Table 5.

Meta-regression to identify factors that account for significant heterogeneity in the overall standardized mean difference (SMD) between PUFA treatment and placebo upon depression in patients with major depressive disorder or bipolar disorder.

Factor in Model	Factor category	Between-study variance (τ2)	SMD for category	P(difference = 0)
No factor		0.39	0.91	NA
Baseline mean HDRS score	<17	0.31	0.48	0.291
	>17		1.01
Depression type	Bipolar	0.48	0.80	0.761
	MDD		0.98
PUFA species utilised	Predominantly DHA	0.18	0.06	0.009
	Predominantly EPA		1.18

Meta-regression models the effect of study variables on the between-study variance (τ²) in a random effects meta-analysis. To conserve degrees of freedom, each factor was tested independently. A smaller τ² compared to the no-factor base model indicates that the factor explains more between-study variance. A meta-regression model also allows us to calculate a Wald test for the difference in the SMD for each category of the explanatory factor. A p-value less than 0.05 indicates that there is a statistically significant difference between the SMD for each category. All trials listed in Table 4 were included in the analysis of PUFA species utilised and depression type. For those studies using the HDRS, we used a baseline mean depression score of 17 as the cut-off value for moderate vs. mild depression [89]. The analysis of baseline depression scores excluded two trials – the trial conducted by Keck and colleagues [81] who did not report baseline scores, and the trial by Nemets and colleagues [71] because it used the Children's Depression Rating Scale and there is no agreement on what score constitutes moderate depression. NA – not applicable.