Does Treating Hearing Loss Prevent Dementia? The Trials Say No, but Not for the Usual Reason

Precise question

Does treating age-related hearing loss with hearing aids and audiological rehabilitation slow cognitive decline or prevent dementia in older adults?

This is narrower than "is hearing loss associated with dementia," answered by cohorts decades ago, and narrower than "should older adults get hearing aids," which is settled on grounds unrelated to cognition. It asks whether the treatment changes the trajectory, and it requires randomization, because almost everything that makes older adults hard-of-hearing also makes them likelier to develop dementia. Hearing loss tracks age, income, education, vascular disease, social isolation, and access to care; dementia tracks all of those too. An observational study cannot tell whether the ear is a lever or a symptom.

Four labels are used throughout:

• Observation: a measured difference with no causal isolation.
• Quasi-causal estimate: a natural experiment, sibling comparison, instrumental-variable design, or risk-stratified randomization that removes some but not all confounding.
• Causal estimate: a randomized contrast, or a Mendelian randomization standing in for one.
• Extrapolation: extending a result beyond the tested population, mechanism, or outcome.

Why it matters

Hearing loss is the most common chronic sensory condition in the world. The World Health Organization's March 2026 fact sheet reports that over 5% of the global population — 430 million people — already need rehabilitation for disabling hearing loss, that more than 700 million will require it by 2050, and that unaddressed hearing loss costs almost US$1 trillion a year. The 2021 World Report on Hearing put the treatment gap more bluntly: of those who could benefit from a hearing aid, only 17% use one.

Why bring dementia into it? The 2024 Lancet standing commission listed 14 modifiable risk factors and concluded that roughly 45% of dementia might be preventable or delayed. Hearing loss is on that list, and a widely cited summary attributes about 7% of cases to it, third among the listed factors. Few other entries are this cheap to treat relative to their burden, so a null result here needs reading carefully rather than as a verdict.

Best current answer

Treat hearing loss because treating it works, not because it has been shown to prevent dementia. The randomized evidence does not support a broad dementia-prevention claim. It supports something narrower: the cognitive benefit, if it exists, is concentrated in people already at high risk, and the same intervention reliably improves falls, participation, and communication.

What replicated across trials, cohorts, and instruments sat downstream of hearing itself — balance, communication, participation. What did not was the distal outcome, dementia incidence, in a population that mostly lacked the baseline risk to show it.

What the randomized trials found

ACHIEVE: a precise null in a well-calibrated trial

ACHIEVE (NCT03243422) randomized 977 adults aged 70 to 84 with untreated hearing loss and without substantial cognitive impairment, at four US field sites, to a best-practice hearing intervention — audiologic consultation, fitting, hearing aids, and rehabilitation counselling — or to an active health-education control of four sessions on chronic disease and disability prevention. The active control isolates hearing-specific content from the attention and expectations any active intervention delivers. Mean four-frequency pure-tone average was 39.4 dB (SD 6.9); of the 977, 238 came from the Atherosclerosis Risk in Communities (ARIC) cardiovascular cohort and 739 were recruited de novo.

The primary result was a near-perfect null. Over three years, global cognition declined by 0.200 standard deviation units in the intervention group (95% CI −0.256 to −0.144) and 0.202 in control (−0.258 to −0.145). The difference was 0.002 (95% CI −0.077 to 0.081), p = 0.96. Power calculations assumed a three-year change of −0.24 SD units and control-arm decline was −0.202, so the trial was calibrated as intended and found the intervention shifted the trajectory by approximately nothing. It is not a study underpowered into obscurity.

The pre-specified subgroup analysis holds the interesting result. Among the 238 ARIC participants — measurably older, more female, more Black, less educated, and in worse cardiometabolic health — the intervention group showed a 48% reduction in three-year cognitive decline, with an interaction p of 0.010. Among the 739 de novo participants there was no benefit at all.

