HN Debrief

Vitamin D3 During Pregnancy and Cognitive Performance at 10 Years

  • Public Health
  • Science
  • Statistics
  • Nutrition

The paper is a post hoc secondary analysis of a Danish randomized controlled trial that originally tested prenatal vitamin D3 for asthma-related outcomes, not cognition. It followed roughly 500 children to age 10 and reported small positive associations between higher-dose maternal supplementation and a few cognitive measures, mostly in memory. After correction for multiple testing, the abstract says visual memory and verbal memory remained significant while flexibility or set shift did not. That framing drove most of the reaction.

Treat this as a hypothesis-generating result, not a basis for product claims, health guidance, or broad neurodevelopment conclusions. If you work with research-heavy health topics, watch for how many outcomes were tested, how multiple-comparison correction was done, and whether the study was actually designed for the endpoint now being promoted.

Discussion mood

Mostly skeptical and a bit irritated. The main reasons were that the paper reuses an asthma trial for a different endpoint, tests many cognitive measures, reports only small effects, and still invites a much stronger takeaway than the data justify in a vitamin D area that commenters already see as prone to hype.

Key insights

  1. 01

    Vitamin D findings live in a publication-bias swamp

    Vitamin D gets studied against an enormous range of outcomes because it plausibly touches many body systems. That makes any single modest positive signal hard to trust on its own, since many null follow-ups or failed exploratory analyses may never get published, and this paper does not stand out as a field-settling result in that larger literature.

    Do not evaluate a vitamin D paper in isolation. Look for systematic reviews, preregistered replication, and whether the endpoint was primary rather than one more positive result surfacing from a crowded field.

      Attribution:
    • medymed #1
    • scythe #1
  2. 02

    Geography matters through baseline vitamin D, not nationality

    Location is relevant here because sunlight, fortification, and time outdoors change vitamin D biology before supplementation even starts. Several commenters pointed out that the study did measure maternal baseline vitamin D and adjusted for it, while others added that northern countries can offset low winter sun through fortification and that indoor sunlight through glass does not make vitamin D because UVB is filtered out.

    If you want to apply this result to another population, match on baseline vitamin D status and local fortification patterns before thinking about dose. Latitude alone is too crude and national origin is the wrong variable.

      Attribution:
    • xwowsersx #1
    • cientifico #1
    • doikor #1
    • wrsh07 #1
    • HexPhantom #1
    • roncesvalles #1
  3. 03

    Randomization does not rescue a weak secondary endpoint

    Randomization handles confounders like socioeconomic status only for outcomes the trial was properly designed and followed to assess. Once a study pivots years later to a secondary endpoint with limited detail on selection, attrition, and measurement, the fact that the original trial was randomized no longer gives the new claim the same level of credibility as a clean primary analysis.

    When you read follow-up analyses of old trials, separate the strength of the original randomization from the strength of the new endpoint. Demand attrition details and a prespecified analysis plan before treating the result like causal evidence.

      Attribution:
    • bootsmann #1
    • p-o #1
  4. 04

    The useful reading is neurodevelopment, not IQ

    The least misleading interpretation is that prenatal vitamin D may affect some narrow neurodevelopmental outcomes, not that it makes children broadly smarter. Commenters highlighted that the effect, if real, is modest and local to a couple of memory measures, which is a very different claim from a general improvement in cognition or intelligence.

    Avoid broad labels like 'smarter kids' in any downstream use of this study. If you follow the space, watch for targeted replication on specific memory outcomes rather than omnibus cognition claims.

      Attribution:
    • HexPhantom #1 #2

Against the grain

  1. 01

    Exploratory analyses still have value

    Looking across many outcomes in an existing trial is not automatically misconduct. Used honestly, it is a way to kill weak hypotheses and surface one or two candidates worth testing in a study that is actually built for them.

    Do not throw out every post hoc analysis. Classify it correctly and use it to prioritize the next experiment, not to justify practice changes.

      Attribution:
    • marcosdumay #1
  2. 02

    Multiple-testing correction may not be as damning

    The criticism that any correction should wipe out the result depends on how independent the cognitive measures really are. If several scores are correlated for the same child, a blunt whole-battery correction like Bonferroni can be too harsh, and the paper's false discovery rate approach may be more defensible than critics first assumed.

    When assessing 'p-hacking' claims, inspect the dependence structure among outcomes and the exact correction method. Bad correction can inflate findings, but overly harsh correction can also hide a real signal.

      Attribution:
    • senkora #1
    • tdb7893 #1
  3. 03

    Homogeneous cohorts are not automatically flawed

    A single-country cohort does limit generalization, but it does not invalidate the internal comparison if treatment and control groups start from the same place. For a biology question tied to vitamin D levels, a clean estimate within one population can still be useful, and later studies can test whether the same dose-response appears elsewhere.

    Separate internal validity from external validity. A narrow cohort can still answer a narrow question well, but you should resist exporting the dose or effect size without replication in other settings.

      Attribution:
    • cenamus #1
    • nine_k #1
    • dheera #1

In plain english

false discovery rate
A statistical method that tries to limit how many reported positive findings are actually false positives when many tests are run.
multiple testing
Running many statistical tests in one study, which raises the chance of finding a seemingly significant result just by luck.
post hoc
An analysis designed after the original study data were collected, rather than planned in advance as a primary test.
secondary analysis
A new analysis of data from a study that was originally run for a different main question.
UVB
Ultraviolet B, the part of sunlight that helps human skin produce vitamin D.

Reference links

Statistical skepticism and explainers

Vitamin D, sun exposure, and human adaptation

Tangential biology references