Alcohol and Longevity: Applying First Principles to an Uncomfortable Dataset
By Akash S. Chauhan | First Principles Healthspan, Issue 03
Few topics in nutrition epidemiology have been more thoroughly distorted by wishful thinking than the health effects of moderate drinking. For decades, the data appeared to show that light to moderate drinkers had lower rates of cardiovascular disease than abstainers — the so-called J-curve. That finding was widely reported, enthusiastically embraced, and turned into a cultural permission slip. It was also, as we now have strong evidence to believe, largely an artifact of how the research was conducted rather than a real biological effect.
This issue is not a moral argument against drinking. It is an attempt to apply the same rigorous standards we would apply to any other health claim — asking not "what did the observational data show?" but "does the causal evidence support it?"
Why this matters
The stakes are not trivial. Alcohol is a Group 1 carcinogen (International Agency for Research on Cancer), meaning there is sufficient evidence that it causes cancer in humans — specifically cancers of the oral cavity, pharynx, larynx, esophagus, liver, colorectum, and breast. At the same time, the cardiovascular benefit hypothesis gave many people a biologically plausible reason to maintain a habit they already enjoyed. If that hypothesis was an artifact, then the risk-benefit calculation changes substantially — and people making long-term decisions about healthspan deserve to know that.
The J-curve was not a discovery that moderate drinking is healthy. It was a finding that moderate drinkers are healthier. Those are different claims, and the difference matters enormously.
What the old evidence showed: the J-curve
The J-curve hypothesis was based primarily on large observational cohort studies. Ronksley et al. (2011) published an influential meta-analysis in the BMJ synthesizing 84 studies and finding that moderate alcohol consumption was associated with a 25% reduction in cardiovascular mortality and a 29% reduction in cardiovascular incidence compared to abstinence (PMID: 21343207). The shape of the relationship — lowest risk at light-to-moderate consumption, higher risk at both abstinence and heavy drinking — produced the characteristic J-curve.
This looked like a real biological signal, and researchers proposed plausible mechanisms: alcohol raises HDL cholesterol, has some antiplatelet effect, and may reduce fibrinogen. The finding was replicated across multiple cohorts, different populations, and different outcome measures. It was the kind of convergence that, in observational epidemiology, usually signals something real.
But the methodology had a fundamental flaw that took decades to fully appreciate.
The sick quitter problem
The most serious confound in the J-curve literature is what epidemiologists call the "sick quitter" or "abstainer bias" problem. The abstainer reference group — the people who do not drink — is not a healthy comparison group. It includes a substantial fraction of people who previously drank heavily and stopped because of illness, people who cannot drink due to medication interactions, people with pre-existing chronic conditions, and people in recovery from alcohol use disorder. When you compare moderate drinkers to this composite abstainer group, the moderate drinkers look healthier — not necessarily because alcohol is protective, but because the abstainer group is sicker on average.
Fillmore et al. (2006) demonstrated this directly by separating "lifetime abstainers" from "former drinkers" in the abstainer category. When only lifetime abstainers were used as the reference group, the apparent cardioprotective benefit of moderate drinking was substantially attenuated or disappeared entirely. This single methodological correction destabilizes much of the J-curve literature.
Additional confounds compound the problem: moderate drinkers tend to have higher socioeconomic status, better healthcare access, more social connection, and healthier diets than the abstainer population as a whole. These are exactly the variables that predict cardiovascular health, and disentangling them from alcohol consumption in observational data is not reliably possible.
What the new evidence shows: Mendelian randomization
Mendelian randomization is a technique that uses genetic variants as natural instruments to estimate causal effects, bypassing the confounding inherent in observational studies. The logic is clean: some people carry genetic variants that cause them to drink less (for example, variants in the ALDH2 or ADH1B genes that make alcohol metabolism aversive). These variants are assigned at conception, before health status is established, and are not associated with the confounders that plague observational research. If alcohol genuinely protects cardiovascular health, people with variants that cause them to drink less should have worse cardiovascular outcomes.
Millwood et al. (2019) applied this approach to 512,715 adults in the China Kadoorie Biobank, using genetic variants affecting alcohol metabolism to construct an essentially confound-free natural experiment (PMID: 31076257). The findings were stark: among men (who drink heavily enough in this population for the analysis to have power), genetic variants associated with lower alcohol consumption predicted lower blood pressure and lower stroke risk in a linear, dose-dependent fashion. There was no J-curve. There was no cardiovascular protection. The relationship between alcohol and these outcomes was monotonically harmful — more alcohol, more risk.
Biddinger et al. (2022) applied a similar design specifically to atrial fibrillation in JAMA Network Open, using Mendelian randomization in a large UK Biobank sample (PMID: 35133425). Even moderate alcohol consumption — as few as one drink per day — was associated with a significantly increased risk of AF. The effect was linear and causal. Given that AF is a major contributor to stroke risk and cardiovascular mortality, this finding closes another door on the cardioprotective hypothesis.
The global burden conclusion
The GBD 2016 Alcohol Collaborators published a comprehensive analysis in The Lancet in 2018 combining data from 195 countries on the relationship between alcohol consumption and 23 health outcomes (PMID: 30146330). Their conclusion, which generated considerable public attention, was that the safest level of alcohol consumption is zero — that when cardiovascular effects, cancer risk, injury, and other outcomes are combined, no level of consumption produces a net health benefit.
This does not mean that one drink a week is meaningfully dangerous for a healthy individual — the absolute risk at low consumption levels is small. But it does mean that the framing of "a drink a day is heart-healthy" has no rigorous causal support, and that the total-health risk curve, across all outcomes, does not have a J-shape. It starts at zero and goes up.
What the evidence does and does not say
It is worth being precise about what this research concludes. It does not conclude that moderate drinkers are doomed. It concludes that the apparent cardiovascular benefit of moderate drinking was almost certainly a methodological artifact, and that when causal methods are applied, the true risk relationship is linear and unfavorable, particularly for cancer and AF.
The cancer signal is worth emphasizing specifically because it is independent of the cardiovascular debate. Alcohol is metabolized to acetaldehyde, a direct DNA-damaging agent. This mechanism does not require heavy consumption. It operates at any dose. The International Agency for Research on Cancer has classified alcohol as a Group 1 carcinogen since 1988, and the dose-response for several cancer types — particularly breast cancer — shows no safe threshold in the data.
For someone optimizing healthspan over a 40-year horizon, this is the more important calculation than the cardiovascular question. The cardiovascular risk from light drinking, if it exists, is small at low doses. The cancer risk accumulates differently and does not have the same threshold structure.
This Week's One Thing to Do
Apply first principles to your own risk calculus. If you drink regularly, write down honestly: what do you believe the health effect to be, and what is that belief based on? If it is based on the observational J-curve literature — the "a glass of red wine is good for you" finding — you now have the Mendelian randomization evidence to weigh against it. The question is not whether you choose to drink. It is whether your choice is based on accurate information about the tradeoffs.
This is not a call to abstinence. It is a call to informed decision-making, which is what first-principles healthspan requires.
Until next week, Akash S. Chauhan
Education only. Not medical advice. Always consult a licensed clinician for individual decisions.