Omega-3 supplementation reduces aggressive behavior: A meta-analytic review of randomized controlled trials

Abstract

There is increasing interest in the use of omega-3 supplements to reduce aggressive behavior. This meta-analysis summarizes findings from 28 RCTs (randomized controlled trials) on omega-3 supplementation to reduce aggression, yielding 35 independent samples with a total of 3,918 participants. Three analyses were conducted where the unit of analysis was independent samples, independent studies, and independent laboratories. Significant effect sizes were observed for all three analyses (g = .16, .20, .28 respectively), averaging .22, in the direction of omega-3 supplementation reducing aggression. There was no evidence of publication bias, and sensitivity analyses confirmed findings. Moderator analyses were largely non-significant, indicating that beneficial effects are obtained across age, gender, recruitment sample, diagnoses, treatment duration, and dosage. Omega-3 also reduced both reactive and proactive forms of aggression, particularly with respect to self-reports (g = .27 and .20 respectively). It is concluded that there is now sufficient evidence to begin to implement omega-3 supplementation to reduce aggression in children and adults - irrespective of whether the setting is the community, the clinic, or the criminal justice system.

1. Introduction

There is growing interest in the use of nutritional supplements to ameliorate aggressive and antisocial behavior (Olivia Choy, 2023; Hibbeln & Gow, 2014). This interest is in part stimulated by long-standing research showing that poor nutritional status is a risk factor for externalizing behavior problems (Neugebauer, Hoek, & Susser, 1999; Raine, Mellingen, Liu, Venables, & Mednick, 2003). In particular, omega-3 has been hypothesized as one nutritional component that could explain the link between poor nutrition and aggressive/violent behavior (Choy & Raine, 2018). Correlational research has also shown that fish consumption is negatively associated with cross-country homicide rates (Hibbeln, 2001). Importantly, experimental research in humans based on randomized controlled trials (RCTs) have argued that omega-3 supplementation can reduce aggression in a wide variety of human populations (Adams et al., 2018; Hamazaki et al., 1996; Raine, Leung, Singh, & Kaur, 2020). Nevertheless, the variability in results from RCTs raise questions on whether omega-3 supplementation can reduce human aggression.

The notion that omega-3 supplementation can produce a decline in aggression has been lent credence by a meta-analysis conducted by Gajos & Beaver (2016) which observed an overall effect size of .24 for intervention studies based on a random effects model. Nevertheless, there are two important limitations that give rise to caution over this conclusion. First, although the analysis was argued to be based strictly on aggression, 12 of the 30 intervention studies instead consisted of oppositional defiant disorder, conduct disorder, and externalizing behavior outcomes – variables which did not include explicit measures of aggression. Of the remaining 18 studies, one did not have a control group, leaving only 17 studies out of the 30 original studies measuring aggression as an outcome. As such, it is unclear whether the effect size of .24 pertains to aggression per se, or instead to these antisocial constructs, as analyses were based on all 30 aggregated studies without sub-analyses on only studies with aggression as an outcome.

A second methodological limitation of this initial meta-analysis consists of the handling of studies with multiple aggressive/antisocial outcomes. Where a single study had several different aggression measures as outcomes, effect sizes were calculated for each measure which were then entered as separate effect sizes, resulting in 54 effect sizes from the 30 studies contributing to the overall effect size. Entering multiple effect sizes from several aggression measures from one study creates dependence among effect sizes which in turn results in a violation of standard meta-analytic models (Lipsey & Wilson, 2001), a fact that Gajos & Beaver recognized. A different approach in which each study contributed just one effect size to the analysis would be of value, as those laboratories utilizing multiple aggression measures would then not exponentially weight the overall effect size, and statistical independence would not be lost. Despite these limitation, Gajos & Beaver (2016) have provided a valuable initial contribution that highlights the issue of whether omega-3 can be viewed as a viable treatment strategy to reduce aggression.

A question remains as to whether omega-3 is effective for all forms of aggression. One prominent distinction in the field is that between reactive, impulsive aggression, and proactive, predatory aggression (Blair, 2010). Several RCTs on omega-3 have found that there may be better evidence for treatment efficacy for reactive aggression compared to proactive aggression. For example, one study of prisoners found stronger evidence for treatment efficacy for a reactive-impulsive form of aggression compared proactive, psychopathic-like aggression (Raine et al. 2020), and similar effects have been found with children (Raine et al. 2016). Finding conflict however, with other studies finding equal efficacy for both reactive and proactive aggression (Raine et al. 2015). One goal of the current study therefore was to evaluate more systematically whether reactive aggression is indeed more influenced by omega-3 supplementation, or whether both forms of aggression can benefit from this intervention.

