3 The body under sound
The reclassification of noise from nuisance to risk rests here, on physiology and on epidemiology that agree with each other. The body does not need to be consciously annoyed to be affected: a passing lorry at night raises heart rate and blood pressure through the autonomic nervous system whether or not it wakes the sleeper, and chronic exposure keeps the stress response switched on. The pathway runs through sympathetic activation, stress hormones, inflammation, oxidative stress, endothelial dysfunction and fragmented sleep, and over years it leaves a cardiovascular and metabolic mark the epidemiology can see.
3.1 The pathway, not the annoyance
The injury at stake is not the obvious one. It is not damage to the ear. Environmental noise at the levels found beside a motorway or beneath a flight path is, with rare exceptions, far too quiet to threaten hearing; the resident losing sleep to the road outside is not going deaf. The cardiovascular toll is entirely non-auditory. It travels not through injured hearing but through the body’s stress machinery, which treats a sound as a signal to be reacted to independently of, and long before, any conscious verdict on whether it is wanted. That distinction is what moves noise out of the audiologist’s clinic and into the epidemiologist’s cohort. A sound need not be loud to matter; it needs only to be there, and to be the kind of thing a vigilant nervous system keeps answering.
Exposure to transport noise activates the sympathetic nervous system and the hypothalamic–pituitary–adrenal axis, releasing adrenaline, noradrenaline and cortisol. These raise blood pressure and heart rate and, sustained across years, begin to remodel the vessels themselves. Translational and experimental work has traced the intermediate steps with some precision: chronic exposure drives oxidative stress in the vascular wall through activation of the NADPH oxidase and uncoupling of the enzyme that makes nitric oxide, the molecule healthy arteries depend on to relax. What follows is endothelial dysfunction, inflammation and a measurable stiffening of the vessel — the substrate on which cardiovascular disease is built (Münzel et al., 2020; Münzel et al., 2024). None of it requires the sound to be loud enough to annoy. The response runs through the autonomic system rather than the deliberating mind, which is why a sleeper who never consciously registers a passing lorry can accumulate its effects all the same.
Two features of this response make it more dangerous than intuition allows. The first is that it does not fade. Common sense expects the body to acclimatise, to stop reacting once a resident has got used to the road outside. The evidence points the other way. Chronic exposure sustains the reaction rather than exhausting it, and the habituation that does occur is conscious, not autonomic (Münzel et al., 2024, p. 30). A person can cease to notice a sound while their vasculature goes on answering it. The second is that the burden falls hardest where there is least reserve to absorb it: those already carrying cardiovascular disease are among the most susceptible to noise-related harm, so exposure compounds an existing vulnerability rather than spreading itself evenly across the population (Münzel et al., 2024, pp. 25–26). The reassuring folk belief — you get used to it — is, at the level that decides cardiovascular risk, close to exactly wrong.
Sleep is where much of that accumulation happens, and it is the reason the night carries a regulatory penalty of its own. A passing vehicle need not wake a person to leave a trace: it fragments the architecture of sleep, shifting its stages and provoking brief cortical arousals and autonomic surges that the waking mind rarely recalls. The systematic review underpinning the WHO guidelines, pooling some seventy studies, found that transportation noise disturbs both self-reported sleep and objectively recorded sleep physiology in adults, with the odds of a cortical awakening rising by roughly a third for each ten-decibel increase in the noise reaching the ear (Basner & McGuire, 2018, Section 3). Because these responses recur night after night without ever surfacing as a memory, the body integrates a dose the mind never notices. That is the route by which an unremembered sound becomes a cardiovascular one.
3.2 A graded relation with no safe floor
If the mechanism explains how noise reaches the heart, the epidemiology measures how much. What it finds is not a single number so much as the shape of a relationship: as exposure rises, risk rises with it, smoothly and without an obvious step. For road-traffic noise and ischaemic heart disease, the systematic review underpinning the WHO guidelines pooled seven longitudinal studies and returned a relative risk of 1.08 (95% CI 1.01–1.15) for each additional 10 dB of \(L_\text{den}\), rated as high-quality evidence, with the reference category centred near 53 dB (Kempen et al., 2018). Eight per cent per ten decibels sounds modest, and for one individual it is. But the relation is roughly multiplicative, so it compounds with exposure, and it applies across a population numbered in the tens of millions. A small per-person increment becomes a large collective one.
The phrase that has come to summarise this literature is no safe threshold: no level below which the excess risk reliably falls away to nothing. It is an important claim and an easily overstated one. It does not say that a whisper is dangerous. It says that within the range the studies actually observed, the data give no warrant for drawing a line and calling everything beneath it safe. That qualification matters, because the evidence thins at the bottom: the pooled estimate is anchored to a 53 dB reference category, and below it the measured curve runs out. The calculator that follows is built to show this boundary rather than paper over it. It sets a communicative baseline at 45 dB, a plausibly low-exposure reference, but marks the 45–53 dB stretch explicitly as extrapolation, a dashed continuation consistent with the no-threshold finding but not itself measured. The honest reading of the low end is not harm proven down to silence but no floor found within the evidence, and beneath it, assumption.
