Vitamin D at 5,000 IU reduced depression symptoms most in new dose-response analysis

A dose-response meta-analysis of 15 randomized controlled trials, published in Frontiers in Nutrition in March 2026, found that vitamin D supplementation reduced depressive symptoms by a standardized mean difference of −0.98 (95% CI −1.28 to −0.68, p < 0.001) compared to placebo. Higher daily doses produced larger reductions: the dose-response curve, fitted by the authors using restricted cubic splines, reached its maximum effect at 5,000 IU per day (SMD = −1.44; 95% CI −1.81 to −1.06). The analysis was led by Hsuan-Hsien Liu of Linkou Chang Gung Memorial Hospital in Taiwan and drew on 962 participants across the 15 included trials.

It is the second dose-response meta-analysis on this question in two years. A larger 2024 analysis by Ghaemi and colleagues (31 RCTs, 24,189 participants) also found that vitamin D cut depression symptoms in a dose-dependent manner, with most of the benefit captured by roughly 5,000 IU per day. The Liu analysis is smaller but adds a mechanistic layer the Ghaemi paper did not have: the authors measured significant drops in both parathyroid hormone (mean difference −4.19 pg/mL; 95% CI −8.18 to −0.2) and tumour necrosis factor-alpha (mean difference −0.3 pg/mL; 95% CI −0.44 to −0.16) in the vitamin D groups. Neither body weight, BMI, nor interleukin-6 changed.

An SMD of −0.98 is large by the standards of psychiatric intervention research. It approaches what you would expect from a moderate-dose antidepressant in an RCT. But that comparison needs a caveat: heterogeneity across the 15 trials was high (I² = 79%), which means the individual studies did not all agree in direction or magnitude. The pooled estimate carries real uncertainty.

What the meta-analysis actually examined

Liu and colleagues searched PubMed, EMBASE, and the Cochrane Library from database inception to June 2024. They included RCTs that enrolled adults with depressive disorders diagnosed by DSM-IV, DSM-5, or ICD-10 criteria, or adults with clinically significant symptoms on validated scales like the Hamilton Depression Rating Scale or Beck Depression Inventory. Pregnant participants, patients under 18, and depression secondary to another medical condition were excluded.

The 15 qualifying trials enrolled a combined 962 participants. Individual samples ranged from roughly 26 to 106 people. Vitamin D was given at doses spanning a few hundred IU per day to single injections of 150,000 or 300,000 IU; control groups received placebo or no treatment. Most trials ran 8 to 24 weeks.

The primary outcome was change in depressive symptom scores. The pooled SMD of −0.98 translates to about a one-standard-deviation improvement in the treatment group relative to controls, a shift that would matter clinically for a patient starting at moderate depression. On the secondary outcomes, serum PTH and TNFα both fell in the vitamin D groups, while body weight, BMI, and interleukin-6 did not budge.

The dose-response curve

The authors used restricted cubic splines to plot the relationship between daily vitamin D intake and depression symptom scores. The curve rose steeply from the lowest doses through about 2,000 IU per day, climbed more gradually, and reached its maximum at 5,000 IU per day. Beyond that, the confidence intervals widened enough to make the curve uninterpretable. Two studies that used single megadoses of 150,000 IU or more were excluded from this analysis because their pharmacokinetics differ fundamentally from daily oral dosing.

“Upon reaching a daily dose of 5,000 IU of vitamin D, the SMD value reached its lowest point,” the authors wrote. The Ghaemi 2024 meta-analysis found that doses up to 8,000 IU per day continued to produce incremental benefit, though most of the effect was already present at 5,000 IU. The two analyses converge on the same dose neighbourhood despite different scopes and inclusion criteria.

That dosage sits in tension with safety guidance. The Tolerable Upper Intake Level for vitamin D in adults, set by the US National Institutes of Health, is 4,000 IU per day. Prolonged intake at or above that level has been tied to increased fall risk in older adults. The Liu authors address this directly: 5,000 IU “may be acceptable under appropriate monitoring,” they write, but “safety concerns remain for certain subpopulations.”

Who benefited most

Two subgroup findings stand out. Female participants showed a larger effect than the pooled estimate (SMD −1.26; 95% CI −1.5 to −1.01; p < 0.001). Participants with obesity showed the largest benefit of any subgroup (SMD −1.83; 95% CI −2.4 to −1.26; p < 0.001).

