictor, dichotomized into two groups: EPA < 60% or EPA ≥ 60% of EPA + DHA. Secondary analyses explored relevance of treatment duration, age, and EPA dose.
Supplements with EPA ≥ 60% showed benefit on standardized mean depression scores (SMD, for EPA ≥ 60% = 0.558, 95% CI = (0.277, 0.838), z = 4.195, p = 0.001; for EPA < 60% = −0.026, 95% CI = (0.200, 0.148), z = −0.316, p =0.756), with negligible contribution of random effects or heteroscedasticity, and no effects of treatment duration or age. Supplements with EPA < 60% were ineffective. Exploratory analyses supported a non-linear model, with improvement determined by the dose of EPA in excess of DHA, within the range 200 to 2200 mg EPA.
Supplements containing EPA ≥ 60%, in dose range 200 to 2200 mg EPA in excess of DHA, were effective against primary depression. Translational studies are needed to determine mechanisms of EPA’s therapeutic benefit.
Low levels of omega-3 polyunsaturated fatty acids (PUFA) have been linked to depression1,2 and suicide3, as well as to cardiovascular4,5 and inflammatory disorders6, and thus may impact comorbidity of depression with diseases such as coronary heart disease7,8 and diabetes9. Previous meta-analyses disagree as to the benefit of omega-3 fatty acid supplementation for depression.10–14 However, the trials performed to date vary in important methodological aspects, including the type of placebo, diagnoses, monotherapy vs. augmentation, and doses and proportions of eicosapentaenoic acid (20:5n-3; EPA) and docosahexaenoic acid (22:6n-3; DHA) in the supplements. Two factors have recently been proposed to account for discrepancies between studies: a greater efficacy of EPA than DHA11,13, and greater effectiveness in patients with a diagnosed depressive disorder13,14. The a priori goal of this meta-analysis was to test the hypothesis that EPA is the active component of omega-3 PUFA treatment in depressive disorders. This study extends previous work by including recent clinical trials15,16 not reviewed in prior meta-analyses, and by proposing a novel model to explain the effects of EPA dosing. Determination of the most effective omega-3 PUFA supplementation regimen is important for treatment of depression and for design of future research studies.
Published studies eligible for this analysis were identified through a search of clinical trials in PubMed/MeSH (1960 through June 2010) using the following terms: “Fish Oils”[Mesh] AND (“Depressive Disorder”[Mesh] OR “Bipolar Depression”) AND “Randomized Controlled Trial”[Publication Type] and limited to published articles written in English. The reference lists within the resulting publications and relevant review articles were also examined to check for completeness of the assembled list of studies.
Trials were included if they met the following inclusion criteria: (1) prospective, randomized, double-blinded study design; (2) depressive episode as the primary complaint (with or without comorbid medical conditions); (3) administration of omega–3 PUFA supplements; (4) appropriate outcome measures to assess depressed mood; (5) a placebo comparison group, and (6) published in English.
Data extracted were study design, sample size, dose and percentage of EPA and DHA, subject mean ages, mean baseline and endpoint depression ratings and standard deviations (SDs) for PUFA and placebo groups, and p values. Mean and SD values not included explicitly in the published reports of Grenyer et al.17, da Silva et al.18 and Silvers et al.19 were provided electronically by the authors (Howe, P., PhD, written communication, May 28, 2010; Ferraz, A.C., PhD, written communication, June 24, 2010; Silvers, K.M., PhD, written communication, August 14, 2010, respectively). EPA was quantified as a percentage of (EPA + DHA) in the supplement, ranging from 0 (in one trial with DHA alone) to 100% (ethyl-EPA alone).
The clinical outcome of interest was standardized mean difference in the change from baseline to endpoint scores on a depression rating scale, in subjects taking PUFA supplements vs. subjects taking placebo. Trials used the Hamilton Depression Rating Scale20 for the main outcome measure except for 4 studies17,18,21,22 that used the Beck Depression Inventory23, the Montgomery–Asberg Depression Scale24, the Childhood Depression Rating Scale25, and the short form of the Depression, Anxiety, and Stress Scale26, respectively.
