I am sick of people rejecting good evidence about vitamin D because they are confused about the bad evidence and can't be bothered to investigate, so I am going to explain it.
Let's look at this like a 19th century physician who woke up from a coma this morning to trawl the public internet for info (I helped), knowing about evolution and bodies and counting and skepticism but not about "metastudies" or "scientific consensus" or "USDA guidelines." How much vitamin D do we need?
Hunter-gatherers in the environment where most of our evolution happened might have been outside all day shirtless. On average the sun's halfway from peak, so that might be equivalent to 8 hours of peak sunlight at the equator. If you're Fitzpatrick Type III like me, your skin adapted to absorb sunlight somewhat more efficiently since your ancestors were exposed to less over an evolutionarily relevant period, so let's cut that in half and say your body evolved for the equivalent of 4 hours of peak sunlight a day. An experimental study says:
Peak ultraviolet B irradiation for vitamin D synthesis occurs around 12 pm Eastern Standard Time (EST). In Boston, MA, from April to October at 12 pm EST an individual with type III skin, with 25.5% of the body surface area exposed, would need to spend 3 to 8 minutes in the sun to synthesize 400 IU of vitamin D. It is difficult to synthesize vitamin D during the winter in Boston, MA. For all study months in Miami, FL, an individual with type III skin would need to spend 3 to 6 minutes at 12 pm EST to synthesize 400 IU. Vitamin D synthesis occurs faster in individuals with lighter Fitzpatrick skin types.
There's another one that's paywalled if someone wants to pay or steal and check.
400 IU per 5 minutes is 8000 IU per hour is 32,000 IU (800 micrograms) per day by this estimate.
When deciding how much is actually appropriate to supplement, we need to take into account diminishing returns; eventually the sunlight starts producing other secondary metabolites which are also good for us, so a 16,000 IU supplement is lower-quality than sunlight but similar in the effective dosage of the most important chemical our evolutionary ancestors' bodies would have made from sunlight; in practice I wouldn't take more than that.
If someone wants to pay some people to have their blood drawn and analyzed before and after a full sunny day at the beach, great, but I can't be bothered. It would also be nice to see a study done on black people.
We know we get MUCH LESS vitamin D than this, and we already know what kinds of effects to look for. If you get no sunlight or supplemental vitamin D, your bones break down. If you live in an area with very little sun you're much more likely to be depressed and get sick. We know mechanistically that vitamin D plays a role in immune function. We observe that if you take a huge amount of supplemental vitamin D (>100,000 IU/day) you end up with too much calcium in your blood.
Let's use something where we have better intuitions about effective doses: exercise. Vitamin D supplementation equivalent to 4 hours of peak sunlight is like 4 hours of brisk treadmill walking - plausibly similar in dosage but lower in quality to the kind of movement our ancestors would actually get in a day.
Now let's look at the object-level studies that Scott Alexander says show that vitamin D doesn't work. I'm just going to look at the randomized controlled trials because observational studies for or against vitamin D are trash for anything except hypothesis generation unless they have a very carefully selected instrumental variable.
The largest randomized trial ever done on the subject, a 36,282-person behemoth, found zero effect of vitamin D on its two measured endpoints of colon cancer or breast cancer and in fact the vitamin D group had nonsignificantly more cancer than controls.
The colon cancer link is broken but the breast cancer study reports a dosage of 400 IU/day. On the exercise scale that's FIVE MINUTES of brisk walking. FIVE MINUTES is not very long at all compared with FOUR HOURS. If you went to see Lord of the Rings and only saw the first five minutes you would not be satisfied with a partial refund. If you showed up fasted to Thanksgiving dinner in the expectation of a full day's calories (the official US ration is 2000), this is like getting a single banana.
And a randomized controlled trial of 2700 people investigating all-cause mortality found zilch.
The intervention is 100,000 IU every four months. On the exercise scale this is like saying that because we live totally sedentary lives, instead of four hours of brisk walking every day, we should RUN A DOUBLE MARATHON ONCE EVERY FOUR MONTHS IN OUR OTHERWISE SEDENTARY LIVES. The marathon is named after the time when a messenger ran 26 miles home from the battle of Marathon, delivered the news, ran all the way back on the same day, delivered instructions to the general, and then died of exhaustion on the spot.
