← Longevity & Supplement Guides
The drift you can’t hear
Nobody remembers the year the restaurants got louder.
That is the whole problem in one sentence. Age-related hearing loss does not arrive; it recedes. There is no morning you wake up and notice a missing frequency. Instead the world subtly reorganises itself around the gap: people start mumbling, the television is badly mixed, group dinners become tiring in a way you put down to age, and you quietly stop suggesting the place with the tiled floor. The deficit reports itself as a complaint about the world.
It is also, on paper, one of the most promising entries on the list of things you might change about your own ageing. And the trial built to settle it produced, at first glance, absolutely nothing.
Why hearing loss hides
Presbycusis — the gradual, age-related version — is not a volume knob turning down. It takes the high frequencies first, and it takes them slowly, over decades. Vowels are low and loud. Consonants are high and quiet. The letters that carry the actual information in a word, the s and f and th and k, live in exactly the band that goes first.
Which is why the classic complaint is not "I can't hear you." It is "I can hear you, I just can't understand you." The speech is arriving. The consonants are not. Your brain fills in the gaps from context, which it is superb at, and which is precisely why you do not notice: the gap-filling is invisible from the inside. What you feel instead is effort.
Where speech lives
Consonants sit in the band that goes first
The trial built to settle it
ACHIEVE was the study everyone had been asking for: not another cohort showing that deaf people do worse, but a randomised trial that actually gave some people hearing aids and not others. Nine hundred and seventy-seven adults aged 70 to 84, all with untreated hearing loss, none with substantial cognitive impairment, at four sites across the United States. Half got audiological counselling and hearing aids. Half got a health educator talking about chronic disease. Three years of follow-up, cognition measured every six months.
The result, published in The Lancet in 2023: three-year cognitive change was −0.200 standard deviations in the hearing group and −0.202 in the control group. The difference was 0.002, with a confidence interval running from −0.077 to 0.081. The p-value was 0.96.
A p of 0.96 is not a near miss. It is two lines lying on top of each other.
The number that travelled
And yet the headline that went around the world that summer was that hearing aids slowed cognitive decline by 48 percent. Both things are true, which is the interesting part.
The 977 participants were not one population. They were recruited from two: 238 came from ARIC, a long-running cardiovascular cohort, and 739 were healthy volunteers who answered a call. The ARIC participants were older, carried more risk factors, started with lower cognitive scores, and declined faster over the three years. The volunteers were, in the clinical sense, doing fine. A prespecified analysis tested whether the intervention behaved differently in the two groups, and it did: the interaction was significant at p = 0.010. Within the ARIC group, decline in the hearing arm ran about 48 percent slower.
A subgroup result rescued from a null trial is the most oversold object in clinical research, and you should be reflexively suspicious of one. But this subgroup was specified in advance and the interaction test — the test that asks whether the difference between groups is itself real — cleared the bar. That is the line between a finding and a fishing trip.
A null average can hide a real gradient
A secondary analysis published in Alzheimer's & Dementia in 2025 pushed further and found the effect was not a two-group switch but a slope. The higher a participant's baseline predicted risk of decline, the larger the benefit. In the top quartile of predicted risk, three-year decline in the hearing arm was 61.6 percent slower than control, with a confidence interval from 33.7 to 94.1 percent.
Put plainly: you cannot slow a decline that is not happening. Three quarters of the trial were healthy volunteers whose cognition was barely moving. Average them together with the people who were actually declining and the result is zero — not because nothing happened, but because the arithmetic dissolved it.
This is worth carrying beyond hearing. It is the same structure that turns up whenever an intervention is tested on a population that mostly did not need it, and it is a good reason to read "no effect" as "no average effect" until you have checked who was in the room. We have made the same argument about what biological-age tests can honestly tell you.
Two outcomes, two answers
The same intervention, measured two ways
Seven years later, two answers from the same people
In February 2026, Neurology published an analysis that stretched the question out to seven years. Researchers used data from Australian participants in the ASPREE study to emulate a target trial: roughly 2,777 eligible people, mean age 75, of whom about 664 received a new hearing aid prescription, followed with cognitive assessments every six months.
Cognition, again, did not move. The seven-year mean difference was 0.03 standard deviations, interval −0.14 to 0.21. Nothing.
