Skip to content
What Causes What
SMOKE~ NON-LINEAR

Caffeine intake → All-cause mortality

Higher Caffeine intake moves All-cause mortality in a non-linear way. Dose or timing can flip the sign.

Li 2020 pooled 21 cohort articles (10,103,115 people, 240,303 deaths) and found a nonlinear inverse between coffee and all-cause death: three cups a day against none or rare sat at relative risk 0.87 (0.84 to 0.89). Tsujimoto 2017 saw the same U in caffeine milligrams. Nobody assigned a dose.

Meta-analysis · Low · N=10103115

The association disappears when you account for other factors, like socioeconomic status or healthy-user bias.

Mechanism

Coffee drinkers tend to smoke more, drink more, and move less. After those habits are named, cohorts often show a U: fewer deaths around a few cups or a mid caffeine dose, then a flatten or a rebound. Authors point at chlorogenic acid and insulin. They do not isolate one path, and they do not agree caffeine is the actor. Li, Liu, Sun and colleagues (2020) pooled 21 cohort articles, 10,103,115 people, 240,303 deaths. Splines were nonlinear (P < 0.001). Three cups a day versus none or rare: relative risk 0.87 (0.84 to 0.89). The abstract reports a similar inverse for caffeinated coffee and for decaf. Lopez-Garcia, van Dam and Hu (2008) followed 41,736 men and 86,214 women on cumulative coffee. After age, smoking, and diet, all-cause relative risks in women across cups were 1.0, 0.98, 0.93, 0.82, 0.74, and 0.83 at six or more (P for trend < 0.001). Men sat at 1.0, 1.07, 1.02, 0.97, 0.93, and 0.80 (P for trend 0.008). Pooled sexes: 0.86 at two to three cups, 0.79 at four to five, 0.83 at six or more. The inverse ran through cardiovascular death, not cancer. In women on two or more cups a day it sat below 1 in every caffeine quintile. Decaf also tracked a small inverse, mainly in women. Zhou, Ruan and Zhang (2025) followed 46,222 NHANES adults, 7,074 deaths. Total coffee versus none: 0.89 under one cup, 0.84 at one to under two, 0.83 at two to under three, 0.85 at three or more (P-trend 0.004). Splines were nonlinear (P < 0.001). Black coffee 0.86. Low-sugar, low-saturated-fat coffee 0.86. High sugar and cream did not sit below 1. Caffeinated cups tracked the total. Decaf under three cups sat near 1; three or more sat at 0.74. Tsujimoto, Kajio and Sugiyama (2017) followed 17,594 NHANES 1999-2010 adults, 1,310 deaths. Caffeine milligrams versus under 10 mg a day: 0.81 at 10 to 99, 0.63 at 100 to 199, 0.69 at 200 or more. The same U appeared in people who drank under one coffee cup a week. No cardiovascular link.

Caveats

The stamp is smoke because nobody randomized caffeine or coffee against death. These are cohorts and a meta of cohorts. Observation alone cannot close the backdoor, so the stamp stays smoke. More papers of the same design do not settle the claim. Smoking sits on the backdoor. In Lopez-Garcia, age-only models in the heaviest drinkers sat above 1. After smoking, the sign flipped. Residual smoking and income, education, and diet remain. The exposure here is milligrams. Three of the four files count coffee cups. Li reports a similar inverse for decaf. Lopez-Garcia says the coffee inverse was independent of caffeine. Zhou's inverse was mostly caffeinated cups and mostly black or lightly sweetened coffee. Tsujimoto is the file that isolates milligrams, including in people who barely drank coffee. Zhou 2025 and Tsujimoto 2017 share NHANES. They are not two countries. Li's search closed in November 2017 and likely already holds Lopez-Garcia. Omitted van Dongen 2017 and the 2020 post-MI review: coffee after a heart attack, not all-cause death in the general population. Omitted : atrial fibrillation events, not death. Omitted : all-cause death in French HIV-HCV co-infection. Omitted : tea in chronic kidney disease. Omitted : green-tea cups in Vietnam, not a caffeine dose. Omitted : UK Biobank tea cups. A caffeine-metabolism score was a modifier, not the exposure. This is a journal entry, not medical advice and not a reason to drink coffee, to skip cream, or to treat a number.

Effect

qualitative. Li 2020: 21 cohorts, 10,103,115, 3 cups/d vs none/rare all-cause RR 0.87 (0.84-0.89), Pnonlinearity <0.001. U, not a linear drop. Design does not identify.

Nodes

Lifestyle & environmentCaffeine intakeBiology & endocrinologyAll-cause mortality

Named confounders

Cite this page

What Causes What. “Caffeine intake → All-cause mortality.” https://whatcauseswhat.org/edges/e-caffeine-mortality (atlas updated 2026-09-02).

https://whatcauseswhat.org/edges/e-caffeine-mortality

Papers