One person in five on Earth has some degree of hearing loss. That is not a figure of speech: the Global Burden of Disease study put it at 1.57 billion people in 2019, or 20.3 percent, and projected 2.45 billion by 2050 — a 56.1 percent increase driven almost entirely by population ageing rather than by any rise in age-standardised risk [1]. It is the commonest sensory disorder there is, and among the most neglected.
Two processes account for most of it. Presbycusis is the gradual, bilateral loss that arrives with age [2] [3]. Noise-induced hearing loss is damage from sound exposure, occupational or recreational; occupational noise alone is estimated to account for 16 percent of disabling adult hearing loss worldwide, over 4 million disability-adjusted life years [4]. The two are not independent — noise exposure early in life accelerates the age-related loss that follows, which is why the animal literature on it is titled Evidence of a Misspent Youth [5].
The cellular target is the hair cell of the cochlea, and in mammals it does not come back. Hair cells are produced before birth and are not replaced, which is why noise and ototoxic damage are permanent [6]. But the modern picture is more unsettling than that, and it is the subject of this review's centrepiece: the cells that fail first are not the hair cells at all, and the standard hearing test cannot see the failure.
What makes hearing loss more than an inconvenience is what follows it. Uncorrected hearing loss is associated with isolation, reduced social activity, a feeling of exclusion and a higher prevalence of depressive symptoms [7] [8], with poorer quality of life on population measures [9] [10] [11], with declining physical function — in the Health ABC cohort, women with moderate or greater impairment had a 31 percent increased risk of incident disability and a 31 percent increased risk of needing nursing care [12] — and with cognitive decline and dementia. Age-related hearing loss is now treated as a modifiable risk factor for dementia [3] [13], which is a strong claim, and this review examines what actually supports it.
And most of it goes untreated. In United States adults aged 70 and over, hearing aids were used by 40.0 percent of those with moderate hearing loss but only 3.4 percent of those with mild loss [14]. Of the people who are fitted with an aid, many stop wearing it: roughly 80 percent of adults aged 55–74 who would benefit from a hearing aid do not use one, and the reasons given by those who were fitted and abandoned them are mundane and fixable — value, fit and comfort, maintenance, attitude, device factors, finance [15].
Centerpiece: what the audiogram measures, and what it misses
The standard test of hearing is pure-tone audiometry: tones are presented at a range of frequencies and the quietest audible level at each is recorded in decibels hearing level (dB HL), a scale referenced to normal young hearing, so 0 dB HL is normal and larger numbers are worse. The results are summarised as the pure-tone average across the speech frequencies — conventionally 0.5, 1, 2 and 4 kHz in the better ear — and graded: normal at 25 dB HL or below, mild at 26–40, moderate or severe above 40 [16] [9]. Disabling hearing loss in adults is defined internationally as above 40 dB HL in the better ear [17].
A note on what the figure does not show. The task that produced this review asked for an audiogram with a presbycusis curve and a noise notch plotted on it. This substrate carries the severity grading, the averaged frequencies, the notch literature and the extended-frequency findings — but it does not carry a measured presbycusis threshold-versus-frequency curve. Drawing one would mean inventing the dB values, so none is drawn. What is plotted instead is the audiogram's frame, which turns out to be the more interesting object.
The left panel is a subtraction nobody performs. The Global Burden of Disease study reports, in one sentence, that of the 1.57 billion people with hearing loss, 403.3 million had moderate or higher loss after adjusting for hearing aid use, and 430.4 million without adjustment [1]. The difference — 27.1 million, or 6.3 percent — is the number of people worldwide whose hearing loss is reclassified below "moderate" because they wear an aid. That is the global treatment gap, sitting inside a prevalence statistic. Ninety-four percent of the world's moderate-or-worse hearing loss is not being moved by the intervention that exists for it.
A second check on the same paper. It gives a count (1.57 billion) and a share (20.3 percent) in separate clauses. Dividing one by the other recovers a world population of 7.73 billion — correct for 2019. Neither figure is presented as a population estimate; the internal consistency is there to be found.
