No level of drinking has been shown to be good for you. The appearance of benefit at low volumes is an artefact of how studies define non-drinkers, and risk rises with dose from the first drink.
What it rests onSystematic review and meta-analysis of 107 cohort studies and 4.8 million people, in which the protective association disappears after adjusting for cohort age and sex and for abstainer bias; 86 of the 107 studies put former or occasional drinkers in the abstainer reference group and only 21 were free of both biases.
Heavy drinking is plainly harmful, but small amounts remain a legitimate open question, and 'no amount is safe' overstates what the evidence supports.
What it rests onPooled cohort evidence showing lowest coronary heart disease mortality at one to two drinks a day; a 2025 American Heart Association scientific statement he co-authored reading the low-intake cardiovascular evidence as no risk to possible risk reduction; and a published methodological argument that Mendelian randomization is genetic instrumental variable analysis rather than randomization.
What it rests onThe Surgeon General's Advisory identifying seven cancers linked to alcohol, breast cancer in particular, and the population-health argument that lower consumption means fewer health and social problems. She notes cancer warnings appeared in dietary guidelines 25 years ago and have since been dropped.
The three are frequently not discussing the same drinkers, the same doses, or the same outcomes, and most of the apparent disagreement lives in that gap. Dose: Mukamal's ceiling is one to two drinks a day, roughly 14 to 28 g of ethanol; Canada's low-risk band, which Stockwell defends, is two drinks a week.
What one of them calls moderate sits at or above what another calls high risk, so each can be right about their own number. Outcome: Mukamal's protective signal is specific to coronary heart disease and type 2 diabetes, while the same body of work, including his own papers, finds harm for colorectal and breast cancer, and the BMJ meta-analysis he co-authored found nothing at all for stroke. Stockwell's null is all-cause mortality, which nets harms against benefits by construction. Nestle is not arguing about the curve at all: she is arguing about what national dietary guidance should say, and her outcome is cancer, from the Surgeon General's Advisory naming seven of them.
Population: the strongest J-curve cohorts are US nurses and male health professionals, wealthier, better educated and better insured than the general population, while Stockwell's pooled analyses span 107 cohorts of varying quality and his policy work concerns whole populations including dependent drinkers. Pattern: Mukamal's distinctive finding is about frequency, small amounts spread across the week, while Stockwell's analyses are built on volume. Reference group: the dispute between those two reduces, more than either usually says out loud, to who counts as a non-drinker.
Where they agree
None of the three recommends starting to drink, or drinking more, for health; the AHA statement Mukamal co-authored says it remains unknown whether drinking is part of a healthy lifestyle and tells clinicians to reinforce exercise, avoiding tobacco, and healthy weight instead. All three accept that three or more drinks a day, and binge drinking, are consistently harmful.
All three accept that alcohol causes cancer: it is Nestle's whole argument, Stockwell's headline finding, and something Mukamal co-authors papers reporting. Mukamal and Stockwell both say the question needs a long-term randomized trial, and both agree no such trial has been completed.
What would settle it
A completed long-term randomized trial of low-volume drinking against abstention, which is what MACH15 was designed to be before NIH ended it in June 2018 after 32 randomizations.
Short of that: replication of the bias-corrected reanalyses inside the Harvard cohorts by their own investigators; Mendelian randomization work with instruments that survive the pleiotropy and replication objections; or an agreed reference-group standard across the field, which would remove most of the argument between Mukamal and Stockwell. Worth revisiting in a year: the pilot remote abstinence trial from Mukamal's own work (PMID 41686020, 2026) is the most likely near-term mover.
Nestle's case is about disclosure rather than dose, so what would move her is different again: evidence that the cancer attribution in the Advisory does not hold at low intakes, or a guidelines committee restoring the cancer warning she says was dropped.
Practice does not pick a winner here
You do, and your call is dated, kept, and yours to change. Follow the people you already trust and what they publish turns into a short daily practice.
Record where you land on this, and change your mind later without losing the history
The 2 practices that follow from this question, graded and linked to the source
Everything else these people publish, read and graded, in one place
Tim StockwellPhD · alcohol epidemiologist and addictions researcher