The current study aims to address these two limitations by excluding studies that do not specifically measure aggression (antisocial / externalizing behavior) and by averaging the results of multiple measures of aggression within one study, in addition to expanding the number of studies for inclusion. Several other issues were addressed. In addition to calculating an overall effect size where the unit of analysis is independent samples, a second analysis was conducted in which the unit of analysis was independent studies. In this latter approach, where one publication reported on several independent samples, these were averaged to contribute one effect size. A third analysis used laboratory as the unit of analysis, whereby effect sizes from several studies emanating from the same laboratory were averaged, thus preventing a few laboratories overly influencing the overall effect size. Moderators were also explored in order to examine any heterogeneity in findings; important questions concern whether findings apply to males and females and to children and adults, as well as whether treatment duration and dosage influence effects. A separate analysis was also conducted on studies reporting on both reactive and proactive aggression as this distinction has not previously been examined in a meta-analysis, together with a moderator analysis of reporter source (self-report vs observer report) for these two sub-types.

2. Methods

2.1. Inclusion Criteria

Studies were included in the meta-analysis if they met the following inclusion criteria: (1) consisted of an RCT, (2) explicitly measured aggression, (3) provided supplementation with omega-3, (4) conducted on humans, (5) provided sufficient information to compute an effect size, (6) published in English. We excluded studies involving prenatal supplementation to the mother during pregnancy and effects on the offspring. We did not exclude studies which also included additional nutritional supplements (in the majority of cases calcium and vitamin D), but we examined this as a potential moderator.

2.2. Search Strategy.

Searches were conducted in 7 data-bases as follows: PubMed, PubMed Central/NIH National Library of Medicine, PsychInfo, Cochrane, Medline, Web of Science, and ProQuest. Searches were conducted up until December 2023, with no limits on publication date, country, and age in order to include as many eligible studies as possible. Two coders (AR and LB) independently extracted the data according to a predetermined list of interests. A list of potentially relevant studies was produced for a later full-text analysis for meeting inclusion criteria.

Search terms consisted of the following: Docosahexaenoic OR Eicosapentaenoic OR Docosapentaenoic OR alpha-linolenic acid OR Omega-3 OR “fish oil” OR DHA OR EPA OR DHA OR ALA AND (aggress* OR viol* OR anger OR angry OR hostil*) AND (“randomized controlled trial” OR RCT). In addition, we checked reference sections of articles included in the meta-analysis in addition to other reviews for links to other relevant articles.

2.3. Outcome Measure

The search was focused on aggression in particular and not broader traits such as anger and hostility, even though the latter were included in the search term (see below). We attempted to focus on aggressive behavior, and excluded anger which is viewed more as a feeling (emotion) and hostility which is more viewed as an attitude (cognition). We also excluded studies assessing mood, suicide, self-harm, emotional lability, irritability, and affect dysregulation.

2.4. Data Extraction

Because only one laboratory out of 19 followed up participants after the cessation of supplementation, the outcome analyzed here consists of change in aggression from beginning to end of treatment for experimental and control groups. Means, SDs, and sample sizes were extracted from each study. Effect sizes consisted of group mean differences and standard deviations (SD). If change from baseline scores were not given, these values were computed using baseline and post-test means and SDs for each group. If SDs were missing, we applied methods outlined in the Cochrane Handbook for Systematic Reviews of Interventions in order to estimate SDs and calculate effect sizes (Cumpston, McKenzie, Welch, & Brennan, 2022). Authors were also contacted to ascertain additional information where necessary. The effect size consisted of Hedge’s g which corrects for bias due to small sample size. A positive effect size indicates a stronger reduction in aggression in the omega-3 group compared to controls. For each study the effect size and its associated standard error were calculated for entry into SPSS.

Where several measures of aggression were reported, effect sizes were calculated for each measure and a meta-analysis run on these effect sizes to produce an averaged effect size. Multiple effect sizes from several aggression measures from one study were not entered separately because this creates dependence among effect sizes which in turn results in a violation of standard meta-analytic models (Lipsey & Wilson, 2001). Similarly, when the aggression measure was completed by two informants (e.g. parent and child), effect sizes were averaged.

The following variables were also extracted as potential moderators for either sub-group (grouping) or meta-regression (continuous): child vs adult sample, clinical vs community sample, added vitamins/minerals to omega-3 supplement (yes/no), diagnosis (yes/no), externalizing / internalizing / non-clinical population, age (child, adult), gender (% females), treatment duration (weeks), omega-3 dose (g), ratio of DHA to EPA, publication year, country-wide fish consumption estimate, and study quality. Estimated fish consumption for the country in which the study was conducted was calculated based on cross-national seafood data (Micha et al., 2014). Study quality was based on the following 8 variables: large sample size (>173 based on a median split), use of a placebo (yes/no), double-blind employed (yes/no), sensitivity analyses run (yes/no), data imputed (yes/no), low dropout rate (< 5%), number of aggression measures averaged (> 1), and intention-to-treat analysis (yes/no).