The right panel is the blind spot, in octaves. The pure-tone average is built from four frequencies spanning 0.5 to 4 kHz — exactly three octaves. The region where early damage first becomes measurable is 10–16 kHz [18], which lies entirely above the highest of them: 4 to 16 kHz is a further two octaves, an interval two-thirds as wide as the entire averaged range and never sampled by it.
That is not an abstract complaint. It is what a direct study found. College students were divided into low-risk and high-risk groups by self-reported noise exposure and use of hearing protection. Both groups had normal thresholds at standard audiometric frequencies. The high-risk group showed significant threshold elevation at 10–16 kHz [18]. The audiogram said they were fine. They were not.
The mechanism behind that is the important finding of the last twenty years. In mice, acoustic overexposure causing moderate but completely reversible threshold elevation left the sensory hair cells intact — and caused acute loss of afferent nerve terminals and delayed degeneration of the cochlear nerve [19]. The threshold recovered; the wiring did not. The same synaptic and neural degeneration occurs with age in mice never exposed to loud noise [20]. The synapses between hair cells and cochlear nerve terminals degenerate first, and this primary neural degeneration does not affect hearing thresholds [18]. The condition has a name: hidden hearing loss — hidden precisely because the test cannot see it.
Human evidence for it is indirect but consistent. In people with tinnitus and a normal audiogram, auditory brainstem responses show a significantly reduced wave I — generated by the primary auditory nerve fibres — with normal wave V, generated more centrally: reduced neural output from the cochlea, renormalised further up the pathway [21].
And here is the consequence patients actually report. Twenty-one older adults with normal hearing (aged 60–79, thresholds ≤20 dB HL from 0.125 to 6 kHz) were matched to nine young adults (18–27) on mean audiogram, years of education and performance IQ — three variables held constant by design [22]. The older group still identified speech worse in every background tested [22]. When the audiogram, the education and the IQ are all matched and the difference persists, the difference is not in any of them.
That is the shape of this disease. The complaint is I can hear you, I just can't understand you in a restaurant, and it is a real physical finding that the standard test is not built to detect. Listening under those conditions costs effort, and there is a formal framework for treating that effort as a limited resource being spent [23].
Pillar 1: measurement and diagnosis
The audiogram, and reading it properly
Beyond the pure-tone average, the audiogram distinguishes sensorineural loss (cochlea or auditory nerve) from conductive loss (outer or middle ear failing to deliver sound to a working cochlea) by comparing air-conduction thresholds against bone conduction, which bypasses the middle ear. The distinction decides everything downstream: conductive losses often have a mechanical fix, sensorineural ones generally do not. Objective measures of middle-ear function such as ear-canal reflectance and umbo velocity have been compared directly in conductive loss [24], and conductive loss is not benign for central processing either — it produces a reversible binaural hearing impairment [25].
Dead regions are a specific trap. Where inner hair cells have lost function completely over a stretch of the cochlea, the audiogram still records thresholds there — detected through neighbouring regions — and amplifying into a dead region can make things worse rather than better. Their diagnosis, perceptual consequences and implications for hearing-aid fitting are set out in detail [26].
The notch, and a caution about it
The classic sign of noise damage is a notch at 4 kHz: a localised dip in thresholds with recovery at higher frequencies [27]. It is real, and it is less useful than it looks. In 357 electrical transmission workers, assessors identified a notch at some frequency in 175 (49 percent). Only 15 (4 percent) had a notch specifically at 4 kHz, and only that small group showed the expected associations with noise exposure — a significant odds ratio of 4.25 for firearms. The much larger group with 6 kHz notches (110, 31 percent) showed no such association [27]. The conclusion the authors drew is worth stating plainly: diagnosing noise-induced hearing loss requires a detailed exposure history, the 4 kHz notch may help confirm it, and the 6 kHz notch is variable and of limited importance. A sign present in 31 percent of a population and unassociated with the exposure it supposedly indicates is not a diagnostic test.
Speech in noise
Because the audiogram misses what patients complain about, speech testing matters. Speech recognition in noise is degraded in presbycusis beyond what thresholds predict, with possible neural mechanisms [28], and both auditory and cognitive factors contribute in older listeners with normal hearing [29] [30]. The Abbreviated Profile of Hearing Aid Benefit is a validated self-report inventory for documenting the outcome of a fitting across everyday listening situations [31], and self-reported disability of this kind captures things the audiogram does not.
