The sick-quitter effect

Don't start drinking for your health. When a study says light drinkers live longer, check who it counted as a non-drinker before you believe it.

What it is

For about thirty years, observational studies kept producing a J-shaped curve: light drinkers died less often than people who did not drink, and death rates only climbed at heavier intakes. The explanation offered was that alcohol raises HDL ("good") cholesterol and makes blood slightly less prone to clotting.

The flaw sat in the comparison group. Many people counted as non-drinkers were former drinkers, and many of them had stopped because they were already ill. Putting them in the reference group makes not drinking look dangerous and light drinking look protective, with no help from the alcohol itself. That contamination of the abstainer group is the sick-quitter effect, also called abstainer bias. It is not the only distortion: moderate drinkers in Western countries also tend to be better off and live more healthily, which no cohort fully adjusts away.

What happens when you correct for it

A 2023 review pooled 107 cohort studies covering about 4.8 million people and over 425,000 deaths. Taken at face value, the data reproduced the familiar J-curve. Once the analysis corrected for studies that counted former drinkers as non-drinkers, and for study quality, the low-dose benefit went: occasional drinkers died at essentially the same rate as lifetime non-drinkers, a 4% difference well within the range expected from chance (relative risk, RR, 0.96, 95% confidence interval 0.86–1.06 — the range where the true effect most plausibly lies; when it crosses 1.0, the result is compatible with no effect). Low-volume drinkers, at 1.3–24 g a day, also showed no significant difference.^[1]

Harm, by contrast, showed up clearly. Mortality ran about 19% higher at 45–64 g a day and about 35% higher at 65 g a day or more. Women crossed the line sooner: about 21% higher mortality than female lifetime non-drinkers at 25–44 g a day, against a threshold of 45 g a day in men, where risk ran about 15% higher. Pooled across every level of intake, female drinkers as a whole ran about 22% above female lifetime non-drinkers.^[2]

A follow-up on the same 107 studies sorted them by quality instead of adjusting for it. The studies that enrolled younger people, followed them past 55 and kept former and occasional drinkers out of the abstainer group found low-volume drinkers dying at essentially the same rate as abstainers, a 2% difference (RR 0.98, 0.87–1.11). The studies that failed those tests produced the familiar 16% benefit (RR 0.84, 0.79–0.89).^[3] The J-curve, in other words, shows up mainly in the weaker studies.

An independent Australian group reached the same place from outside that research team. Of 25 published reviews of drinking and death from any cause, only five had kept former drinkers out of the reference group, and 17 of the 20 that had not reported lower mortality at low intakes. The correction does not erase the J-curve everywhere, though: three of the five reviews that excluded former drinkers still found one, while the only review judged at low risk of bias found risk rising steadily with every level of drinking.^[4]

The correction is disputed

The bias adjustments have been argued over in print. A defence of moderate drinking, published as a formal comment on the 107-study review, disputes them; its authors disclose extensive past funding and consultancies from wine, beer and spirits bodies.^[5] Conflict-of-interest flags run through this field on the advocacy side as well as the industry side.

The strongest dissent is a congressionally mandated review by the US National Academies of Sciences, Engineering, and Medicine (NASEM). It pooled eight studies and found that moderate drinkers had about 16% lower mortality from any cause than people who never drank (RR 0.84, 0.81–0.87), graded moderate certainty.^[6]

This is not the crude sick-quitter error. The review excluded studies that lumped former drinkers in with never-drinkers, and says its results "are not directly comparable to past evidence that does include such abstainer bias." Its limits are subtler. Every study it used was observational; there was no randomised trial in the report at all. People who report never drinking are unreliable about their own history, and lifelong abstainers differ from drinkers socioeconomically in ways no cohort fully adjusts away. And drinking more versus less within the moderate range made no measurable difference to mortality (RR 0.96, 0.87–1.06), graded low certainty.^[7]

A parallel federal review, released in January 2025 a month after the NASEM report, reached the opposite conclusion, that "at no age was there a significant net benefit of alcohol consumption on all-cause mortality." It got there by a different route, modelling deaths from conditions alcohol is known to cause rather than pooling mortality studies. It never went beyond a public-comment draft.^[8]

The same trap beyond mortality

The pattern turns up wherever people change their drinking because of their health.

• Dementia. In over half a million older adults in the US and UK, a U-shaped curve appeared in the ordinary analysis and disappeared in a genetic one, where risk rose with intake. People who went on to develop dementia had already been cutting down, and when intake was measured closest to diagnosis, moderate drinking looked protective again.^[9]
• Brain scans. In about 25,000 UK adults, never-drinkers had the same grey-matter volume as the lightest drinkers, while former drinkers had less than the lightest drinkers. Grey matter was still steadily lower with higher intake, with smaller volumes from 56–112 g a week.^[10]
• Atrial fibrillation. A 2026 pooling found 28% lower risk of this irregular heartbeat below 12 g a day than in people reporting no intake — built on the same kind of no-intake reference group. The same analysis found 75% higher risk above 60 g a day.^[11] The genetic evidence points the other way, and the one randomised trial found that cutting down helps people who already have it.

Two designs get around it. Comparing drinkers with other drinkers keeps abstainers out entirely: across three large US cohorts, people drinking on 5–6 days a week had about a quarter lower risk of type 2 diabetes than those drinking on 1–2 days. That is a real association, but frequent light drinkers may differ from occasional ones in many other ways, and the genetic analyses do not reproduce it.^[12] Genetic studies, which use inherited variants that make people drink more or less, are much less open to confounding by illness, and in several of them the protective curves seen for heart disease, diabetes and dementia vanished. The Alcohol article covers what they show.

What to do

1. Don't start drinking for a health benefit. The benefit the J-curve promised largely disappears once the comparison is corrected for former drinkers.
2. Read "light drinkers live longer" headlines with one question: were former drinkers kept out of the non-drinking group? If not, discount the result.
3. Give more weight to studies that compare drinkers with drinkers, or that use genetics, than to those that compare drinkers with everyone who reports not drinking.

Further reading

• Zhao J et al. Association Between Daily Alcohol Intake and Risk of All-Cause Mortality: A Systematic Review and Meta-analyses. JAMA Netw Open 2023.^[13]
• Stockwell T et al. Why Do Only Some Cohort Studies Find Health Benefits From Low-Volume Alcohol Use? A Systematic Review and Meta-Analysis of Study Characteristics That May Bias Mortality Risk Estimates. J Stud Alcohol Drugs 2024.^[14]
• Sarich P et al. The association between alcohol consumption and all-cause mortality: An umbrella review of systematic reviews using lifetime abstainers or low-volume drinkers as a reference group. Addiction 2024.^[15]
• NASEM. Review of Evidence on Alcohol and Health. National Academies Press 2025.^[16]