Hacker News Reader

Comment by seizethecheese | Hacker News Reader

> For the primary comparison of any vitamin D versus placebo, the intervention did not statistically significantly affect overall ARI risk (OR 0·94 [95% CI 0·88–1·00], p=0·057; 40 studies; 61 589 participants; I2=26·4%).

Okay, so the 95% confidence interval is reduction from 0% to 12%, but since this barely is on the other side of p=0.05 the conclusion is fully that the claim is not supported.

View on HN

The more interesting finding is that people with low baseline Vitamin D levels (<25 nmol/L) did not show any improvement. Odds ratio of 0.98 (0.80–1.20).

A lot of previous Vitamin D research was dismissed with the claim that Vitamin D would be more effective if you only looked at people with low baseline Vitamin D status. This paper does not show that to be true.

The proper claim is that raising people from deficient (<20ng/ml) to a sufficient level (>30ng/ml ideally) should improve outcomes. Sometimes looking at a subset of those who started deficient is good enough to find such an effect, but sometimes it isn't, such as if the dose used was not high enough to raise people up enough, particularly common if the subjects weights were high which requires higher intakes to move the needle a lot, so other subset analysis look at norm-weight subjects. Proper study guidelines were published in the mid 2010s and meta-analyses like this one that aggregate many studies that don't conform to these guidelines are just as invalid as individual studies that don't.

For detailed treatment of this see this thread on X: https://x.com/KarlPfleger/status/1732514710715514883

if we're teetering on significance seems like a subgroup analysis is worth considering

Reply on news.ycombinator.com