Screening
Universal neonatal hearing screening is established practice, and the evidence for early detection and intervention, the possibility of harm, and cost have all been examined [32] [33]. It has a documented limit. In the United Kingdom, ascertainment of permanent childhood hearing impairment (>40 dB HL) across birth cohorts found prevalence rising from 1.07 per 1000 at age 3 to 2.05 per 1000 by ages 9–16 after adjustment for under-ascertainment. Against neonatal screening yields close to 1 per 1000, that means 50–90 percent more children are diagnosed by age 9 than the newborn screen finds [34] — because a substantial share of childhood hearing loss is progressive or late-onset, so a clean newborn screen is not a permanent clearance.
In adults, screening for handicapping hearing loss in the elderly has an established approach [35], and smartphone audiometry has been validated against conventional equipment: in 1,070 school-age children, smartphone screening showed equivalent sensitivity (75.0 percent) and specificity (98.5 percent) to conventional screening audiometry [36]. Community health workers have run it during routine home visits in an underserved community [37], and mobile devices have been evaluated as screening tools where hearing services are scarce [38].
Causes that need naming, not just grading
An audiogram grades severity; it does not give a cause. Several diagnoses change management:
Otitis media is a leading cause of preventable hearing loss, particularly in developing countries, with global incidence and sequelae quantified within the Global Burden of Disease framework [39]. Chronic suppurative otitis media has its own pathogenesis and treatment [40], bacterial biofilms have been directly detected on the middle-ear mucosa of children with chronic disease [41], and inflammatory mediators drive the sequelae [42].
Sudden sensorineural hearing loss is an emergency. Its cause is unknown; proposed mechanisms include vascular occlusion, membrane breaks and viral cochleitis, and temporal-bone histopathology in 17 affected ears found predominantly loss of hair cells and supporting cells of the organ of Corti [43]. It has an international consensus on treatment [44], and irreversible cases following COVID-19 have been reported [45].
Ménière's disease has formal diagnostic criteria agreed across five international societies: episodic vertigo lasting 20 minutes to 12 hours, with low- to medium-frequency sensorineural hearing loss and fluctuating aural symptoms in the affected ear [46].
Auditory neuropathy is the diagnosis the audiogram most misleads on. Ten patients had preserved otoacoustic emissions and cochlear microphonics — normal outer hair cell function — with absent or severely distorted auditory brainstem potentials. Pure-tone thresholds were only mildly to moderately elevated, far less than the functional impairment warranted [47]. Its commonest recessive non-syndromic form is caused by mutations in otoferlin (OTOF) [48] [49] — which matters greatly for what comes later in this review.
Ototoxic drugs are covered under management below. Genetic causes are numerous: large-scale mouse screens have revealed an extensive and still largely unexplored genetic landscape for auditory dysfunction [50] [51] [52], and the architecture of human deafness genetics has been reviewed alongside its therapeutic implications [53].
Pillar 2: management
Hearing aids
Amplification is the primary intervention, and the randomised evidence is thinner than the ubiquity of the devices suggests. A Cochrane review of hearing aids for mild to moderate loss in adults found five randomised trials, 825 participants, published between 1987 and 2017, with a large beneficial effect on hearing-specific health-related quality of life [54]. Five trials is a small evidence base for the standard treatment of the commonest sensory disorder in the world. Modern devices are sophisticated signal processors — directional microphones, noise reduction, feedback cancellation, wide dynamic range compression [55] — and algorithmic improvement of speech recognition in noise for hearing-impaired listeners is an active engineering problem [56].
The gap between fitting and use is the practical problem, and it has been mapped [15]. Tele-audiology addresses part of it: remote screening, diagnostic testing, intervention and rehabilitation can each be completed reliably, with improved access, higher follow-up rates and reduced travel time and cost, against real constraints on connectivity, ambient noise control, licensure and reimbursement [57].
A gap worth stating. Over-the-counter hearing aids are the most consequential recent change in access to amplification, and this substrate contains no work on them — no trial, no regulatory analysis, no uptake study. The review cannot characterise their effect, and does not.
