Do Hearing Aids Slow Cognitive Decline? What ACHIEVE Found

Abstract

Hearing loss is the single largest modifiable risk factor for dementia in the Lancet Commission's accounting. The obvious question is whether treating it helps.

ACHIEVE is the trial that asked. Nine hundred and seventy-seven adults aged 70 to 84 with untreated hearing loss, randomised to hearing aids with audiological support or to a health education control, followed for three years.

The primary outcome was null. A prespecified subgroup showed a 48 per cent reduction in cognitive decline. Which of those two sentences you lead with determines what a reader believes, and the honest answer requires both.

1.The Design

Frank Lin, James Pike, Marilyn Albert and colleagues ran ACHIEVE across four sites in the United States, publishing in The Lancet in 2023.

The control arm is the first thing worth noticing. Participants were not left alone: they received a successful ageing health education programme, matched for contact time and attention. That controls for the effect of being enrolled, visited and taken seriously, which in a three-year trial in older adults is not a small thing.

The second and more consequential design feature is where participants came from. Two hundred and thirty-eight were drawn from the ARIC study — an existing observational cohort of older adults with established cardiovascular risk factors. Seven hundred and thirty-nine were community volunteers recruited de novo.

Those two groups differ in a way that turns out to matter more than anything else in the trial.

2.The Primary Result

Across the whole trial, three-year cognitive change did not differ between the hearing intervention and the control. The difference was 0.002 standard deviation units, at p = 0.96.

That is about as null as a result can be. Not an underpowered near-miss, not a trend in the right direction — a point estimate essentially indistinguishable from zero in a trial of nearly a thousand people over three years.

Any account of this trial that does not begin here is misleading, and a good many accounts of it do not begin here.

3.The Subgroup

The trial prespecified an analysis by recruitment source, and it found a significant interaction at p = 0.010 — meaning the effect of the intervention genuinely differed between the two populations rather than varying by chance.

In the ARIC participants, at higher risk of cognitive decline, hearing intervention reduced three-year cognitive decline by 48 per cent, at p = 0.027. In the de novo community volunteers, at lower risk, there was no difference, at p = 0.18.

The authors' conclusion is stated with corresponding care: hearing intervention may reduce cognitive change over three years in populations of older adults at increased risk for cognitive decline, but not in populations at decreased risk.

Three features make this subgroup finding more credible than most. It was prespecified rather than found afterwards. The interaction test was significant, which is the correct statistical basis for claiming a subgroup difference. And the direction was predictable in advance — an intervention that slows decline can only demonstrate that in people who are declining.

4.What Still Argues for Caution

The ARIC subgroup is 238 people. A 48 per cent reduction estimated in a subgroup of that size, at p = 0.027, is not a settled quantity, and the confidence around it will be wide.

There is also a less comfortable reading of the same data. The de novo volunteers were healthy, motivated people who signed up for a hearing study; their cognition may simply not have declined enough over three years for any intervention to slow it. That is a floor effect, and it makes the trial less a test of who benefits than of who had room to.

Both readings are compatible with the numbers, and the practical implications differ. Under the first, hearing aids help the at-risk specifically. Under the second, they might help anyone given enough time, and three years was too short to show it in healthy volunteers.

The trial cannot distinguish them and neither can we.

5.How This Sits With the Observational Evidence

Our companion article covers the observational side: a meta-analysis of 383,326 people finding hazard ratios of 1.18 for hearing impairment alone and 1.53 for combined hearing and vision loss, with Mendelian randomisation offering borderline support for a causal contribution.

That literature establishes that hearing loss predicts dementia. ACHIEVE tested whether correcting it changes the trajectory, and returned a null overall with a positive signal where decline was actually occurring.

The two are not in conflict. A risk factor can be real and its correction can still fail to help, either because the association was not causal or because the intervention came too late — and ACHIEVE was conducted in people aged 70 to 84, after decades of untreated loss.

What nobody has tested is treating hearing loss at fifty-five and following people for twenty years, which is the trial the observational data actually motivate and which may never be run.

Editorial Comment

MindHeaven® has no product related to hearing, and nothing we sell influences anything in this article.

We wrote it because ACHIEVE is the clearest example we have found of a result whose meaning depends entirely on which sentence gets quoted, and because the temptation to quote the flattering one is exactly the temptation this library is built to resist.

We would want the same treatment applied to a trial of ours: null primary outcome first, subgroup second, and the caveats about subgroup size attached to the subgroup rather than left off.

Practically, the case for treating hearing loss does not depend on dementia at all. Hearing aids improve conversation, participation and quality of life, which are reasons enough. Difficulty following speech in company is worth an audiologist, and the benefit arrives whether or not it does anything for cognition.