2.5. Statistical Analyses

We conducted a random-effects meta-analysis using SPSS version 29.01 because we anticipated significant variability across studies due to differences in age, gender, aggression measure, dosage, and treatment duration. The fixed effects model which assumes that the true effect size is the same in all studies is viewed as implausible in cases of different studies on different populations from different laboratories using different measures (Borenstein, Hedges, Higgins, & Rothstein, 2021). An inverse variance method was employed to weight studies based on sample size to yield an estimate of the mean effect size. Standardized mean differences were based on the mean pre-post treatment change of the experimental group minus that of the control group, divided by the pooled standard deviation. Two-tailed tests and 95% confidence intervals (CI) were employed.

Heterogeneity was assessed using the Q statistic. The I² statistic, the ratio of genuine heterogeneity as a function of total variation, was used as a signal-to-noise measure to evaluate how much of any heterogeneity observed can be attributed to true variation in study effect sizes as opposed to sampling error (Borenstein et al., 2021).

Publication bias was evaluated in two ways. First, Egger’s regression was calculated, with a significant test indicating potential bias. Second, Duval and Tweedie’s trim-and-fill test was run, with asymmetry in the funnel-plots suggesting publication bias. If bias was detected, new studies were added to the plot and the overall effect size recalculated.

We conducted moderation analyses to probe how effects sizes for omega-3 supplementation varied as a function of both group and continuous moderators. Sub-group tests were based on the Q statistic tested using χ². For continuous moderators, univariate meta-regression analyses were conducted. The resulting beta value indicates how treatment effectiveness changes as a function of the moderator variable, with a significant coefficient indicating a linear relationship between the moderator and treatment efficacy (Page et al., 2021).

Sensitivity analyses were conducted to test robustness of findings. Findings from REML (restricted maximum likelihood) were re-run using maximum likelihood (ML), Empirical Bayes, Hedges, Hunter-Schmidt, DerSimonian-Laird, and Sidik-Jonkman estimators. These analyses were conducted first using no standard error adjustment, and then re-run applying the Knapp-Hartung adjustment which provided two adjustments to the standard error for the random effects model. This modification results in an expansion of the confidence interval and as such is a conservative approach (Borenstein et al., 2021).

3. Results

3.1. Study selection and characteristics of included studies.

Full details of the number of records screened and those excluded are given in the flow diagram in Figure 1. Out of 1,853 records screened, 28 published articles and one unpublished study met the inclusion criteria, yielding 35 independent samples with a total sample of 3,918 participants. Publication year ranged from 1996 to 2024, spanning almost three decades. Regarding child/adult composition, 51.4% were conducted on children (defined as below 18 years). Regarding population type, 48.5% of studies were community-based, 15.2% were on internalizing-spectrum populations (e.g. autism), and 33.3% were externalizing-spectrum populations (e.g. prisoners). Mean age was 21.74 (SD = 14.44). The percentage of females was 34.12% (SD=28.43). Mean duration of supplementation was 16.37 weeks (SD=9.04). Mean omega-3 dosage was 1.18g (SD = .56). Surprisingly, only two authors reported ethnicity. Diagnostic groupings were as follows: none/community recruited (17), ADHD (4), CD/ODD (6), intellectual disability (1), borderline personality disorder (2), autism spectrum (2), suicidality (1), psychosis (2).

Figure 1.

Flow diagram of literature search resulting in the 28 studies included in the omega-3 – aggression meta-analysis, resulting in a final N of 3,918 participants.

3.2. Meta-analytic results and risk of bias

3.2.1. Independent samples.

Effect sizes from the 35 independent samples from the 28 studies are given in the forest plot in Figure 2. The overall effect size was statistically significant (g = .162, p = .006, SE = .055, t = 2.94, CI = .050 to .275), with the prediction interval ranging from −.271 to .596. The Q statistic indicated significant evidence of heterogeneity in findings (χ² = 82.48, df = 34, p < .001). The I² was 67.8%, indicating that approximately two-thirds of this heterogeneity can be attributed to true differences between studies.

Figure 2.

Forest plot of the 35 independent RCT samples. Blue boxes indicate the effect size for each study along with the confidence interval (horizontal lines). The red diamond indicates the overall effect size (g = .162) along with its confidence interval. Some individual studies yield more than one independent sample (e.g. Hamazaki 2002a, 2002b). A positive effect indicates that omega-3 supplementation is associated with reduced aggression.

