The problem with using Mendelian Randomization to study alcohol and heart disease
This is a summary of an influential journal article written in 2020 by three well-known professors at the Harvard School of Public Health and Harvard Medical School:
“Genetic instrumental variable analysis: time to call mendelian randomization what it is. The example of alcohol and cardiovascular disease” (Eur J Epidemiol, Feb 2020), by Kenneth J Mukamal, Meir J Stampfer, Eric B Rimm
Paper · Eur J Epidemiol · 2020Genetic instrumental variable analysis: time to call mendelian randomization what it is. The example of alcohol and cardiovascular diseaselink.springer.comThey argue that there are many problems with using Mendellian randomization to study the relationship between alcohol and heart disease.
The same criticisms apply to using Mendellian randomization to study other kinds of foods and beverages, as well.
Also see this previous post on the subject: ↪ Mendelian randomization is a poor tool for studying alcohol and CVD.
————-
SUMMARY
To understand their critique, we first need to break down the complex tools they are discussing into everyday concepts.
- What is Mendelian Randomization?
The authors argue that we should stop using the fancy name Mendelian Randomization (MR) and use the more technical, honest name: Genetic Instrumental Variable (GIV) analysis.
The "Tool" (The Instrument)
In a standard experiment, researchers split people into two groups at random. In alcohol research, you can't easily force thousands of people to drink (or not drink) for 30 years to see who gets a heart attack.
Instead, scientists use genes as a "natural experiment." For example, some people are born with a genetic variant (like a "glitch" in the ALDH2 gene) that makes drinking alcohol very unpleasant—it causes flushing, nausea, and a racing heart. Because these people are genetically "assigned" to drink less from birth, researchers can use that gene as an Instrumental Variable (a proxy) to study the effects of low alcohol consumption.
- The Three Big Assumptions
For this genetic "short-cut" to work, three things must be true. The authors explain why these are harder to prove than they look:
A. Relevance (The "Strong Link")
The gene must actually change behavior.
The Problem: While a gene might make you likely to drink less, it doesn't force you to. Social pressure, culture, and personal choice still play a huge role. If the link between the gene and the actual behavior is weak, the whole math equation falls apart (Weak Instrument Bias).
B. Exclusion Restriction (The "No Side-Door" Rule)
The gene should only affect the heart by changing how much you drink.
The Problem: This is called Pleiotropy. A gene might make you drink less, but it might also make you more likely to exercise or change your cholesterol levels directly. If the gene affects the heart through these "side doors," we can't blame (or credit) the alcohol for the result.
C. Independence (The "No Hidden Factors" Rule)
The gene shouldn't be tied to other lifestyle factors.
The Problem: Population Stratification. Genetic variants aren't spread evenly. If a certain "low-drinking gene" is more common in a specific ethnic group that also happens to eat a very healthy diet, the study might wrongly conclude that the gene (and thus less alcohol) is what's keeping them healthy.
- The "J-Shaped Curve" Controversy
This is the heart of the debate. For decades, traditional studies have shown a J-shaped curve:
Heavy drinkers have high heart disease risk.
Non-drinkers have moderate risk.
Moderate drinkers (1–2 drinks a day) have the lowest risk.
Why GIV Challenges This
Recent GIV (Mendelian Randomization) studies often show a straight line instead: suggesting that as soon as you start drinking even a little, your risk goes up.
Mukamal and his colleagues are skeptical of this straight line.
They argue that the genetic method is "blunt." It looks at a person’s risk over their entire lifetime, but it can’t distinguish between someone who has one glass of wine with dinner for 5 weeknights in a row verses someone who binges five drinks all at once on a Friday night. Mendelian randomization (GIV) provides a lifetime average effect. It cannot distinguish between “moderate daily” drinking and heavy episodic" (binge) drinking. For example, if a genetic variant makes group Group A less likely to drink 5 extra units of alcohol per week than Group B, the model looks at the health outcomes of Group B and attributes them to those 5 more units. However, it cannot see HOW those 5 extra units were actually consumed by Group B. Were they spread out over the entire week? Or were they consumed in binge fashion all at once? One behavior might be healthy while the other is likely harmful.
Additionally, being free of genetic impediments to drinking (e.g. not having the ALDH2 gene variant that causes flushing) does not guarantee that you WILL actually drink. Lots of people voluntarily choose to abstain. Mendelian randomization has no way to distinguish these abstainers genetically. It treats them as being prone to drinking, when they actually do the opposite. This is a major problem. They are analyzed as a “drinker" in the intent-to-treat style analysis. This can dilute the results or bias them, especially if those people differ in other health-conscious ways.
- Why the Name Matters
The authors are worried about the word "Randomization." In a Randomized Controlled Trial (RCT), everything is controlled. In Genetic Instrumental Variable (GIV) analysis, nothing is controlled—we are just observing nature and using complex math to guess at cause and effect.
By calling it "randomization," scientists might give the public (and doctors) a false sense of certainty. The authors suggest that GIV is a great clue, but it shouldn't be treated as the final "truth."
- The Takeaway
The article concludes that while using genetics to study health is a brilliant innovation, it has "blind spots." We shouldn't throw away 50 years of traditional nutrition science just because a new mathematical model says otherwise. Instead, we should:
Use genetics as one piece of the puzzle.
Be honest about the math's limitations.
Remember that a gene is a "proxy" for behavior, not the behavior itself.
#alcoholHD #mendelianrandomizationHD #mrHD #meirstampferHD #harvardHD



