The no supplementation group was associated with 14.3 times the risk of death compared to those who regularly supplemented with vitamin D

Sunday, January 31st, 2021

Peter Attia wouldn’t claim that vitamin D is a magic pill against infection, given the evidence we have:

While it is easy to fool ourselves, quasi-experimental studies like this one, for example, shouldn’t be overlooked completely. The study reported the 14 day mortality of 77 elderly (mean age 88 years) hospitalized patients comparing those that regularly supplemented vitamin D in the preceding 12 months and those that started supplementing after COVID-19 diagnosis. Both groups were compared to a third group that didn’t supplement with vitamin D at all. Long-time supplementers had a 93.1% survival rate compared to 81.7% survival rate in the more recent supplementers, and there was a 68.7% survival rate in the group that didn’t take vitamin D. Given the hazard ratio 0.07 in the first group, the study reported a 93% reduced associated risk for those that regularly supplemented vitamin D. In other words, the no supplementation group was associated with 14.3 times the risk of death compared to those who regularly supplemented with vitamin D.

Randomized controlled trials (RCTs) do, nonetheless, continue to be the gold standard. There is one pilot RCT that looked at the rate of ICU admission and death for 76 people with and without in-hospital vitamin D supplementation. It reported that 98% of the treatment group did not get admitted to the ICU compared to 50% admission in the untreated group, of which 15% (2 people) later died. After adjusting for confounding variables, patients treated with vitamin D had 0.03 times the risk for ICU admission compared to non-treatment. Put another way, patients not treated with vitamin D had 33.3 times the risk of ICU admission compared to patients treated with vitamin D. And if you want further commentary on the importance of RCTs to distinguish signal from noise on issues like this, my conversation with Vinay Prasad gets to the heart of the matter.


  1. Gwern says:

    An HR of >14 or >33 is all you need to know that it’s massively confounded. Has anything at all been established to work on coronavirus that well? That magnitude of silver bullets almost never actually exists in the real world. (For comparison, smoking and lung cancer has a HR <30. I'm not sure you would get a HR of 33 for vitamin D and even rickets!) Sometimes people argue that "these correlates are *so* large that they must reflect at least *some* causation", but they are a huge red flag: if the hidden and unmeasurable biases and confounding are so bad that, despite your best efforts, they can inflate the HR by 10+ from what is plausible, why can't they happen to inflate it by another 4 or so? (Similar to Feynman's observation about the Space Shuttle o-rings.)

    Incidentally, Attia does not mention the later larger vitamin D RCT which turns in HRs closer to 1.

Leave a Reply