The intervention worked as an intervention. At year three, participants reported 7.2 hours per day (SD 5.2) of device use, and self-reported communication impairment fell from 15.7 to 7.8 while control rose from 14.9 to 16.2. This disciplines interpretation: the de novo group entered with greater impairment, improved more in absolute terms, and got no benefit.

CHOICE: the null on the outcome that matters most

CHOICE, published in JAMA Neurology in late September 2026, randomized 703 older adults with mild cognitive impairment and coexisting hearing loss in Shanghai to a hearing-aid intervention or sham control, with 24 months of follow-up and 610 (86.8%) completing; mean age 74.5 years (SD 7.2), range 60 to 96. The primary endpoint — conversion to dementia-level impairment — did not differ significantly: 3.09% (95% CI 1.62%–5.74%) with intervention versus 4.74% (2.60%–8.29%) in control, risk difference −2.07 percentage points (−4.74 to 0.61), relative risk 0.59 (0.25–1.38), p = .23. No study-related adverse events were reported.

The point estimate is not nothing: a relative risk of 0.59 is what a modest protective effect looks like, and the interval contains both a 75% reduction and a small increase in harm. A trial of 703 people over 24 months is not powered for conversion rates in the low single digits — an endpoint-sample mismatch dressed as a negative result. The prespecified secondary analysis was striking — cognitive improvement in 15.34% (11.84%–19.62%) versus 2.36% (0.99%–5.22%), relative risk 5.94 (2.46–14.37) — but "improved" versus "declined" is a coarse dichotomy on a continuous scale, and in a sample selected for impairment regression to the mean alone can manufacture an appearance of improvement. It also runs opposite to ACHIEVE, where the group with the most impairment to fix showed no benefit.

The parallel findings that did replicate

Across the same trials, on outcomes they were not powered for, the signal is consistently and statistically clearly positive.

Falls: adjusted mean falls were 1.45 (95% CI 1.28–1.61) with intervention versus 1.98 (1.82–2.15) in control, a 27% reduction, mean difference −0.54 (−0.77 to −0.31). The effect held in both the ARIC and de novo populations — precisely the internal heterogeneity that made the cognitive result ambiguous.

Social participation: adjusted three-year differences favoured the intervention on network size (+1.05, 95% CI 0.01–2.09), diversity (+0.19, 0.02–0.36), embeddedness (+0.27, 0.09–0.44), and loneliness on the UCLA scale (−0.94, −1.78 to −0.11). The authors were appropriately careful: it is unknown whether the network changes are clinically meaningful, and the loneliness change is not.

Biomarkers: in 164 ARIC participants with plasma at baseline and year three, the intervention was associated with a slower three-year rise in glial fibrillary acidic protein, a blood marker of neuroinflammation: −0.026 (95% CI −0.144 to 0.093) versus +0.149 (0.039 to 0.260) in control, difference −0.175 (−0.322 to −0.028), p = 0.019. The neurodegeneration marker neurofilament light chain moved the same way but missed significance (difference −0.159, 95% CI −0.348 to 0.029, p = 0.098). Note the sample: the analysis drew exclusively from the ARIC subsample that showed the cognitive benefit — coherence within a subgroup, not independent confirmation of it.

Two trials that bound the picture

SENSE-Cog randomized 252 people with mild-to-moderate dementia across five European countries to a sensory intervention bundle including hearing aids. Among the 251 analyzed, quality of life after a median 37.7 weeks was identical (DEMQoL 92.8 both arms; adjusted difference 0.18, 95% CI −2.13 to 2.30, p = 0.87), with all intervention-related adverse events grade 1 — a null in people who already have dementia. That is a different question from prevention, but it removes the inference that treating hearing in dementia improves outcomes patients can feel.

