Vitamin D and Depression: Systematic Review and Bayesian Meta-analysis of Mendelian Randomization Evidence

Background: Observational links between low 25(OH)D and depression may reflect confounding or reverse causation.
Methods: We systematically reviewed Mendelian-randomization (MR) studies through November 2024, appraising instrument strength, pleiotropy safeguards, bidirectionality, and dose–response/nonlinearity; certainty was graded using an MR-adapted GRADE framework. Because of between-study heterogeneity and scale differences, we did not perform conventional frequentist pooling. Instead, we conducted a de novo Bayesian random-effects meta-analysis of study-level MR estimates (harmonized to log-odds per 10 nmol/L where possible) and synthesized available nonlinear contrasts.
Results: Sixteen studies met inclusion criteria. Most linear MR estimates were null or near-null. Method-dependent signals appeared under correlated-instrument models and in low-quantile/nonlinear analyses. Bidirectional MR more often supported the reverse direction (depression → lower 25(OH)D). Certainty for nonlinearity was downgraded owing to missing stratum-level instrument strength and incomplete multiplicity control. In the Bayesian meta-analysis, the pooled average effect suggested a small protective association of higher 25(OH)D with depression risk (OR ≈ 0.988 per 10 nmol/L; posterior P [μ < 0] = 0.997). Available nonlinear syntheses indicated stronger benefit at low 25(OH)D with flattening beyond ~50–60 nmol/L.
Conclusions: Current MR evidence does not support a large population-wide linear effect of 25(OH)D on depression. Any average effect appears small; signals confined to severely deficient strata should be considered hypothesis-generating and require replication across methods, datasets, and ancestries. Universal supplementation for depression prevention is not supported; testing and correction of marked deficiency remain reasonable.
Keywords: Vitamin D; Depression; Mendelian Randomization; Systematic Review; Threshold Effect; Causal Inference