Alcohol intake → All-cause mortality

Mechanism

Observational papers have long sold a J-curve: people who drink a little look safer than people who drink nothing, so a glass looks protective. Sick quitters, smoking, money, and who is still alive to join a cohort sit on that backdoor. Researchers use a genetic shortcut called Mendelian randomization. It is good at ruling out diet, income, and the decision to stop drinking. It can still be fooled when one gene affects two traits. Kassaw, Zhou, Mulugeta, Lee, Burgess, and Hyppönen (2024) instrumented habitual grams per day with 94 SNPs from Liu's alcohol GWAS, keeping those that pointed the same way in UK Biobank. The sample was 278,093 unrelated white-British adults aged 37 to 73. Death came from NHS registers through 12 November 2021: 20,834 deaths, median 12.6 years. Linear Mendelian randomization used a genetic risk score and five summary methods (inverse-variance weighted, Egger, weighted median, weighted mode, MR-PRESSO). Nonlinear Mendelian randomization used the doubly-ranked method in 25 strata, because the gene-to-intake link is stronger where people already drink more. Reported intake against death was J-shaped. Genetically predicted intake was not. Each extra 8 grams a day (one UK unit) raised all-cause death (odds ratio 1.27, 1.16 to 1.39), cardiovascular death (1.30, 1.10 to 1.53), cancer death (1.20, 1.08 to 1.33), and digestive death (2.06, 1.36 to 3.12). Respiratory and COVID-19 death were imprecise. Tests for curvature did not reject a straight line (P of 0.21 or higher on every outcome). Cutting the follow-up before the pandemic did not change the all-cause result.

Caveats

This page is adult alcohol intake and later death from any cause. It is not a heart attack someone survived, not prostate cancer, and not colorectal methylation. The sample is white-British UK Biobank. A 5 percent volunteer cohort is healthier than the country. Intake is self-report. The genetic score still associated with smoking and Townsend deprivation; adjusting for both did not move the estimates. Mendelian randomization assumes no pleiotropy and a lifelong linear effect. It can miss a binge, a decade of abstinence, or a threshold in very heavy drinkers. An earlier nonlinear Mendelian randomization of alcohol and cardiovascular disease reported a curve, including for all-cause death. Kassaw argues that method assumed a constant gene-to-intake link, which fails here. We take their linear genetic file, not the observational J-curve. The design is Mendelian randomization. Holmes 2014 and Millwood 2019 are vascular events, not all-cause death. Wood 2018 is an observational pool of current drinkers. Those files stay off this page. This is not medical advice and not a recommendation to drink, to abstain, or to treat a number.

Effect

odds_ratio = 1.27. Kassaw 2024: OR 1.27 (1.16-1.39) all-cause death per +8 g/day genetically predicted alcohol. Linear; P for nonlinearity 0.21 or higher. Observational J-curve in the same sample.

Nodes

Named confounders

• Educational attainment
• Disposable income
• Sleep duration and regularity

Papers

• Alcohol consumption and the risk of all-cause and cause-specific mortality-a linear and nonlinear Mendelian randomization study.

  NA Kassaw, A Zhou, A Mulugeta, SH Lee, S Burgess, E Hyppönen · 2024 · International journal of epidemiology