If on the other hand we amortize the dosage uniformly over the four months, that amounts to about 800 IU/day, or ten minutes of brisk walking on the exercise scale. Not nothing, but not at all the same sort of thing as four hours.
Now for the meta-analyses:
A meta-analysis by the Systematic Evidence Reviews people (who know their stuff) concluded that "Vitamin D and/or calcium supplementation also showed no overall effect on CVD, cancer, and mortality."
Let's forget about the fact that "vitamin D and/or calcium" is a stupid metric, except let's not, because it is a stupid metric we should not care about for anything except bone health. If vitamin D has specific effects on immune function or general wellness, there's no reason to expect that Calcium is a valid substitute. I also looked up the Study Characteristics of Included Studies, to see which ones had a treatment arm with vitamin D and found:
- Dean 2011 gave 5,000 IU/day to 63 people and measured "cognitive and emotional functioning", treatment period unspecified. Reasonable dosage - on the exercise scale that's an hour of brisk walking - but terribly underpowered.
- Graat 2002 used 200 IU/day of vitamin D. On the exercise scale that's TWO AND A HALF MINUTES. Also it was in a multivitamin with lots of confounding stuff, and group size was slightly more than 150.
- Lappe 2007 used 1000 IU/day. Twelve and a half minutes of walking.
- RECORD used 800 IU/day (ten minutes).
- Trivedi 2003 used 100,000 IU (DOUBLE MARATHON) every four months.
- WHI used 200 IU/day (two and a half minutes).
Wang et al found much the same (although their conclusions section does a terrible job elucidating this).
Broken link but I found the study on Internet Archive, and the abstract describes "moderately high doses" as averaging to about 1000 IU/day (twelve and a half minutes).
Autier looks at 172 randomized trials (!) and finds "Results from intervention studies did not show an effect of vitamin D supplementation on disease occurrence". [sic]
The highest dosage mentioned in the abstract is 2000 IU (25 minutes), which is at least getting somewhere. Maybe that's why "supplementation in elderly people (mainly women) with 20 μg vitamin D per day seemed to slightly reduce all-cause mortality." Also, a SIGNIFICANT REDUCTION IN ALL-CAUSE MORTALITY is a huge deal. That is the gold-standard outcome metric that everyone wishes their study was adequately powered to hit. Last time I checked statins don't reach all-cause mortality significance and they're widely prescribed. I don't independently find this meta-analysis very persuasive because the effect is barely significant, but summarizing this as a negative result is totally bonkers.
This is starting to get out of scope, and I'm not really interested in a conversation about whether Scott's characterizations are honest, so I'm going to stop here. Let's look for comparison at a study I liked.
The Spanish RCT studying vitamin D for COVID used a dosage regimen that - according to Chris Masterjohn's summary - was "equivalent to 106,400 IU vitamin D on day 1, 53,200 IU on days 3 and 7, and 53,200 IU weekly thereafter." Some of these are heroic doses, and the dosage regimen hardly seems optimal, but this is for people who had already been hospitalized with COVID, a situation of acute illness where the body might be churning through a tremendous amount of vitamin D. (For similar reasons I've started taking vitamin C megadoses when I get sick, because several grams per day could easily make a big difference even though the much smaller doses in RCTs don't.) On the exercise scale, this is equivalent to walking a double marathon on the first day, a single marathon on days 3 and 7, and weekly thereafter.
Intense, and not spaced out as much as I'd like, which is some evidence against the validity of the study - but neither obviously too small to make a difference nor spaced too far apart to be meaningful. I would expect better results from an equivalent daily dosage schedule. Masterjohn's calculation: "If this were given as daily doses, it would be the equivalent of 30,400 per day for the first week, followed by a maintenance dose of 7,600 IU per day."