There was no evidence of publication bias. Egger’s regression test was non-significant (p = .89, intercept = −.012, SE = .088, t = −0.14, p = .89, CI = −.191 to .166). Similarly, Duval and Tweedie’s trim-and-fill test indicated no evidence of asymmetry in the funnel plot (see Figure 3).

Figure 3.

Funnel plot of the effect sizes drawn from the 35 independent samples (blue circles), indicating no evidence of publication bias. Dotted lines indicate pseudo confidence intervals.

3.2.2. Independent studies.

Effect sizes from the 28 independent studies are given in the forest plot in Figure S1. The overall effect size was statistically significant (g = .204, p = .007, SE = .070, t = 2.94, p = .007, CI = .063 to .347), with the prediction interval ranging from −.342 to .751. The Q statistic indicated significant evident of heterogeneity in findings (χ² = 73.98, df = 27, p < .001). The I² was 81.7%, indicating that most of this heterogeneity can be attributed to true differences between studies.

There was no evidence of publication bias. Egger’s regression test was non-significant (p = .78, intercept = .034, SE = .120, t = 0.28, p = .89, CI = −.191 to .166). Similarly, Duval and Tweedie’s trim-and-fill test indicated no evidence of asymmetry in the funnel plot (see Supplement, Figure S2).

3.2.3. Independent laboratories.

Effect sizes from the 19 independent laboratories are given in the forest plot in Figure S3. The overall effect size was statistically significant (g = .278, p = .008, SE = .092, t = 3.00, CI = .083 to .472), with the prediction interval ranging from −.396 to .952. The Q statistic indicated significant evidence for heterogeneity in findings (χ² = 45.66, df = 18, p < .001). The I² was 66.9%, indicating that two-thirds of this heterogeneity can be attributed to true differences between studies.

There was no evidence of publication bias. Egger’s regression test was non-significant (p = .76, intercept = .071, SE = .228, t = 0.31, p = .89, CI = −.410 to .551). Similarly, Duval and Tweedie’s trim-and-fill test indicated no evidence of asymmetry in the funnel plot (see Figure S4).

3.3. Sub-group and meta-regression analyses.

Because the meta-analysis of independent analyses produces the largest number of independent effect sizes, we based sub-group (moderator) analyses on these 35 samples to maximize power.

Results of categorical (subgroup) and dimensional (meta-regression) moderators are shown in Table 1. Moderators were largely non-significant, with the exception of publication year (p = .03), with larger effect sizes for earlier studies. Study quality was marginally significant (p = .051) with larger effect sizes in less rigorous studies (study quality scores for each study are provided in the Supplement).

Table 1.

Results of moderator analyses based on categorical moderators (Table 1a) and meta-regression results for continuous covariates (Table 1b). The results of moderation are indicated by Χ² while effect sizes for the sub-groups are also provided. Significant or trending moderators are in bold.

TABLE 1a
	Χ2	p	Sub-group	Effect Size (g)	SE	p	CI
Moderator
Child vs Adult	.437	.509	Child	.193	.090	.047	.003 to .382
			Adult	.132	.043	.008	.040 to .223
Clinical vs Community	1.998	.158	Clinical	.105	.035	.007	.032 to .178
			Community	.275	.114	.029	.033 to .517
Added vitamins / minerals	.392	.531	Yes	.194	.061	.051	.000 to .252
			No	.126	.104	.090	−.036 to .424
Diagnosis	1.795	.180	Yes	.104	.038	.006	.029 to .179
			No	.244	.097	.012	.053 to .435
Externalizing / Internalizing	1.415	.493	Normal	.104	.038	.006	.029 to .179
			Internalizing	.181	.158	.251	−.128 to .490
			Externalizing	.261	.135	.053	−.004 to .526

TABLE 1b
Covariate	β	SE	t	p	CI
Age	−.002	.004	−.389	.699	−.010 to .007
Gender	.002	.002	.808	.425	−.003 to .006
Treatment duration	−.002	.007	−.335	.740	−.016 to .012
Omega-3 dose	−.117	.113	−1.033	.309	−.347 to .113
DHA/EPA ratio	.024	.018	1.308	.202	−.013 to .061
Fish consumption	−.0004	.003	−.126	.900	−.007 to .006
Study quality	−.048	.024	−2.030	.051	−.096 to .000
Publication year	−.017	.008	−2.264	.030	−.033 to −.002

CI = confidence interval. SE = standard error.

3.4. Sensitivity analyses

Sensitivity analyses were conducted to test robustness of findings. Findings from REML (restricted maximum likelihood) were re-run using maximum likelihood (ML), Empirical Bayes, Hedges, Hunter-Schmidt, DerSimonian-Laird, and Sidik-Jonkman estimators. Analyses were conducted either using no standard error adjustment or applying the Knapp-Hartung adjustment.