Statistical analyses were performed using R27 (R Foundation for Statistical Computing, Vienna, Austria). Where means and SDs for baseline and endpoint were available for both groups, the effect size was calculated according to the method of Hedges.28 The difference in mean baseline-endpoint change between PUFA and placebo groups was divided by an estimated SD of the change, calculated by pooling baseline and endpoint SDs in each group and multiplying by 21/2. This technique assumes that baseline and endpoint values are uncorrelated, whereas in actuality they are probably positively correlated. This conservative assumption, therefore, is likely to overestimate SD’s of the change and result in smaller estimated magnitude of effect sizes. In one study, use of the Hedges method was not possible due to limited specificity of information29, so effect size was calculated from p-values30, and standard error (SE) was imputed via a regression of SE on the reciprocal of the square root of the study size, which in this sample strongly correlated with SE (r = 0.96). The SD of the group difference was obtained by pooling SDs of the placebo and treatment groups.
A regression analysis was used to study the contribution of the EPA proportion to the effect size for omega-3 PUFA supplementation compared to placebo. The predictor variable for the fixed-effect part of the model was the percentage of EPA in the supplement, dichotomized into two groups: EPA less than 60% of DHA + EPA concentrations, or EPA greater than or equal to 60% of EPA + DHA concentrations. This cutoff was chosen based on the empirical observations that all significant positive studies used at least 60% EPA and all studies with less than 60% EPA were negative (the remaining studies used at least 60% EPA but were negative).
Two reports29,31 tested different doses of 100% ethyl-EPA; individual dose analyses within each paper were treated for purposes of meta-analysis as separate trials. This and other cases of studies by the same author may be statistically dependent, violating a basic assumption of ANOVA. To test for author effects, two classes of models were generated for the random part of the mixed model: models including ‘author’ as a random effect and those in which all studies were regarded as independent. Another concern was whether the precision of the estimated effect size might depend on study size, e.g. studies with fewer subjects might have larger variance. This and two other potential conditions of heteroscedasticity in study-wise SD were tested a priori in the mixed models: study-wise SD depended on 1) sample size or 2) EPA dichotomized at 60%, or 3) was constant. Using EPA dichotomized at 60% as the fixed-effect term and all combinations of the 2 random-effects and 3 heteroscedasticity possibilities, a family of 6 mixed-effects models was generated. One additional model was fitted, a weighted least squares regression32 with weights proportional to the reciprocal of estimated study effect size SE. The model with the smallest Bayes Information Criterion (BIC) value33 was chosen. As a further check, a Welch Two-Sample t-test was also performed, which is not based on an assumption that the SD’s in the high and low percentage EPA groups are equal.
Two sources of heterogeneity were feasible to test, given the information available in the trials included in the meta-analysis: treatment duration and mean age, included as covariates in separate regression analyses. In these analyses, the same family of 7 models was utilized, the dependent variable remained effect size, and the predictors were EPA dichotomized at 60% plus one of the covariates; interactions were also tested. Publication bias was assessed with a funnel plot.
Given the observed 60% threshold for significance, it was hypothesized that EPA was effective to the extent that it was in excess of DHA in the supplements. Therefore, correlations were examined between effect size and EPA dose in excess of DHA (EPA dose – DHA dose), where positive numbers represent EPA in excess, and negative numbers represent DHA in excess. A second observation was negative outcomes in 2 published studies29,34 at doses of pure ethyl EPA greater than or equal to 4,000 mg. Therefore, a non-linear relationship of EPA dose to effect size was proposed. Linear and non-linear regression models were empirically fit to the data, using the curve-estimation module from SPSS (Release 17.0.0, 2008, SPSS Inc., Chicago, IL, for Mac [Apple, Inc., Cupertino, CA]). Weighted linear and quadratic least squares regression analyses were also performed, using weights proportional to the reciprocal of estimated study effect size SE. No correction was made for multiple testing.