ETA: In the comments on LessWrong Scott Alexander found a study estimating Hadza hunter-gatherer sunlight exposure equivalent to daily oral supplementation of 2000 IU. This seems like a better target for long-run supplementation than the maintenance dose in the Spanish RCT, and it's at a more comfortable distance from the estimated tolerable upper limit of 10,000 IU per day.
The main point here, that many negative results are from studies that are ruined by underdosing and by severely unnatural dosing schedules, seems correct to me. I think the root cause of the error is that people's ideas about how much vitamin D people need came from nutrition, with anchors set relative to amounts found in food, rather than thinking of it as a substitute for sun exposure, with small amounts coincidentally also occuring in food but not as the main source.
That said, I think comparing to 4 hours of peak sun exposure per day yields a significant overestimate. IIRC skin vitamin D synthesis shuts off after a certain point, and evolution would have tuned the rate of vitamin D synthesis such that that shutoff point would still be reached under below-average sun exposure conditions, ie during winter and in people who live under a forest canopy. I also suspect that use of clothing, loss of fur, and lighter skin arose close together in time, which also somewhat messes up the evolutionary argument.
There are plenty of places where ordinary clothing is still fairly light, and because we store vitamin D in our fat cells, it's not obvious to me that vitamin D production wouldn't be somewhat seasonal in the ancestral environment.
The last time I looked into it in more detail, there's not a discrete shutoff point, but a curve of diminishing returns where sunlight starts breaking vitamin D down into other (useful) metabolites at an increasing fraction of the rate at which new vitamin D is synthesized. I agree that 4 hours is an overestimate, which is why I suggested that a 16,000 IU supplement would better approximate the optimal amount of vitamin D than a 32,000 IU one. In practice I now take 10,000 a day (except when I forget, so my true average is lower), in part because there are rumors that this is the highest dosage widely shown to be safe for adults.
I support your cause 110%!
My 10% augmentation invokes two aspects of vitamin D supplementation: (1) enormous human variation in oral vitamin D uptake, and (2) nutrient interaction, particularly with vitamin K2.
1. The “Serum 25(OH)D vs Vitamin D Intake” chart that Scott Alexander cites in his substack response to your post shows that oral Vitamin D absorption varies enormously across humans.
As a result, the most direct way to assess the health impact of vitamin D supplementation requires individual subject measurement of serum 25(OH)D. I’m not aware of any such studies with mortality as a dependent variable. Even then, a satisfactory study would need to address nutrient interaction (#2 below).
So studies lacking individual subject serum 25(OH)D observation require, at minimum, statistical analysis acknowledging error in the independent variable. Again, I’m not aware of any such studies.
That said, even the most careful statistical treatment might be fruitless because of the extreme measurement error of the independent variable, spanning a range from insufficient to toxic: for example, we can see in Scott’s chart that at 10,000 IU/day intake, (a) a substantial fraction of subjects fail to reach 30ng/mL serum 25(OH)D, the current standard of vitamin D sufficiency, while (b) a similar fraction of subjects exceed 100 ng/mL, which could be toxic (see #2 below).
So, ironically, Scott’s attempted rebuttal undermines everything he ‘knows’ about vitamin D!
2. In a comment to your LessWrong linkpost, ChristianKl mentions his post advising vitamin K2 supplementation in conjunction with vitamin D supplementation in order to reduce hypercalcemia risk. We can see from this that an individual's vitamin K2 and calcium consumption will affect the benefit or harm from vitamin D supplementation.
Randomized controlled studies get stretched very thin when they need (or at least ought) to address causal interactions. Vitamin D, Vitamin K2 and calcium intake all produce non-monotonic health consequences -- good up to some point and then adverse beyond that. Furthermore, the location of the local maximum with respect to any one input will be a nonlinear function of the other two inputs. So even a 3x3x3 design will at best produce a crude rendition of the 'harm/benefit' landscape, while substantially reducing the study's statistical power compared to a basic treatment vs. control study.
In these circumstances, naive meta-analysis -- that is, adding up individual study results -- accomplishes little or nothing. To provide us with reliable guidance, we need a more sophisticated synthesis.