A risk-stratified re-analysis of ACHIEVE sharpened the subgroup finding. A predictive model built on 2,692 dementia-free ARIC participants was applied to the 977 trial participants; among those in the top quartile of predicted risk, three-year decline under hearing intervention was 61.6% slower than control (95% CI 33.7%–94.1%). This is a more elegant hypothesis than cohort membership, because it is prospectively derived rather than attached to a recruitment route. It is also a hypothesis: the interval is enormous, it is a secondary analysis of one trial, and it has not been independently replicated.

What the observational record adds

The randomized null sits on a large and genuinely consistent observational literature, which should be interpreted rather than dismissed.

A 2024 BMC Medicine synthesis paired UK Biobank cohort analysis, a meta-analysis of 31 prospective cohorts, and Mendelian randomization. In UK Biobank, mild and severe impairment carried progressively higher risk of all-cause dementia than normal hearing (HR 1.52, 95% CI 1.31–1.77; HR 1.80, 1.36–2.38) and of Alzheimer disease (HR 1.63; 2.18). Across 31 cohorts, single hearing impairment tracked both at HR 1.30 (1.21–1.40), dual sensory impairment higher still. Mendelian randomization returned odds ratios of 1.74 (1.01–2.99) for all-cause dementia, 1.56 (1.09–2.23) for Alzheimer disease, and 1.14 (1.02–1.26) for neurovascular dementia.

A 2018 systematic review in JAMA Otolaryngology–Head & Neck Surgery covering 40 studies in 12 countries found the same gradient — cross-sectional odds of 2.00 (1.39–2.89) for cognitive impairment and 2.42 (1.24–4.72) for dementia, but prospective estimates far smaller, 1.22 (1.09–1.36) and 1.28 (1.02–1.59). Even there the prospective estimate for Alzheimer disease specifically was null (OR 1.69, 95% CI 0.72–4.00). That cross-sectional-to-prospective gradient is itself diagnostic: much of the large observational effect is probably reverse causation — prodromal trouble processing speech in noise, or the same neurodegeneration degrading auditory processing — rather than hearing loss causing dementia.

The largest synthesis specific to hearing aid use rather than hearing loss is a 2026 review in the Journals of Gerontology series B, pooling 46 studies and 231,565 participants. It reports improved global cognition (SMD 0.27, 95% CI 0.08–0.45) and memory (SMD 0.13, 0.09–0.18) among users versus non-users, and lower risk of cognitive impairment (RR 0.84, 0.80–0.88). Two features deserve scepticism: it pools predominantly observational studies, exactly the designs that inflate effect sizes, and its executive-function estimate is negative (SMD −0.12, 95% CI −0.20 to −0.04), which is hard to reconcile with a global-cognition benefit. Point estimates that contradict one another across domains indicate a mixture of real effects and residual confounding.

Mechanisms: three candidates and a fourth rival

The ACHIEVE investigators named the candidate pathways themselves, and they are worth stating as hypotheses: cognitive load, in which a degraded signal forces attention away from memory and onto decoding speech; structural effects on brain integrity, in which reduced auditory input produces measurable cortical change; and reduced social engagement, in which communication barriers shrink participation in cognitively stimulating activity. The evidence is uneven: The social pathway has the clearest experimental support, from the ACHIEVE network analysis against an active control. The cognitive-load pathway is plausible and largely untested, since no trial manipulated listening effort in isolation. The structural pathway is least constrained, because no trial reported imaging.

Against these stands a rival that is not a mechanism of hearing loss at all. Hearing loss marks accumulated vascular injury, socioeconomic disadvantage, and reduced educational opportunity, and dementia is downstream of the same factors. If shared causes dominate, hearing loss is an excellent sentinel — a cheap, early, physically measurable marker of neurological and vascular risk — without being a lever. Under that account a hearing intervention improves communication and balance while leaving the trajectory untouched, close to what both trials found.

What would separate the accounts: if hearing loss is a lever, benefit should appear where baseline risk is high enough for trajectories to separate within a few years, which is what the ACHIEVE ARIC and top-quartile analyses suggest. If it is a sentinel, the association should be steepest at low levels of hearing loss — disproportionate to the sensory deprivation present — with no further gradient at severe levels. The observational literature does not cleanly separate these, though the UK Biobank mild-versus-severe gradient is at least consistent with a dose-response.

