The Role of BMI in Disease Risk

I have a first-author paper out in Nature Communications: Sequence variants associated with BMI affect disease risk through BMI itself. This is something I worked on for a long time at deCODE, and I want to write about it here in a way that goes a bit beyond the press release.

The core question

Obesity is associated with a long list of diseases: type 2 diabetes, heart failure, fatty liver disease, and others. The association is clear. Pinning down whether BMI itself is the culprit, or whether it proxies for something else, has been harder. Physical inactivity, diet quality, or metabolic factors that travel alongside high BMI could be the real drivers. If so, telling someone to lose weight might not reduce their risk, because they would still carry those underlying factors.

For prevention and treatment, the difference between “BMI causes disease” and “BMI correlates with things that cause disease” is enormous.

Using genetics as a natural experiment

The tool we used is Mendelian randomization. The logic: we have identified many genetic variants that influence BMI. These variants are assigned randomly at conception. They are not correlated with diet choices, socioeconomic status, or any of the usual confounders that make observational studies tricky to interpret. So if we find that people who carry more BMI-raising variants also have higher rates of a disease, and if that association disappears once we account for their actual BMI, we have good evidence that BMI is the pathway. The gene variants don’t know anything about your lifestyle; they raise BMI, and the disease risk follows.

We applied this to data from Iceland and the UK Biobank, covering a wide range of obesity-related diseases.

What we found

The results fell into a few categories.

For fatty liver disease, glucose intolerance, and knee replacement, the genetic association with disease was fully explained by BMI. Once you condition on BMI, there is essentially nothing left. This suggests these conditions are driven quite directly by the mechanical and metabolic consequences of carrying excess weight.

For type 2 diabetes, heart failure, myocardial infarction, atrial fibrillation, hip replacement, and stroke, the association was largely but not entirely attenuated. BMI explains most of the genetic risk, but something is left over, which could mean that BMI earlier in life matters more than a single measurement, or that closely related metabolic traits are doing some independent work.

We also found some sex differences: the attenuation pattern was broadly similar between men and women, but myocardial infarction stood out as differing by sex, which is consistent with what is already known about cardiovascular disease and sex.

For a wide set of diseases, reducing BMI would reduce risk. That strengthens the clinical case for weight management interventions, including the GLP-1 receptor agonists now achieving large BMI reductions in patient populations.

The interview

Shortly after the paper came out, Kári Stefánsson, who leads deCODE, sat down with me to talk through the study. We discuss the methodology, the findings, and what they might mean in practice.

A personal note

This paper took a while to get right, and it went through more iterations than I care to count. Getting a first-author paper out of deCODE, where the collaboration lists are long and the standards are high, means a lot to me. The genetics of complex traits like BMI involve a lot of moving parts (polygenicity, population structure, measurement issues) and I am proud of how the analysis came together.

GLP-1 drugs are reshaping weight management, and understanding why BMI reduction helps matters more now than when we started this work. Genetics gave us a cleaner handle on that question than diet studies could.