Mendelian Randomization: How it Works, and What it Reveals about Vitamin D / by F. Perry Wilson

An article in the BMJ uses the statistical approach to suggest that Vitamin D supplements might not prevent bone fractures.

This week, how Mendelian randomization works, and what it can tell us about Vitamin D, milk drinking, and bone fractures as we check out this study, appearing in the BMJ:

title.JPG

So – what is Mendelian randomization? Broadly, it’s a statistical framework that attempts to figure out if some exposure causes some outcome.

 Cause, meet effect.

Cause, meet effect.

Causation is the holy grail of medical research, because, for example, if low vitamin D levels cause bone fractures, then treating low vitamin D levels would prevent fractures.

So let’s say you want to know whether low vitamin D levels cause bone fractures.

You measure vitamin D levels in a bunch of people, follow them over time, and see if those with lower levels develop more fractures.

 Confounders, the bane of observational research.

Confounders, the bane of observational research.

But be careful – what you just described was an association between low vitamin D levels and fracture risk. You know very well that other factors, like diet, exercise and sun exposure can affect vitamin D levels and might independently change your fracture risk.

These other factors are known as confounders, variables that are linked with both the exposure of interest (vitamin D level) and outcome of interest (fracture) and they are the bane of observational research. We have a slew of statistical tools to try to account for confounders, but none are perfect.

No if you really want to get at causation, you should do a randomized trial, but these can be tricky and very expensive. It’s not always ethical or feasible to randomize people to certain exposures, which is probably why we have yet to see a definitive trial exploring the efficacy of absinthe use on painting ability.

Mendelian randomization gets around these issues via an interesting epidemiologic trick.

We know there are a host of things – like lack of exercise - that might be associated both with low vitamin D levels and fracture risk. But there are genes that are associated with low Vitamin D levels too.

Some people are born with gene variants that predispose them to lower Vitamin D levels throughout their lives. Whether you get one of those genes or not, practically speaking, is random. It has nothing to do with eating habits or exercise - It’s an accident of birth.

 See how genetic variants have no path to fracture that DOESN'T go through Vitamin D levels? That's the central assumption of Mendelian Randomization, as well as its secret sauce.

See how genetic variants have no path to fracture that DOESN'T go through Vitamin D levels? That's the central assumption of Mendelian Randomization, as well as its secret sauce.

So if low vitamin D levels cause fracture, than a group of people genetically cursed to be predisposed to low vitamin D levels should have higher fracture rates. On the other hand, if low Vitamin D levels are merely a marker of other healthful behaviors, being born predisposed to have low levels will have no effect on fracture risk.  This is the genius of Mendelian randomization. We use the genes people have had their whole lives, genes that never change, to get around the mess of the myriad life choices that affect measured Vitamin D levels.

The genius of the BMJ manuscript is that they used genome-wide association data from over 500,000 individuals to perform mendelian randomization analysis not just on vitamin D levels, but on a host of classic fracture risk factors: including dairy intake, diabetes, age at menopause and more – all things that are associated with fracture but have not been shown to cause fracture. This is a shotgun approach to find something we can change that will actually impact the fracture rate.

What they found was that out of 14 risk factors for fracture, only 1, bone mineral density, had strong evidence of causality. Grip strength, a proxy for muscle mass, had weak evidence of causality.

 A "causality plot". All of these things are  associated  with fracture, but only some appear to  cause  fracture.

A "causality plot". All of these things are associated with fracture, but only some appear to cause fracture.

Remember – they are not saying that these other things aren’t associated with fracture, they are saying that changing the other things won’t change the rate of fracture. So we get a ton of information from this study – it suggests that, for example, while those who have earlier menopause are more likely to have a fracture, delaying menopause is unlikely to make a difference in the fracture rate.

What about vitamin D? Well – this study would suggest that all those vitamin D supplements we take aren’t doing much to decrease the fracture rate, and that finding broadly aligns with randomized trials that have examined vitamin D supplementation.

So that’s Mendelian randomization – I hope I’ve brought a bit of sunlight to the issue.

This post first appeared on medscape.com.