Competing explanations that are not mechanisms

Three alternative readings of the ACHIEVE pattern deserve naming, because each predicts something different and none can currently be excluded.

Selection, not moderation. The ARIC subgroup differed at baseline in age, sex, race, education, and cardiometabolic health, so a subgroup defined by recruitment route is partly defined by baseline risk — and participants with faster baseline decline have more room to show a treatment difference. This predicts that a risk-stratified re-analysis should reproduce the effect, and the top-quartile analysis, stratifying on predicted decline rather than recruitment route, found an even larger one. Encouraging for the moderation story, and still a secondary analysis in a single trial.

Measurement contamination. The primary endpoint was an in-person cognitive battery on which hearing-impaired participants may score worse for reasons unrelated to neurodegeneration, an artifact present at baseline and year three. To the extent it inflates decline in both arms, it attenuates a true effect toward the null. This predicts larger effects on hearing-independent outcomes, yet the biomarker result is itself non-significant for neurodegeneration.

Time horizon. Three years of cognition in adults with high MMSE scores is a short instrument for a disease unfolding over decades; ACHIEVE follow-up continues in a separate study (NCT05532657), designed partly to address this. Neither trial was powered for dementia incidence, and neither should be read as having tested prevention.

Strongest counterevidence to the null reading

A synthesis concluding "probably not" owes the same discipline to the evidence against it.

The randomized nulls are not clean zeros. CHOICE's relative risk of 0.59, with an interval reaching 0.25, is not what an ineffective intervention looks like; it is what an underpowered one looks like. ACHIEVE's ARIC subgroup showed 48% and its top predicted-risk quartile 62%, both pre-specified or model-derived rather than dredged. The biomarker analysis found GFAP — a neuroinflammation marker the trial did not select for — moving in the predicted direction at p = 0.019.

But the trials disagree about which subgroup benefits. ACHIEVE's high-risk cohort benefited and its lower-risk cohort did not; CHOICE's entire sample had mild cognitive impairment, a higher-risk group than ACHIEVE's average participant, and its secondary cognitive-improvement signal ran opposite to ACHIEVE's pattern. Both cannot be right in the way they are usually read.

The simplest reading is that there is no contradiction, only an effect too small to detect. If the true effect on cognition is a few hundredths of a standard deviation — roughly what the observational aid-use meta-analysis would suggest at SMD 0.27, most of it likely confounding — then both trials are correctly reporting near-zero, the subgroups are noise, and the biomarker p-value is one of several comparisons that would cross 0.05 by chance.

Evidence quality and limitations

Strong. Falls reduction in ACHIEVE: randomized, active control, consistent across both recruitment populations, biologically coherent. Social network preservation: randomized, significant on four related measures. Absence of study-related adverse events in CHOICE and anything above grade 1 in SENSE-Cog: strong evidence of tolerability.

Moderate. Hearing loss as a risk factor for dementia: consistent across large cohorts, with a dose-response gradient between mild and severe impairment, supported by Mendelian randomization. Its weaknesses are the all-cause dementia MR interval (1.01–2.99), which barely excludes the null, and genetic proxies for a trait with substantial environmental variance.

Weak. Any claim that treating hearing loss prevents dementia: two trials both negative on their primary cognitive endpoint, one underpowered for its own dementia endpoint, the two discordant on subgroups, subgroup effects from a single trial with wide intervals, pooled observational estimates dominated by non-randomized designs, and no trial powered for dementia incidence.

Absent. Cost-effectiveness on dementia-outcome grounds. No published economic evaluation of that pathway was located, and the WHO's US$16-per-dollar return concerns hearing care itself.