But the diagnostic endpoints did. Estimated seven-year dementia risk was 5.0 percent among those prescribed hearing aids and 7.5 percent among those not, a risk ratio of 0.67 (0.37 to 0.97). Cognitive impairment ran 36.1 percent against 42.4 percent, a ratio of 0.85 (0.70 to 1.00). And risk tracked inversely with how often people actually wore the things.
The authors are admirably plain about what this is: an observational emulation in which residual confounding cannot be ruled out. The journal's own evidence grade states that hearing aids did not change overall cognitive scores. Both sentences belong in the same paragraph, and usually only one of them makes it into the coverage.
Why a score and a diagnosis can disagree
There is a tidy explanation and an uncomfortable one, and honesty requires both.
The tidy one is statistical. A continuous score averages everybody, including the large majority who were never going to decline much. A diagnosis counts only the tail. An intervention that keeps a modest number of people from crossing a clinical threshold will barely shift a mean while visibly changing a rate. The two measures are asking different questions, and there is no law that they must agree.
The uncomfortable one is that a cognitive assessment is administered out loud. A person who cannot hear the examiner performs worse on it. Fitting them with hearing aids may improve the measurement rather than the mind. Nobody has cleanly separated those two possibilities, and anyone who tells you otherwise is selling something. It is the same caution we apply to what a wearable can and cannot actually measure: the number and the thing are not the same object.
What hearing might have to do with thinking
Three mechanisms are usually proposed, and none is settled. Effortful listening may consume cognitive resources that would otherwise be spent elsewhere. Difficulty in conversation may drive social withdrawal — and social connection carries its own association with how we age. Decades of reduced auditory input may change the brain regions that process it.
The 2024 report of the Lancet standing Commission keeps hearing loss on its list of 14 modifiable risk factors, alongside education, blood pressure, smoking, physical inactivity, social isolation and, newly, untreated vision loss. The Commission's much-quoted figure — that around 45 percent of dementia cases are potentially addressable across the life course — is a population-level estimate built on assumptions about causality. It is not a personal forecast, and it should never be read as one.
What to actually do about it
The honest answer is that there is nothing to swallow here. We sell nothing for hearing, and no supplement addresses it. What exists is a measurement almost nobody has taken.
An audiogram takes about twenty minutes, is widely available, and produces a real curve you can repeat in five years. Most people over fifty have a decade of dental records and no audiogram at all. That asymmetry is strange when you say it out loud, and it is the sort of gap a sensible baseline of biomarkers exists to close. A number you have taken once is a fact; a number you have taken twice is a trend.
Early signs that are behavioural, not auditory
- You follow conversation fine one-to-one, and lose it entirely in a restaurant or a car.
- You have started watching television with subtitles on, and would now find it odd without them.
- Other people have started to seem indistinct — a complaint about them, not about you.
- You ask "what?" less often than you used to, because you have got better at guessing.
- Group settings leave you unusually tired, and you have begun to decline them.
- Someone else has mentioned the volume of your television before you noticed it.
Two further things follow from the evidence rather than from us. Noise exposure is cumulative and the damage does not undo itself, which makes protection a decision you make in your forties for a person you will be in your seventies. And in the seven-year data, frequency of use tracked with outcomes — hearing aids in a drawer are not an intervention. If you already have loss, the variable that mattered was wearing them.
The most defensible position is also the least dramatic. Approach hearing the way you approach vision: correct it because living with it uncorrected is worse, not because a trial has promised you a brain in thirty years. That is the same logic behind keeping your cardiorespiratory fitness up — the near-term case stands on its own, and the long-term case is a bonus nobody can yet guarantee.
The bottom line
The best randomised trial we have showed no average benefit of hearing aids on cognitive decline over three years — and a substantial benefit in the participants who were actually at risk, with the size of that benefit rising in step with their baseline risk. A seven-year observational emulation found no movement in cognitive scores and a lower rate of dementia diagnoses, and its authors will not claim cause. None of this establishes that treating hearing loss protects the brain. All of it suggests that the question is real, that the answer depends enormously on who you are, and that the people who stand to gain are the ones already drifting.
Which leaves a smaller, firmer conclusion. You are probably not going to notice this one yourself, because the whole nature of the thing is that it hides inside an explanation about other people mumbling. So take the measurement. It is twenty minutes, it is a number, and it is the sort of quiet drift that a longevity protocol worth the name is supposed to catch before you do.