Results are given in Table 2 for the independent samples unit of analysis and in the Supplement (Tables S1 and S2) for independent studies and independent laboratories units of analysis. All estimators gave very similar effect sizes to the original (REML) results, with g ranging from .149 to .184, all of which are statistically significant (p < .007).

Table 2.

Sensitivity analyses running different estimators, either with or without no standard error adjustment or applying the Knapp-Hartung adjustment. Estimates when no adjustment is applied are z values (not t).

Estimator	Adjustment	g	p	SE	t	CI
REML	Yes	.162	.006	.055	2.943	.050 to .275
	No	.162	.001	.049	3.285	.066 to .259
ML	Yes	.159	.006	.054	2.924	.048 to .269
	No	.159	<.001	.047	3.350	.066 to .252
Empirical Bayes	Yes	.176	.005	.058	3.014	.057 to .294
	No	.176	.003	.058	3.014	.061 to .290
Hedges	Yes	.179	.005	.059	3.033	.059 to .299
	No	.179	.003	.061	2.937	.060 to .299
Hunter-Schmidt	Yes	.149	.007	.052	2.878	.044 to .255
	No	.149	<.001	.043	3.500	.066 to .233
Der-Simonian-Laird	Yes	.152	.007	.053	2.891	.045 to .259
	No	.152	<.001	.044	3.461	.066 to .238
Sidik-Jonkman	Yes	.184	.004	.060	3.063	.062 to .307
	No	.184	.005	.066	2.803	.055 to .313

REML = restricted maximum likelihood. ML = maximum likelihood.

Very similar results were also observed for independent studies with g ranging from .170 to .217 (p < .007), compared to the original g of .204. Similarly, for independent laboratories, g ranged from .270 to .281, compared to the original g of .278.

3.5. Proactive and reactive aggression

Reactive and proactive aggression were each assessed in 16 separate analyses (total of 32). There was no significant moderating effect of reactive versus proactive aggression (χ² = 0.28, df = 1, p = .59). Forest and funnel plots for the separate measures of reactive and proactive aggression are given in the Supplement in Figures S5 and S6.

3.5.1. Reactive aggression.

For reactive aggression the overall effect size was statistically significant (g = .167, p < .019, SE = .064, t = 2.62, CI = .031 to .303), with the prediction interval ranging from −.283 to .617. The Q statistic was significant (χ² = 57.33, df = 15, p < .001), indicating significant heterogeneity in findings.

There was no evidence of publication bias. Egger’s regression test was non-significant (p = .781, intercept = .048, SE = .170, t = 0.28, CI = −.316 to .413). Similarly, Duval and Tweedie’s trim-and-fill test indicated no evidence of asymmetry in the funnel plot (see Supplement, Figure S6).

3.5.2. Proactive aggression.

For proactive aggression the overall effect size was statistically significant (g = .124, p < .026, SE = .050, t = 2.47, CI = .017 to .230), with the prediction interval ranging from −.182 to .429. The Q statistic was significant (χ² = 34.07, df = 15, p = .003), indicating significant heterogeneity in findings.

There was no evidence of publication bias. Egger’s regression test was non-significant (p = .365, intercept = .122, SE = .130, t = 0.94, CI = −.157 to .401). Similarly, Duval and Tweedie’s trim-and-fill test indicated no evidence of asymmetry in the funnel plot (see Supplement, Figure S6).

3.5.3. Self-reports versus observer reports.

We examined whether different findings would emerge as a function of whether self-reports or observer reports of reactive-proactive aggression were employed.

For reactive aggression, there was a significant moderating effect of reporter, χ² = 9.026, df = 1, p = .003). Self-reports rendered a significant effect size (g = .265, p < .015, SE = .088, t = 2.98, CI = .064 to .465), but the effect for observer reports was non-significant (g = −.030, p = .421, SE = .034, t = −0.876, CI = −.118 to .112).

For proactive aggression, there was significant moderation of reporter, χ² = 4.731, df = 1, p = .03). Self-reports rendered a significant effect size (g = .196, p = .027, SE = .074, t = 2.647, CI = .029 to .364), but the effect for observer reports was non-significant (g = .020, p = .201, SE = .014, t = 1.474, CI = −.015 to .055).