Source · exposure-response function from Kempen et al. (2018) (IHD) and Münzel et al. (2024) (all-CVD pooled); baseline set at 45 dB with the 45–53 dB segment marked as extrapolation. Accessed 2026-07-11.
What the dial reports is a property of populations, not a verdict on the reader. A relative risk is a ratio of rates between an exposed group and a reference group; a value of 1.08 does not mean an individual at that exposure is eight per cent ill, but that a cohort like them will, over time, accumulate proportionally more disease than an otherwise similar cohort at the reference level. Switch the outcome to the broader all-cardiovascular endpoint and the per-decibel slope is shallower, but its sign and shape are unchanged: more exposure, more disease, no step down to safety (MITECO, 2024). The distance between what the statistic says about a group and what a reader wants it to say about themselves is not a defect in the number. It is a property of the kind of claim a rate can make.
A graded relation with no floor carries an uncomfortable implication for how noise is governed. If risk began only above some threshold, a limit value could in principle separate the safe from the unsafe, and compliance would amount to protection. Without a threshold, a limit value is an administrative decision about how much risk to tolerate, not a biological boundary. And the largest share of the total harm need not sit with the small number of people at extreme exposure. It can sit instead with the very large number living just inside what the rules call acceptable, each carrying a small excess risk that a continent’s worth of people turns into a large total.
3.3 The European burden
Turning a relative risk into a public-health total is the work of a health-impact assessment: combine the exposure–response function with a map of how many people live at each exposure level, and add up the excess cases the function predicts. The product is an estimate, not a register of named victims — but it is how a diffuse, per-person increment is made legible as a collective loss, and the exercise, applied across Europe, is sobering. Scaled to the population, the European Environment Agency attributes to transport noise on the order of 66,000 premature deaths, 50,000 new cardiovascular cases and 23,000 type-2-diabetes cases per year (2021 health-impact assessment), placing noise among the top environmental health burdens on the continent (European Environment Agency, 2025).1
The diabetes figure signals that this is not only a cardiac story. The same stress pathway has metabolic consequences, and the epidemiology now links transportation noise to cardiometabolic outcomes including type-2 diabetes, not merely to the heart and vessels narrowly construed (Münzel et al., 2025). On this evidence noise behaves less like a single-organ hazard and more like a systemic stressor, surfacing wherever a chronically switched-on stress response does its slow damage. That breadth is part of what justifies treating noise as a risk factor in the epidemiological sense, standing alongside the more familiar ones, rather than as a nuisance with a health footnote attached.
None of these totals should be read as precise. Each rests on choices: which exposure–response function to trust, what counterfactual exposure to compare against, how low to keep counting. Reasonable choices yield different numbers, which is why the Agency’s own figures do not all agree, as the footnote records. The value of the estimate lies not in its third significant figure but in its order of magnitude, and that is not in serious doubt: the burden is large, and it has been hiding in plain sight, spread too thinly across too many people to announce itself the way a chemical spill would.
3.4 What the number can and cannot say
A careful reader will ask whether any of this shows that noise causes disease, or only that the two travel together. Association alone would not settle the question; but the case here is stronger than mere association, because the strands converge. There is a plausible and increasingly specified mechanism, running from the ear’s alerting response through stress hormones to vascular damage; there is a graded dose–response that behaves as a causal relation should, more exposure tracking more disease across independent cohorts; and the pooled epidemiology has been graded as high quality by the bodies whose task is to weigh it (MITECO, 2024). Convergence of that kind is what a causal inference is built from in environmental health, where controlled human trials are neither feasible nor ethical. It does not license certainty about any one person. It does license treating noise as a cause at the level of the population — which is the level at which policy acts.
But a population number is only ever as good as the instrument that feeds it, and here the estimate quietly inherits a limit from two chapters back. The burden figures are built on \(L_\text{den}\), and \(L_\text{den}\) is an A-weighted, source-oriented, long-term average — the instrument that, as we have seen, is least sensitive precisely where the low-frequency, tonal and night-time sources concentrate. A harm the metric cannot register cannot enter an assessment the metric defines. So the 66,000 is best read not as a ceiling but as a floor: the toll visible through the particular lens Europe has chosen to look through. Whatever share of the real burden is carried by the sounds that lens under-weights does not appear in the total — not because it was weighed and found small, but because it was never placed on the scale.
The diagnosis stops short of a prescription, and the restraint is deliberate. To say that noise is a measurable cardiovascular risk factor, graded and without a demonstrated safe floor, is a claim about the world; to say what limit a city should set, or what a resident is owed, is a claim about values — and the two should not be smuggled into one another under cover of a statistic. The second kind of claim is taken up in the open later, where it can be argued as what it is. The work here has been only to establish that there is something real to argue about: that the move from nuisance to risk is earned, not asserted.
One step remains before the argument can move from the body to the world it inhabits. A population statistic names a harm but does not locate it. Acting on tens of millions exposed requires knowing which millions, and where — which streets, which façades, which sleeping rooms. That is the work of noise mapping, and its instruments, its coverage and its silences are the subject of the next chapter.
The EEA’s publication landing page headlines a somewhat higher figure (73,000 premature deaths) than its thematic health-impact briefing (66,000). The 2021 HIA estimate is used here; the two figures are not reconciled, and the difference is left visible rather than resolved.↩︎