Calibrated confidence

• ~95% that treating hearing loss improves communication, participation, and quality of life in older adults with impairment, and is low-risk.
• ~90% that current randomized evidence does not support treating hearing loss as an established method of dementia prevention.
• ~85% that a best-practice hearing intervention reduces falls over about three years by roughly a quarter.
• ~80% that it slows the shrinkage of older adults' social networks relative to an active control.
• ~70% that hearing impairment is causally upstream of at least part of dementia risk, with a modest effect.
• ~55% that hearing intervention slows cognitive decline in adults already at elevated risk of rapid decline. The number most likely to move — and it moves on replication, not on re-analysis of the same trial.
• ~25% that hearing treatment meaningfully reduces dementia incidence at population scale within a decade. Nothing currently supports this; it survives as a hypothesis.

Reversal conditions

Confidence would rise if a pre-registered, adequately powered trial in a population enriched for high predicted risk of decline reported a benefit on global cognition or dementia incidence with imaging or biomarker concordance; if a second independent trial reproduced the ACHIEVE risk-stratified result on a different cohort; if ACHIEVE follow-up (NCT05532657) reported dementia-incidence differences over five to ten years; or if CHOICE reported enough power to say whether its relative risk of 0.59 was signal.

It would fall if a large trial in a low-risk population reported a null on cognition with tight intervals; if the biomarker finding failed to replicate; or if better-adjusted observational work showed the association attenuating under control for shared causes, making hearing loss a sentinel rather than a lever.

Decision implications

For an individual with untreated hearing loss, the rationale for treatment is complete without any dementia argument. Communication, safety, participation, and quality of life are established benefits, the intervention is low-risk, and the cost-effectiveness of hearing care does not depend on the dementia literature resolving in any direction. A clinician may reasonably mention the high-risk subgroup result as an additional reason to treat, provided it is labelled a preliminary finding in one trial's secondary analysis. Individual medical decisions belong with qualified clinicians.

For health systems, adult hearing screening remains, as of this review, an "I" statement from the US Preventive Services Task Force — insufficient evidence to assess the balance of benefits and harms of screening asymptomatic adults aged 50 and older — and that recommendation dates from March 2021 and does not address treatment outcomes. That gap should be stated plainly. The strongest case for expanded screening and coverage rests on direct communication, participation, and safety benefits, with the WHO's cost arithmetic behind it. Citing dementia prevention would overstate the evidence and, if challenged, would damage the stronger case. The commission's inclusion of hearing loss among 14 modifiable factors describes association and plausibility, not a finding that hearing treatment prevents dementia.

For researchers, the decisive missing experiment is specific and affordable relative to the stakes: a pre-registered trial in a pre-enriched high-risk population, powered for dementia incidence rather than three-year cognitive change, with cognitive, fall, participation, imaging, and biomarker outcomes co-registered and an economic evaluation attached. Two smarter targets also exist: enrichment by predicted decline trajectory rather than recruitment route, and a listening-effort manipulation isolating the cognitive-load pathway from device effects.

The most useful thing the last three years established is that a null result on dementia was the right answer to the wrong question. The intervention was effective on several outcomes; the population was too low-risk, the follow-up too short, and the endpoint too distal. Fixing those three things is tractable — waiting for a larger version of the same design is not.

References

1. Lin, F. R., Pike, J. R., Albert, M. S., et al. (2023). Hearing intervention versus health education control to reduce cognitive decline in older adults with hearing loss in the USA (ACHIEVE): A multicentre, randomised controlled trial. The Lancet, 402(10404), 786–797. https://doi.org/10.1016/S0140-6736(23)01406-X

2. Goman, A. M., Tan, N., Pike, J. R., et al. (2025). Effects of hearing intervention on falls in older adults: Findings from a secondary analysis of the ACHIEVE randomised controlled trial. The Lancet Public Health, 10(6), e492–e502. https://doi.org/10.1016/S2468-2667(25)00088-X