4. Discussion

This meta-analysis examined whether omega-3 supplementation can result in a reduction in aggression using three main units of analysis. Analyses from 28 RCTs were based on 35 independent samples, 28 independent studies, and 19 independent laboratories, encompassing 3,918 individuals. Overall effect sizes for these three analyses were as follows: independent samples (g = .162); independent studies (g = .204) and independent laboratories (g = .278). Together these analyses averaged g = .215. There was no evidence of publication bias, and sensitivity analyses confirmed findings. While there was significant heterogeneity, moderator analyses were largely non-significant, indicating that beneficial effects are obtained across age, gender, recruitment sample, baseline diagnoses, treatment duration, and dosage. Because I² values averaged 72.1%, most of the variability between study effect sizes reflects true variance as opposed to sampling error. Effects also applied to reactive as well as proactive forms of aggression. Overall, results from these 35 independent samples obtained from RCTs provide substantial evidence that omega-3 supplementation can lead to modest short-term reductions in aggression.

4.1. Mechanisms-of-action.

Why would omega-3 supplementation be expected to reduce aggressive behavior? At one level it is well-known that there is a significant neurobiological basis to aggressive and violent behavior (Blair, 2022; Williams et al., 2018; Yang & Raine, 2009). At a mechanistic level, it is also known that omega-3 is a long-chain fatty acid that plays a critical role in brain structure and function. It plays multiple roles, making up approximately 35% of the cell membrane, enhancing neurite outgrowth, regulating both neurotransmitter functioning and gene expression, and being involved in neurogenesis and nerve cell signaling (McNamara & Carlson, 2006; Bazinet & Laye, 2014; Diaz, Mesa-Herrera, & Marin, 2021). Omega-3 also reduces inflammatory processes in the brain and plays a significant role in cerebral blood flow (McNamara, Asch, Lindquist, & Krikorian, 2018; von Schacky, 2021). Structural and functional brain imaging studies on humans have further documented that omega-3 can upregulate a variety of brain regions, with no evidence for any detrimental effect (McNamara et al. 2019). As such, given the undeniable fact that omega-3 is pervasively involved in multiple facets of neuronal biology, it is reasonable to believe that omega-3 supplementation could play a causal role in reducing aggression by upregulating brain mechanisms that may be dysfunctional in such individuals, given the assumption that there is, in part, a neurobiological basis to aggression.

4.2. Reactive-proactive aggression.

One novel aspect of the current meta-analysis is that it provides initial findings on the influence of omega-3 on two well-researched forms of aggression – reactive and proactive – albeit with the important caveat that all effect sizes came from just one laboratory. Analyses resulted in significant effect sizes of .167 for reactive aggression and .124 for proactive aggression, effect sizes that did not significantly differ (p = .59). These effect sizes are somewhat lower than those for the analysis as a whole, but are higher than the average effect size of .064 for this laboratory for all measures of aggression.

We have previously suggested that omega-3 supplementation may be more effective for reactive aggression given that this impulsive form of aggression has, broadly speaking, been more associated with brain dysfunction than instrumental, proactive aggression, particularly with respect to prefrontal dysfunction (Blair, 2010; Raine et al., 1998; Siep et al., 2019). In this context it is notable that concentrations of DHA vary throughout the brain and are at their highest in the prefrontal cortex (Laye, Nadjar, Joffre, & Bazinet, 2018), an area critical for impulse control and emotion regulation. Higher levels of omega-3 are also associated with increased functional connectivity in the frontal pole and anterior cingulate, areas which in part subserve executive functions (Talukdar, Zannroziewicz, Zwilling, & Barbey, 2019). Omega-3 supplementation in turn has also been shown to enhance executive functions (McNamara, Asch, Lindquist, & Krikorian, 2018). Given that reactive-impulsive aggression has been associated with reduced prefrontal glucose metabolism (Raine et al., 1998), as well as poor executive functions (Thomson & Centifanti, 2018) and also reduced connectivity between the prefrontal cortex and amygdala (Romero-Martinez et al., 2019), prefrontal upregulation is a viable explanation for why omega-3 reduces impulsive-aggressive behavior.

At the same time, the stronger omega-3 effect size for reactive aggression was only marginally (and non-significantly) higher than that for proactive aggression, indicating that omega-3 could potentially be effective in treating more planned, cold-blooded, regulatory forms of aggressive behavior. Future studies could evaluate whether omega-3 supplementation enhances prefrontal functioning as assessed by either neurocognitive or brain imaging measures, and by additionally assessing whether such prefrontal upregulation mediates any effect of omega-3 supplementation in reducing antisocial behavior.

4.3. Moderation effects.

With some exceptions, significant moderation of treatment effects was not observed. However, effect sizes were lower in more recent years of publication, and there was a trend for lower effect sizes in better quality studies, suggesting that future studies may obtain lower effect sizes. However, these effects were rendered non-significant (p = .15, p = .36 respectively) after excluding one laboratory (Raine), and caution is also warranted given that we tested for 15 moderators, running the risk of Type 1 error.