The obesity result makes mechanistic sense. Vitamin D is fat-soluble and gets sequestered in adipose tissue, so people with higher body fat percentages tend to have lower circulating 25-hydroxyvitamin D at any given oral dose. The authors suggest this may mean obese patients with depression are both more likely to be deficient and more responsive to supplementation, but that hypothesis needs testing in a dedicated trial, not a post-hoc subgroup analysis.

One other result warrants mention. Self-rated depression scales showed significantly greater improvement than clinician-rated scales (p = 0.01 for the difference). The authors attribute this gap to “unrecognized biases in the blinding methods of some of the included studies.” When patients suspect they are receiving a treatment rather than a placebo, their self-reports can inflate the apparent effect.

What the mechanism might be

The TNFα drop is the most interesting secondary finding in the Liu analysis. It comes from two of the 15 trials, so it is thin. The authors themselves call it “preliminary and hypothesis-generating rather than confirmatory.” But the direction is worth paying attention to. TNFα is a pro-inflammatory cytokine, and the neuroinflammation hypothesis of depression proposes that chronic low-grade inflammation drives depressive symptoms in some patients. If that is right, then an anti-inflammatory effect from vitamin D would line up with an antidepressant effect. The biomarker data is too sparse to confirm this, but it gives future trials something concrete to measure.

The PTH side is better understood. Vitamin D regulates calcium absorption; supplement it, PTH drops, calcium homeostasis improves. The meta-analysis found a mean PTH reduction of 4.19 pg/mL in the vitamin D groups. What that means for mood is less clear. There are plausible downstream links to neurotransmitter synthesis and neuronal excitability, but the chain of evidence has gaps. The serotonin mechanism the authors mention, that vitamin D “may enhance serotonin synthesis and availability in the brain,” comes from preclinical work and hasn’t been tested in humans at the doses used in these depression trials.

So you have two candidate pathways, one inflammatory and one endocrine. Neither is settled. The value of the Liu paper is that it measured both and found signals worth following up. Whether they operate independently or interact is an open question.

The limitations

I² of 79% is high. That is the number that should temper enthusiasm about the pooled effect size. The 15 trials differed on study design, dosing, baseline vitamin D status, depression subtypes, and the instruments used to measure symptoms. Major depressive disorder, postpartum depression, and seasonal affective disorder got combined into one analysis. Those conditions do not share the same biology, and a statistical method cannot fully unbake that cake.

Then there is the risk-of-bias problem. One trial rated low risk on the Cochrane RoB 2.0 tool. Six were high risk, eight moderate. A sensitivity analysis that dropped the high-risk studies still found an effect, which is somewhat reassuring. But you are building a case on mostly moderate-to-low quality evidence, and that limits how confident anyone should be.

The biomarker data is even thinner. The PTH and TNFα findings each draw from two trials. Drawing mechanistic conclusions from four data points is closer to hypothesis generation than confirmation. The trials also did not consistently report what other treatments patients were on. Without that, you cannot tell whether vitamin D adds to standard antidepressants or partly compensates for patients who are undertreated.

Where this leaves things

The Liu 2026 paper strengthens the evidence that vitamin D moves the needle on depression symptoms. It does not tell you who to give it to, at what dose, or for how long. Two independent dose-response meta-analyses now point to roughly 5,000 IU per day as where the benefit maxes out. Both are working from the same pool of RCTs, and that pool is small, heterogeneous, and uneven in quality.

A clinician looking at a patient with depression and a documented vitamin D deficiency now has a second reason to supplement. The mood benefit has enough evidence behind it to act on. For someone with normal vitamin D levels, the case is weaker and the safety trade-off is real, particularly above 4,000 IU per day.

What the field actually needs is a trial that has not been done: large, multicenter, baseline 25(OH)D measured, randomized to 5,000 IU per day versus placebo, depressive symptoms and inflammatory biomarkers tracked over at least 24 weeks, adverse events reported systematically. Until that trial exists, the meta-analyses have taken the existing evidence about as far as it can go.

References

1. Liu HH, Liu TH, Liu CY, et al. Efficacy of vitamin D supplementation in patients diagnosed with depression: a dose-response meta-analysis of randomized controlled trials. Frontiers in Nutrition 13. 2026. https://doi.org/10.3389/fnut.2026.1772451
2. Ghaemi S, Zeraattalab-Motlagh S, Jayedi A, et al. The effect of vitamin D supplementation on depression: a systematic review and dose-response meta-analysis of randomized controlled trials. Psychological Medicine 54(15):1-13. 2024. https://doi.org/10.1017/S0033291724002559