Hearing Loss Changes Cognitive Test Scores Before It Changes Cognition

Two claims about hearing and the mind circulate together and are almost always run into one. The first is epidemiological: people with hearing loss are more likely to be diagnosed with dementia later. The second is practical and much less discussed: someone who cannot hear the examiner clearly will score lower on a test administered by voice, whatever the state of their cognition. They are different claims with different evidence and different consequences, and treating them as one is how a hearing problem gets recorded as a cognitive one.

What the dementia evidence actually says

The Lancet Commission on dementia prevention, intervention and care is the standard reference. Its 2020 report identified twelve potentially modifiable risk factors accounting for around 40 per cent of dementia cases worldwide; the 2024 update extended the list to fourteen and the estimate to around 45 per cent. Across both, hearing loss carries the largest single population-attributable fraction of any midlife factor — the share of cases that would in principle not occur if the factor were eliminated across the population.

That framing is worth reading carefully, because a population-attributable fraction is a statement about a population under a counterfactual, not a prediction about a person. It assumes the association is causal, which is precisely the thing under investigation. Several mechanisms have been proposed — degraded input reducing the cognitive stimulation the brain receives, the sustained effort of listening consuming resources that would otherwise go to encoding what was said, social withdrawal following from difficulty in conversation, and shared underlying vascular or neurodegenerative processes producing both. They are not mutually exclusive, and the evidence does not currently separate them.

The trial that tested it directly

Observational associations invite intervention trials, and the important one is ACHIEVE, published in The Lancet in 2023. It randomised older adults to a hearing intervention or a health-education control and followed cognition over three years. The headline result was null: across the whole sample there was no significant difference in the rate of cognitive decline. A prespecified subgroup drawn from an existing cohort study, who were older and at higher cardiovascular and cognitive risk at baseline, did show meaningfully slower decline with the intervention.

• A null overall result does not show that hearing intervention is useless — it shows that the average three-year effect in that sample was not distinguishable from zero.
• A positive prespecified subgroup result is real evidence, but subgroup findings are weaker than whole-sample findings and are the standard way a promising effect later fails to replicate.
• The healthier part of the sample was declining slowly to begin with, leaving little room for an intervention to show a difference over three years.
• The honest reading is that hearing intervention plausibly helps people already at elevated risk, and that three years is a short window for a process measured in decades.

The measurement problem nobody adjusts for

Set the dementia question aside entirely and a second problem remains, one that applies to a 40-year-old with untreated hearing loss as much as to an 80-year-old. A substantial proportion of a standard cognitive battery is delivered orally. Digit span, which is the workhorse measure of working memory — the ability to hold and manipulate information over seconds — is a spoken list of numbers to be repeated. Vocabulary, similarities, comprehension and arithmetic items are typically read aloud. Instructions for the non-verbal subtests are spoken too.

If some of that input arrives degraded, the score drops for a reason that has nothing to do with the ability being measured. The effect is concentrated exactly where an examiner might look for early decline: auditory working memory and verbal comprehension. A profile showing depressed verbal and working-memory indices against intact perceptual reasoning is a recognised pattern in early cognitive impairment. It is also what unaddressed hearing loss produces. The two are not distinguishable from the score sheet alone, which is why the index profile matters here as much as the total — the mechanics of that breakdown are covered in our piece on full-scale scores versus index scores.

Why speech goes indistinct before it goes quiet

The mechanism is more specific than "things get quieter", and the detail explains why cognitive subtests are affected so unevenly. Age-related hearing loss begins at high frequencies and works downward. Vowels carry most of the energy in speech and sit low; consonants carry most of the information and sit high. The practical consequence is that speech stays audible while becoming harder to resolve — the listener hears that something was said and cannot tell whether it was "six" or "thick".

Read that against what an orally administered subtest asks for. Digit span requires exact recall of a spoken sequence, where mishearing one item costs the whole trial. Vocabulary and similarities require the target word to be identified precisely before any reasoning starts. This is also why a quiet room is not a courtesy: background noise degrades consonant discrimination far more for a listener with high-frequency loss than for the examiner sitting opposite them, so the two people in the room genuinely do not share an assessment of whether conditions were adequate.

Scale matters here too. The World Health Organization estimates that more than 1.5 billion people live with some degree of hearing loss, that several hundred million have loss severe enough to benefit from rehabilitation, and that prevalence rises steeply with age — with a substantial minority of people over 60 affected. A risk factor that common, in the age group most likely to be referred for cognitive assessment, is not an edge case to be checked when a result looks strange.

What follows for anyone being assessed

The two problems converge on one procedural point, and it is unglamorous: hearing should be established before a cognitive result is interpreted, not investigated afterwards because the result looked odd. In practice that means a few things:

• Ask whether hearing was screened before the assessment, and whether any aid normally worn was worn during it.
• Expect the examiner to have controlled the room — background noise degrades speech comprehension disproportionately for a listener with hearing loss, well before it affects anyone else in the room.
• Treat a profile with weak auditory working memory and intact visual reasoning as a question rather than an answer, particularly in an older adult who has not had a recent hearing test.
• Check how the test you are taking is administered before you sit it — our explainer on how IQ tests are built and administered sets out which sections are typically spoken.
• Remember that a retest after a hearing aid is fitted is confounded by the practice effect, which raises second scores on familiar formats regardless of any real change — see our note on what a retest actually measures.

For the risk question rather than the measurement one, the position that the evidence supports is moderate. Hearing loss is common, treatment is comparatively cheap and safe, the observational association with later dementia is consistent, and one randomised trial suggests benefit in higher-risk groups without establishing it in general. That is a reasonable basis for treating hearing loss and a poor basis for promising that a hearing aid prevents dementia. The gap between those two statements is where most of the coverage of this topic goes wrong.

The reason this matters beyond audiology is that it is a clean example of a confusion the whole field is prone to: a number produced by a person, an instrument and a room is routinely read as a fact about the person alone. Age, hearing, language, sleep and testing conditions all sit between the ability and the score. Our piece on what changes with age and what does not makes the same point about a different variable. The score is the last step of a chain, and it is worth knowing what else is in the chain before treating it as a measurement of the mind. If you are going to take an assessment, the conditions you take it under are part of the result.

Common questions