Alcohol and Health: Is There a Safe Dose? What New Research Changed

For many years, large observational studies consistently showed the so-called J-curve — a curve where people drinking moderate amounts of alcohol had lower cardiovascular death risk than people who didn't drink at all, while only high consumption was linked to clearly elevated risk. This pattern seemed consistent enough across studies to become the basis for popular recommendations about a "safe," even protective, dose of alcohol.

The problem is that observational studies, regardless of their scale, have a fundamental limitation: they can't fully distinguish cause from correlation. Non-drinkers in such studies are often a heterogeneous group, including people who stopped drinking because of already-existing health problems — which artificially inflates their risk and artificially "improves" the statistics of the moderate-drinking group. A newer research method, Mendelian randomization, was designed specifically to eliminate this problem.

The turning point in this discussion is the 2024 work by Kassaw and colleagues, published in the International Journal of Epidemiology, using Mendelian randomization on a large, multi-year sample.

To understand the scale of this discrepancy, it's worth comparing the above result with a classic review of observational meta-analyses, such as the 2014 work by Roerecke and Rehm, published in BMC Medicine — well representative of the type of evidence that "safe drinking" recommendations were based on for years.

The main suspected mechanism of error is the so-called reference-group problem — in many observational studies, the group of "non-drinkers" includes both people who never drank and people who stopped drinking due to illness, past addiction, or medications. Such a mixed reference group has an artificially elevated death risk from causes unrelated to alcohol itself, which statistically "improves" the results of the moderate-drinking group in comparison to it, creating an apparent protective effect that doesn't actually exist.

It's worth noting that this doesn't mean earlier observational studies were methodologically "bad" — they simply study a different kind of relationship and are more susceptible to a specific type of systematic bias that Mendelian randomization partly avoids. This is a natural, healthy evolution of scientific knowledge, not proof that science "is randomly wrong."

In practice, this means shifting from thinking "what dose of alcohol is safe" to "any amount of alcohol carries some measurable risk increase" — the decision to drink remains personal, but it's worth making it without the false belief that a protective, health-promoting dose exists.

This topic is one of the best examples of how a change in research method — from observational to Mendelian randomization, less prone to specific systematic biases — can reverse decades of seemingly well-established health recommendations. This doesn't mean earlier science was dishonestly false, just that research methods have their own limitations, which become better understood over time.

The decision to drink alcohol remains an individual choice, factoring in social elements and enjoyment, not only a purely health-related risk calculation. But it's worth making that decision aware that the newest, methodologically more solid evidence doesn't confirm the existence of a health-"safe," let alone protective, dose.