The statistical non-significance of other moderators may potentially have some clinical significance. Omega-3 appears to be effective in reducing aggression in children as well as adults, in females as well as males, in clinical samples as well as community samples, in the general population as well as externalizing populations. The fact that it has some efficacy in clinical populations and in populations receiving a clinical diagnosis is also noteworthy because these populations are receiving interventions for their disorders, and as such omega-3 appears to be a useful therapeutic adjunct for reducing aggression. We caution however that there is the potential for Type 2 error, and that non-significance of moderators could be in part a product of power to detect effects. As such there still remains the possibility of subtle but significant sub-group differences.

The most salient moderator effect was observed for self-reports vs observer reports in separate analysis of those studies of proactive and reactive forms of aggression. Moderation was significant for both reactive and proactive aggression, with self-reports yielding significant effects (g = .265 and .196 respectively), whereas observer reports did not (g = −.03 and .02 respectively). The stronger effects for self-reports may be a function of the fact that motivation for perpetrating aggression (e.g. for revenge in the case of reactive aggression; to acquire resources in the case of proactive aggression) is an important distinguishing feature of these two forms of aggression. Those who self-report may have better awareness of the reason why they perpetrate aggression than observers who, as an external witness of the aggressive act, may be less able to deconstruct the motivation for such acts. In this sense, self-report questionnaires may have somewhat better validity compared to observer reports which are more predicated on observing the act rather than understanding the underlying motivation. One important qualification to these findings is that they are based on just one instrument, the Reactive-Proactive Aggression Questionnaire (Raine et al., 2006) and further research is needed on other instruments to further evaluate the moderating effect of reporter.

4.4. Recommendations for future research

There are a number of areas in which future research could provide important additional information. First and foremost, past studies have not examined mechanisms of action by which omega-3 lowers aggression. Given the wide-ranging influence of omega-3 on the brain, there are clearly many possible mediators that could be explored in future RCTs, including neurocognitive functioning, structural and functional brain imaging, neurotransmitter functioning, and inflammatory processes.

Second, this meta-analysis could only examine immediate post-treatment effects because to date only one laboratory has examined longer-term effects after treatment is terminated. While there is value in knowing whether omega-3 reduces aggression in the short-term, the next step will be to evaluate whether omega-3 can reduce aggression in the longer-term. Initial studies do suggest evidence for longer-term efficacy, both for general aggression (Raine et al., 2016) and also more specifically reactive-impulsive aggression (Raine et al., 2019; Raine et al., 2020).

Third, a challenging issue to researchers is the lack of attention to bioavailability of omega-3, a potential source of noise in RCTs. As one example, because absorption of DHA and EPA requires digestion of fat, taking omega-3 supplements on an empty stomach or during low-fat meals such as breakfast has been argued to reduce bioavailability, with individual differences in such bioavailability argued to be large (von Schacky, 2021). Omega-3 uptake is influenced by many factors. Just as it influences the functioning of the gut microbiota, so too do intestinal microbes influence the absorption and bioavailability of omega-3 (Fu et al., 2021). Fatty acid entry into the brain is a very complex process and is influenced by many factors (Bazinet & Laye, 2014). Essentially, although individuals may consume omega-3, the extent to which it has sufficient bioavailability in the brain to have a functional impact on behavior is variable. The maximum dose given so far in trials dealing with aggression is 2.4g, and while we did not observe a significant relationship between dose and effect size within the range employed in the studies we examined, future studies using larger doses may potentially show greater efficacy.

Fourth, heterogeneity in response to omega-3 and inconsistency of effects may also be a function of genetic variation. There is increasing interest in nutrigenetics, the investigation of the effect of genetic makeup on how individuals respond to nutritional intake (Merched & Chan, 2013). The FADS gene cluster is one variant known to significantly impact omega-3 biosynthesis, and it has been argued to be a major factor in the heterogeneity of clinical trials of omega-3 and heart disease (Fernandez et al., 2021). No omega-3 RCT on aggression has so far examined gene variants as moderators of treatment outcome, and clearly this is an avenue worthy of pursuing in the future.

Fifth, there is a clear need for future studies to investigate different forms of aggressive behavior, as this is a major gap in this literature. While the current meta-analysis does provide an initial foray into this field with respect to reactive-proactive aggression, the participation of more laboratories is critically needed, and other sub-types of aggression could be examined. Importantly, do self-report measures of proactive-reactive aggression provide stronger evidence for efficacy compared to observer reports?