Twenty-four reports were identified through the MeSH/PUBMED search strategy, excluding 3 studies that were not clinical trials, 4 in which the primary diagnosis was not depression, and 3 without a placebo arm. One additional article was identified through manual bibliography search, resulting in 15 double-blinded, placebo-controlled trials that fulfilled all criteria, involving 916 participants (see Table 1). Eight studies included participants with diagnosed Major Depressive Disorder16,17,21,35–39. Two studies concerned participants with a Major Depressive Episode in association with a medical illness: Parkinson’s disease18 and coronary heart disease15. One study enrolled participants with a Major Depressive Episode in context of Bipolar Disorder31. The remaining 4 studies defined the diagnostic criteria as “episode of major depression or dysthymia”40, “ongoing depression”29, “a current depressive episode”19, or “mild to moderately depressed”22. In 3 studies, depression occurred in context of pregnancy or the perinatal period35,39,40. PUFA was given as monotherapy in 6 trials 16,21,22,36,39,40, and in one trial18 as one arm of the study. The remainder gave PUFA as adjunctive to pharmacotherapy15,17,19,29,31,37,38 or psychotherapy35. All studies used an intent-to-treat analysis except one study38 that excluded 4 subjects by placebo lead-in and 6 subjects after randomization, and another study18 in which 2 patients dropped out and were not included in the efficacy analysis. The percent composition of the supplements spanned the entire range from 100% EPA to 100% DHA; doses ranged from 400–4,400 mg/day of EPA and 200–2,400 mg/day of DHA.
The overall effect size for 60% or greater EPA in supplements compared with placebo was 0.558 (p<0.001); for EPA at less than 60%, it was non-significant at −0.026, (see Table 2 and Figure 1). Interpretation of these findings should take into account that asymmetry of the funnel plot indicated some negative publication bias (see Figure 2).
For primary and secondary analyses, the p-values of dichotomized EPA in all models were robust, ranging from 0.00046 to 0.00165. Results from models with the lowest BIC are summarized in Table 2. In the primary regression analyses, the best model was the weighted least squares regression, in which an EPA proportion of at least 60% was a significant determinant of superiority of PUFA over placebo (t = 4.19, df = 17, p-value < 0.001). A Welch Two Sample t-test confirmed the significance of the effect (t = 5.10, df = 16.83, p-value < 0.0001). In secondary covariate analyses, neither treatment duration nor age significantly predicted effect size; an EPA proportion of 60% or greater was still significant with either variable in the model. Interactions were not statistically significant.
In exploratory analyses, EPA dose in excess of DHA (EPA (mg) – DHA (mg)) correlated similarly with effect size using either a linear (F=4.054, p=0.060, df1=1, df2=17) or a quadratic (F=3.399, p=0.059, df1=2, df2=16) function; neither reached significance (see Figure 3). However, weighted least squares regression analyses using weights proportional to the reciprocal of estimated study effect size SE, were significant for both linear (F=4.843, p=0.018, df1=1, df2=17) and quadratic (F=3.993, p=0.039, df1=2, df2=16) approaches.
In agreement with Ross et al.11 and Martins13, this study identifies EPA as the effective PUFA component in treatment of depression. This finding is in contrast to the greater face validity of DHA, which is the major brain omega-3 PUFA species and is lower in brains of depressed subjects in postmortem studies2. The lack of DHA efficacy could mean that acute supplementation does not increase brain DHA concentrations. Increases in brain DHA have been reported after supplementation in piglets41 and in rats42. The effect of dietary DHA supplementation on human brain levels has not been studied; however, intravenously injected radiolabelled DHA43 resulted in an extremely low rate of DHA incorporation into brain in healthy humans: 3.8 + 1.7 mg/day, or a whole-brain half-life of 2.5 years. If this is an accurate paradigm for the fate of dietary DHA, then as noted by Umhau et al.43, effects of supplementation would not be evident in clinical trials lasting a few weeks, and the delay would be impractical for a therapeutic agent.
The inflammatory hypothesis of depression is based on the observations that stress precipitates both inflammatory responses and depression, inflammatory markers are increased in depression, and inflammatory cytokines can produce depressive symptoms in humans.51–53
Long-chain PUFAs and their metabolites have immunomodulatory properties.6 There is a functional opposition between omega-3 and omega-6 PUFA, in which higher relative levels of omega-3 tend to reduce the production of pro-inflammatory eicosanoids and cytokines50,52,54. Ratios of omega-6 to omega-3 PUFA are elevated in depression55–59 and in suicide risk3. These findings are in agreement with a theory proposing arachidonic acid (AA) cascade abnormalities as a cause of mood dysregulation.60–62 EPA has also been proposed29, specifically, as an important competitor with AA. For example, 1) Differences in EPA/AA ratios affect membrane fluidity and cellular responsivity50; 2) EPA competes with AA for cyclo-oxygenase, increasing production of anti-inflammatory prostaglandins29,50; and 3) lower EPA levels have been found associated with a genetic variant of phospholipase A2 (PLA2) that increased risk of interferon-induced depression63.
