Alcohol
Evidence BWhat the evidence actually shows
For thirty years the received wisdom was a J-shaped alcohol curve: teetotallers and heavy drinkers both died more, moderate drinkers least, so a daily glass of wine was practically medicine. The GBD 2016 Alcohol Collaborators, analysing consumption and outcomes across 195 countries, took that curve apart. For total health harm the level of consumption that minimises risk turned out to be zero, and the all-cause picture is dominated by the fact that the "protective" dip was an artefact.
The mechanism of the artefact is well understood. The abstainer reference group in the old studies was contaminated with "sick quitters" — former heavy drinkers who stopped because they were already ill, plus people who never drank due to poor health. Comparing moderate drinkers against that unhealthy reference group manufactures a fake benefit. Clean the reference group up, or use Mendelian randomisation with genetic variants that predict alcohol metabolism, and the protective dip shrinks toward nothing.
How big the effect really is, in years
Heavy sustained drinking costs on the order of a few years of life expectancy — we price it at roughly −3 years as a central, hedged figure, with the real number depending heavily on how much, for how long, and alongside what else (heavy drinkers smoke more, and the two effects tangle). Light-to-moderate drinking, honestly, sits close to neutral: a small net harm that is hard to distinguish from zero and definitely not the benefit the older literature advertised.
So the actuarial summary is deflationary in both directions. The glass of red is not extending your life, and the occasional weekend is not visibly shortening it either. The years are lost at the top of the dose distribution, in sustained heavy use, where the liver, the cardiovascular effects, the cancers, and the accidents all compound.
The catch: it is genuinely hard to measure
Two things keep alcohol at grade B rather than A. First, self-reported consumption is systematically under-reported — compare survey totals against alcohol sales and a third of the drinking simply vanishes — so every exposure category is shifted and blurred. Second, the confounding is ferocious in both directions: drinking correlates with income, sociability, smoking, and diet, and the abstainer group is a mix of the very healthy and the very unwell.
Mendelian randomisation helps by using genes as a natural experiment immune to lifestyle confounding, and it consistently fails to reproduce the protective J. That convergence is why we are confident the harm rises with dose and confident that the "moderate benefit" was mostly bad reference groups — even though the exact year-figures remain fuzzy. Confident about the shape, humble about the decimals.
How this feeds your actuarial age
The alcohol modifier applies a dose-dependent adjustment that is essentially flat through the light range and turns negative as intake climbs into heavy territory. Critically, we do not build in a protective dip for moderate drinking, because the best evidence says that dip was a measurement artefact and baking it into the model would just launder a statistical error into your life expectancy.
In practice this means light drinkers see almost no movement, and the modifier only bites at genuinely heavy sustained intake. If you were hoping the calculator would reward your nightly glass with extra years, it will not — but neither will it punish moderate use as if it were a pack of cigarettes. If your drinking is in the range where these years start disappearing, that is a clinician conversation.
This factor feeds directly into your actuarial age. Run the calculator →
Sources
Reviewed 2026-07-06 by Dmytro Dubina, Actuary · MSc Probability & Statistics · 20+ years in insurance. Population statistics, not medical advice.