Effect of 3–6 cups of coffee daily, for 2 to 24 weeks, on resting heart rate. Six randomised trials, 485 participants. The interval spans zero and includes a decrease. The input variable the whole hypothesis rests on is null: everything downstream is being multiplied by nothing.
Where the chain breaks
Caffeine raises resting heart rate
Measured at +0.40 bpm, not significant. Adenosine antagonism raises blood pressure; the arterial baroreflex answers with increased vagal tone and heart rate stays flat. Tolerance to the pressor effect arrives within days.
A faster heart makes cells run faster
The causal arrow is reversed. Resting cardiac output is set by tissue metabolic demand, not the other way round. Local metabolites and endothelial signalling govern capillary perfusion; the heart supplies what the tissue beds draw. Pacing the heart faster without raising demand shortens diastolic filling and cuts stroke volume.
More throughput means more oxidative damage means shorter life
Mice deficient in both MnSOD and GPx-1 accumulated more oxidative damage and more pathology with no reduction in longevity. The last arrow fails under designed conditions.
Beats are a fixed budget you draw down
The mass-invariant heartbeat product is a consequence of two allometries, never a conserved quantity. Lower the number without changing the state underneath it, as ivabradine does, and nothing follows.
1 · The constant is not constant
Lifetime heartbeats, by species
A "law" with a fourfold spread inside the class it is supposed to govern is a summary statistic. Birds break it further still, sustaining high rates alongside long lives.
Rodent and mammalian-mean figures from Zhang & Zhang, Ageing Research Reviews. The human bar is arithmetic, not a claim: 70 bpm × 80 years = 2.94 × 10⁹, consistent with the ~3 × 10⁹ the same review reports.
2 · What the budget model would have to predict
Years to spend 2.94 billion beats
Take the budget literally and the curve below is your life expectancy as a function of resting heart rate. SIGNIFY ran the experiment: ivabradine, a pure funny-current inhibitor with no other haemodynamic action, slowed 19,102 patients by 9.9 bpm. The model says that buys 12.9 years.
At the coffee meta-analysis point estimate of +0.40 bpm, the same model charges you 0.45 years. Propagate the confidence interval and the cost runs from a 0.90-year gain to a 1.75-year loss.
3 · What actually moves resting heart rate
Change in resting heart rate, by intervention
Same units, same measure. Whiskers are 95% confidence intervals where reported, and the reported range across trials otherwise.
Exercise is the awkward case for the budget model, not caffeine. Training lowers resting heart rate by a few bpm while the sessions themselves drive it to 150–180 for hours each week. Lifetime beats come out flat or higher, and exercise remains the best-supported lifespan intervention in humans. The model predicts the opposite sign.
4 · Rate-of-living, inverted
Mice by mass-adjusted metabolic intensity
All three bars are percentage differences against the lowest-intensity quartile, so they share one axis. The high-metabolism mice burned more oxygen at rest and lived longer, with higher mitochondrial proton conductance through the adenine nucleotide translocase and UCP-3. The theory predicted the opposite sign.
Speakman et al., Aging Cell 2004. The middle bar is arithmetic on rate-of-living's own premise, not a figure from the paper: if lifetime energy expenditure per gram were fixed, lifespan would scale as 1 ÷ 1.17, or −14.5%. Bats and birds show the same inversion, high mass-specific metabolic rates alongside exceptional longevity. Rate-of-living did not lose to hormesis; it lost on its own evidence.
5 · A marker, not a lever
All-cause mortality by resting heart rate
40 studies, 1,246,203 participants, 78,349 deaths. The association is real and roughly linear. It indexes autonomic tone, fitness, inflammation and subclinical disease, which is why moving the number pharmacologically, as in SIGNIFY above, does not move the outcome.
Zhang et al., CMAJ 2016. Relative risks versus the lowest heart-rate category; the per-10-bpm estimate is the pooled dose-response slope.
6 · Where the real cost sits
CRAVE: 100 adults, randomised to coffee or abstention, 14 days
Continuous ECG, wrist accelerometry and continuous glucose monitoring. Cardiovascular tolerance to caffeine is near-complete, which is why the resting-heart-rate signal is null. Tolerance across its other effects is not, and the gap is where a plausible cost actually lives.
≈1 hour in slow metabolisers
Statistically significant
Sleep displacement is the largest realistic route by which habitual caffeine could shorten a human life, and it operates entirely outside the heartbeat frame. The activity increase runs the other way.
What the population data show
If caffeine were burning through a budget, mortality should climb monotonically with dose. Over the usual consumption range it does the opposite.
498,158 people · 12.1 yr · 34,699 deaths
Mendelian randomisation: null
Convex curvature; WBE predicts concave
Both cohort figures are observational, so residual confounding and reverse causation (sick people quit coffee) stay live. The Mendelian randomisation is the useful part, because it partly breaks the confounding. The scaling exponent is from Kolokotrones et al., Nature 2010: the 3/4 power that generates the mass-invariant heartbeat product is an average taken through a curve, not a law.
What survives
Not the budget, but three adjacent mechanisms hold up: sleep displacement; sustained blood-pressure elevation at very high chronic doses, an afterload mechanism with a large evidence base behind it; and arrhythmia triggering in people who already carry a substrate, which is what the CRAVE ventricular-ectopy signal points at. All three are dose- and phenotype-dependent, none of them count beats, and a fast CYP1A2 metaboliser drinking three cups before noon is in a different position from a slow metaboliser drinking six after 4pm.