Finally, no trials have compared DHA-only to EPA-only supplementation. DHA is the predominant component of omega-3 in the brain as EPA is either not detectable in neural tissue or exists in a much lower order of magnitude (Bazinet & Laye, 2014). Furthermore, the emphasis in the literature on the beneficial effects of omega-3 on brain functioning leans heavily on DHA, and not EPA (Bazinet & Laye, 2014; McNamara et al., 2018). Because a higher ratio of DHA relative to EPA in the current meta-analysis was non-significantly associated with a larger effect size, future studies which more directly compare DHA-only with EPA-only administration could help further resolve this issue, and potentially potentiate future treatment approaches.

4.5. Limitations and strengths.

Limitations need to be recognized. First, this review was limited to aggression, and findings cannot be generalized either to externalizing behavior per se, or to aggression-related constructs such as irritability, hostility, and anger which we excluded. Second, we only included studies written in English and as such we may have missed international publications in non-English journals. Third, while we searched dissertation abstracts and included findings from one currently unpublished study that we knew of, and while we searched reference lists for unpublished findings, this does not resolve the potential issue of the existence of unpublished non-significant findings. Fourth, we were unable to explore ethnicity as a moderator as only two laboratories reported this variable. Fifth, the multiple moderator analyses run the risk of Type 1 error, and any significant effects must be treated with caution. Sixth, while any disagreements on study inclusion / exclusion were resolved by consensus, inter-rater reliability was not quantified. These limitations should be set against strengths, which include the exclusive focus on aggression and RCTs (the gold-standard methodology for establishing causality), a comprehensive literature search, sensitivity analyses to confirm findings, evaluation of publication bias and moderators, and new findings on proactive-reactive aggression.

5. Conclusions

Results of this study show that omega-3 supplementation significantly reduces aggressive behavior in the short-term, albeit at a modest level. The degree of reduction is modest, with an average effect size of .22. This treatment effect does however apply broadly across a variety of different populations, and cuts across age and gender. Given the enormous economic and psychological cost of aggression and violence in society, even small effects sizes need to be taken seriously.

Regarding clinical implications, based on these findings, our considered opinion is that there is now sufficient evidence to begin to implement omega-3 supplementation to reduce aggression in children and adults at a modest level - irrespective of whether the setting is the community, the clinic, or the criminal justice system. At the very least, we would argue that omega-3 supplementation should be considered as an adjunct to other interventions, whether they be psychological (e.g. CBT) or pharmacological (e.g. risperidone) in nature, and that caregivers are informed of the potential benefit of omega-3 supplementation.

We base this broad conclusion on the following arguments. First and foremost, evidence has now accumulated over 28 years from 28 RCTs documenting some efficacy. Second, there is additional evidence that omega-3 supplementation can help reduces psychopathology that is comorbid with aggression, including depression (Liao et al., 2019), alcohol use (Pauluci et al., 2022), and more controversially schizophrenia-spectrum disorders (Amminger et al., 2010; Raine et al., 2024). Omega-3 supplementation may additionally therefore reduce comorbid psychopathological risk factors associated with aggression. Third, there are well-documented beneficial effects of omega-3 supplementation on other health conditions, including lowering triglyceride levels (Musa-Veloso et al., 2010), coronary heart disease (Shen et al., 2022), rheumatoid arthritis (Gioxari, Kaliora, Marantidou, & Panagiotakos, 2018) and hypertension (Guo, Li, Li, & Li, 2019), pointing to additional benefits of supplementation over and above aggression reductions. Fourth, omega-3 is safe to administer and side effects if any are both mild and minimal. Both the FDA and the European Food Safety Authority concluded that long-term consumption of EPA and DHA supplements at a dosage level of approximately 5g/day appears to be safe – a dose more than twice that of the highest use in RCTs on aggression. Fifth and relatedly, to our knowledge there are no known disadvantages on health outcome in taking omega-3 at non-excessive doses. Sixth, from a practical standpoint this intervention is low-cost and very easy-to implement. We anticipate that a cost-benefit analysis would conclude that benefits outweigh both financial costs and risks.

We recognize that some may reasonably disagree with this conclusion. Such objections however bear the heavy burden of specifying how many more RCTs, over how many more decades, showing what specific effect size, at what level of safety, will it take for any intervention to begin to be deemed suitable for implementation to reduce aggression. Given the additional psychological and physical benefits of omega-3 supplementation and ease of implementation, we believe the time has come both to execute omega-3 supplementation in practice and also to continue to scientifically investigating its longer-term efficacy.

Highlights.

• Omega-3 supplementation has been argued to reduce aggression

• A meta-analysis was conducted on 28 RCTs involving 3,918 participants

• Omega-3 modestly but significantly reduces aggression in children and adults

• Effects are found for different forms of aggression

• Omega-3 is a safe, cheap, and healthy intervention to reduce aggression