Increased PUFA oxidation could increase ketogenesis, producing ketone bodies that could bypass glucose utilization and improve energy supply to the brain.64 Increased fatty acid oxidation decreases production of triacylglycerol in rat hepatocyte cell cultures65 and increases fasting glucose concentrations in hyperlipidemic men66. Despite its low concentration in hepatocytes, EPA is a much stronger activator than DHA of peroxisome proliferator-activated receptor α (PPARα)67, an important regulator of energy homeostasis and PUFA β-oxidation68.
EPA supplementation in Bipolar Disorder has been observed to increase brain N-acetyl-aspartate69, a marker for neuronal health. EPA supplementation for 9 months also increased the ratio of cerebral phosphomonesters to phosphodiesters, an indicator of phospholipid turnover, and reversed brain atrophy, in a subject with Major Depressive Disorder.70 No comparable studies have been performed with DHA.
The role of dose in PUFA supplements has been difficult to understand. Although EPA at ratios greater than 60% positively affected depression outcome, both successful 21,29,31,37–39 and unsuccessful 15–17,19,22,29,35,40 trials used EPA doses in the same ranges (400 to 4,000 mg/day). To address the effects of dose, we propose the following theoretical model:
The right-hand, descending portion of the quadratic curve is supported by a lone point at 4000 mg of ethyl-EPA29. However, we note the existence of another clinical trial34 in BD not included in this meta-analysis (as the sample comprised depressed and rapid-cycling patients), in which 6,000 mg of pure ethyl-EPA was not superior to placebo. It has been puzzling that these two well-designed studies were negative, as they seem to be comparable to similar, successful trials at doses of 4,400 mg of EPA in MDD38 and 6,200 mg of EPA in BD72. The problem was not the use of pure ethyl-EPA, which has been successfully used to treat depression in several clinical trials16,29,31,37. Rather, we note that in the latter, successful studies38,72, the unopposed doses of EPA were actually only 2,200 and 2,800 mg/day, respectively, consistent with our model. Thus, although the linear regression was also statistically valid, we feel that the U-shaped response curve is more likely to reflect the reality of the clinical response, although it is currently unknown why high doses of EPA may not be effective.
In a more broadly defined population, Martins13 found greater PUFA effects with shorter treatment length. In this meta-analysis, which included studies ranging from 4 to 16 weeks in duration, treatment length was not a predictor of outcome, suggesting that for patients who have a diagnosed depressive illness, effects of EPA may not be limited to the initial treatment period.
This meta-analysis did not take into account unpublished clinical trials that would be predicted by the asymmetric funnel plot to exist. The number of potential moderators examined was limited by considerations of statistical power and inconsistent information in the source articles. Unexamined covariates that might be relevant include baseline level of depression, presence of stabilizing antioxidant in the supplement47, response by sex or ethnicity, baseline plasma PUFA levels, and dietary intakes. The selection of a diagnostic phenotype for study was limited by the relatively small number of clinical trials primarily focusing on depression, and by a lack of diagnostic clarity in some of the studies. Thus no inferences can be made about depressive episodes occurring within Major Depressive Disorder as opposed to Bipolar Disorder. The theoretical model to explain dose effects is based on a small number of studies and must be tested prospectively.
Recently, experts have called for more widespread use of omega-3 supplementation in patients at risk for depression10,73. However, there are no current agreed-upon guidelines concerning the optimal balance of constituents in omega-3 supplements. This meta-analysis finds no evidence that DHA is acutely effective against depression, and in fact, it may block beneficial effects of EPA at about a 1:1 dose ratio. Thus the amount of EPA unopposed by DHA may be critical for effective PUFA supplementation in depressive episodes. These findings argue against additional brief clinical trials of DHA for depression. At present, our knowledge base supports the use in acute depression of omega-3 supplements containing at least 60% EPA, with a ceiling at around 2,000 mg of EPA in excess of DHA, although the therapeutic effects of different unopposed EPA doses should be tested further in prospective studies that take into consideration diet and other potential confounds. We note that long-term efficacy and health effects of PUFA supplementation in depression have yet to be evaluated. Translational studies are also required to understand mechanisms underlying EPA effects in depression.
Sources of Financial Support: This work was partially supported by NIMH grants MH040695 and K08MH079033-03.