Psychosomatic Medicine previously published an erratum to a meta-analysis by Firth et al. () to correct an estimated effect size for one of the 16 studies included in the meta-analysis. The erratum corrected the effect size for the included study by Wardle et al. () and provided updated effect sizes of the meta-analysis ().
After publication of this erratum, it was noticed that a mislabelling of subgroup information had resulted in slight imprecision of the main effect estimates used for three other studies (4–6; k = 3; n = 583 out of k = 16, n = 45,826), and various changes to subgroup analyses were required. After correcting these oversights, meta-analyses were rerun, figures reproduced, and the tables were remade (see Supplemental Digital Content, https://links.lww.com/PSYMED/A714). The overall findings of the meta-analyses as published in 2019 remain unchanged (), but the Results paragraph of the original article’s Abstract should read as follows (changed text is in bold):
“Results: Sixteen eligible randomized controlled trials (published in English) with outcome data for 45,826 participants were included; the majority of which examined samples with nonclinical depression (n = 15 studies). Nonetheless, dietary interventions significantly reduced depressive symptoms (g = 0.162, 95% CI = 0.055 to 0.269, p = 0.003). Similar effects were observed among high-quality trials (g = 0.171, 95% C.I.=0.057 to 0.286, p=0.003) and when compared with both inactive (g = 0.114, 95% C.I.=0.008 to 0.219, p=0.035) and active controls (g = 0.224, 95% C.I.= 0.052 to 0.397, p = 0.011). No effect of dietary interventions was observed for anxiety (k = 11, n = 2270, g = 0.085, 95% C.I. = -0.031 to 0.202, p=0.151). Studies with female samples observed significantly greater benefits from dietary interventions, for symptoms of both depression and anxiety.”
Regarding details of the secondary analyses, the following should be noted: While indication for publication bias remained (Egger’s regression intercept = 0.99, p = .018), the random effects trim-and-fill analysis now found no missing studies, producing the same results as the new main analysis (i.e., g = 0.162, p = .003, as presented above). Whereas pooled effects in the inactive control subgroup were previously reported to be larger than the active control subgroup, the inactive subgroup pooled effects are now reduced, and smaller than the active subgroup. Additionally, whereas the “reducing % fat intake” subgroup was previously significant for depressive symptoms, and the “nutritional professional” subgroup was previously significant for anxiety symptoms, these two subgroups are now non-significant. Further details on the revised analyses are presented in the Supplemental Digital Content, https://links.lww.com/PSYMED/A714.