3. Reed, N. S., Chen, J., Huang, A. R., et al. (2025). Hearing intervention, social isolation, and loneliness: A secondary analysis of the ACHIEVE randomized clinical trial. JAMA Internal Medicine, 185(7), 797–806. https://doi.org/10.1001/jamainternmed.2025.1140

4. Pike, J. R., Huang, A. R., Reed, N. S., et al. (2025). Cognitive benefits of hearing intervention vary by risk of cognitive decline: A secondary analysis of the ACHIEVE trial. Alzheimer's & Dementia, 21(5), e70156. https://doi.org/10.1002/alz.70156

5. Pike, J. R., Huang, A. R., Reed, N. S., et al. (2026). Effects of hearing intervention on blood-based biomarkers of neuroinflammation and neurodegeneration: A secondary analysis of the ACHIEVE randomised controlled trial. eClinicalMedicine, 99, 104164. https://doi.org/10.1016/j.eclinm.2026.104164

6. Chen, Y., Shi, H., Xiang, M., et al. (2026). Hearing intervention for older adults with mild cognitive impairment: The CHOICE randomized clinical trial. JAMA Neurology. Advance online publication. https://doi.org/10.1001/jamaneurol.2026.3395

7. Leroi, I., Armitage, C. J., Camacho, E. M., et al. (2024). Hearing and vision rehabilitation for people with dementia in five European countries (SENSE-Cog): A randomised controlled trial. The Lancet Healthy Longevity, 5(11), 100625. https://doi.org/10.1016/j.lanhl.2024.07.008

8. Jiang, F., Dong, Q., Wu, S., et al. (2024). Associations between hearing and vision impairments and risk of all-cause and cause-specific dementia: cohort study, meta-analysis and Mendelian randomization. BMC Medicine, 22, 518. https://doi.org/10.1186/s12916-024-03748-7

9. Loughrey, D. G., Kelly, M. E., Kelley, G. A., Brennan, S., Lawlor, B. A. (2018). Association of age-related hearing loss with cognitive function, cognitive impairment, and dementia: A systematic review and meta-analysis. JAMA Otolaryngology–Head & Neck Surgery, 144(7), 115–120. https://doi.org/10.1001/jamaoto.2017.2513

10. Liu, G., Zhang, Z., Yang, Y., et al. (2026). Effects of hearing aid use on memory, executive function, and cognitive impairment among older adults with hearing loss: A systematic review and meta-analysis. The Journals of Gerontology, Series B, 81(5), gbag033. https://doi.org/10.1093/geronb/gbag033

11. Livingston, G., Huntley, J., Liu, K. Y., et al. (2024). Dementia prevention, intervention, and care: 2024 report of the Lancet standing Commission. The Lancet, 404(10452), 572–628. https://doi.org/10.1016/S0140-6736(24)01296-0

12. Wise, J. (2024). Dementia: almost half of cases could be prevented by tackling 14 risk factors. BMJ, 386, q1723. https://doi.org/10.1136/bmj.q1723

13. US Preventive Services Task Force; Krist, A. H., Davidson, K. W., et al. (2021). Screening for hearing loss in older adults: US Preventive Services Task Force recommendation statement. JAMA, 325(10), 1196–1207. https://doi.org/10.1001/jama.2021.2566

14. World Health Organization (2026). Deafness and hearing loss. Fact sheet, 3 March 2026. https://www.who.int/news-room/fact-sheets/detail/deafness-and-hearing-loss

15. World Health Organization (2021). World report on hearing. https://www.who.int/publications/i/item/9789240020481

16. US Preventive Services Task Force. Hearing loss in older adults: Screening. Final Recommendation Statement, 23 March 2021. https://www.uspreventiveservicestaskforce.org/uspstf/recommendation/hearing-loss-in-older-adults-screening

17. ClinicalTrials.gov. NCT03243422, Aging and Cognitive Health Evaluation in Elders (ACHIEVE). https://clinicaltrials.gov/study/NCT03243422