Cochlear implants
For severe to profound loss, an implant bypasses the hair cells and stimulates the cochlear nerve directly. In a scoping review of 201 articles, with efficacy data synthesised from the 102 studies with more than 10 participants, average word perception improved from 8.2 percent to 53.9 percent after implantation, self-reported benefit improved by 21.5 percentage points, and at the individual level 82.0 percent of adults with postlingual hearing loss improved speech perception by at least 15 percentage points, against 53.4 percent of adults with prelingual loss [58]. That postlingual-versus-prelingual split is the central fact about implants in adults: the device delivers a degraded signal that an auditory system with prior linguistic experience can learn to interpret, and one without that experience finds much harder.
Outcomes vary widely between recipients, and the sources of that variance have been examined repeatedly — duration of deafness, biographic and audiologic factors, electrode position within the cochlea and cognitive ability [59] [60] [61], electrode design and scalar location [62] [63], and resting cortical activity before surgery [64]. Preserving residual low-frequency hearing during implantation allows combined acoustic and electric stimulation, which improves recognition of speech, voice and melodies [65] [66], and the meta-analysed rate of hearing preservation is established [67]; minimising surgical trauma is the means [68].
In children the case is developmental. Language development in profoundly deaf children with implants [69], communication development in those implanted by 12 months of age [70], and spoken English development among native signing children [71] have each been studied, along with working memory and verbal rehearsal speed after implantation [72]. In older adults, quality-of-life benefit is documented in patients aged 70 and over [73], and whether implantation improves neurocognition in the ageing population has been asked directly [74]. A systematic review with an economic model examined cost-effectiveness for unilateral and bilateral implantation in children and adults [75].
Access is uneven and the constraint is money, not medicine. An international survey across 17 countries found unilateral implants for adults and children, and bilateral implants for children, covered by national funding in about 60 percent of countries — but bilateral implants for adults nationally funded in only 22 percent, with 37 percent using medical insurance and 41 percent self-funded [76].
Bone conduction, and fixing what can be fixed
Where the cochlea works and the conductive pathway does not, bone-conduction devices transmit vibration through the skull. The range now spans softband and passive transcutaneous skin-drive devices, percutaneous bone-anchored aids, and active transcutaneous implants with the transducer under intact skin [77]. Conductive causes — wax, effusion, perforation, ossicular disease — should be identified and treated before anything else is considered.
Ototoxicity: the avoidable share
Two drug classes cause a large and predictable amount of permanent hearing loss.
Platinum chemotherapy. Pooled prevalence of objectively measured ototoxic hearing loss after cisplatin and/or carboplatin was 43.17 percent across 87 records and 5,077 individuals — 49.21 percent for cisplatin alone, 56.05 percent for both agents, and 13.47 percent for carboplatin alone. Scaled to global treatment rates, roughly one million people are exposed annually, implying almost half a million cases of hearing loss per year [78]. In children this has been repeatedly described as underestimated despite lifelong consequences [79], and part of the reason is physical: cisplatin is retained in the cochlea indefinitely after chemotherapy [80]. The mechanisms — reactive oxygen species generation via NADPH oxidases, apoptotic pathways — are well characterised [81] [82] [83].
There is now a randomised protective intervention. In children with standard-risk hepatoblastoma, sodium thiosulfate given after cisplatin reduced the incidence and severity of ototoxicity [84]; a separate randomised phase 3 trial in children with cancer tested the same agent against observation [85]. Amifostine has been tested in children with average-risk medulloblastoma [86], and proton radiotherapy shows low early ototoxicity rates in paediatric medulloblastoma [87]. Prospective group-wide ototoxicity assessment during cisplatin treatment has been formalised [88], and risk factors have been analysed in adults with testicular cancer [89].
Aminoglycoside antibiotics. The mechanisms and targets for hair-cell protection are established [90] [91] [92]: entry requires functional hair cell mechanotransducer channels [93], and mitochondrial calcium uptake underlies the reactive oxygen species generation that kills the cell [94]. Audiologic monitoring during long-term aminoglycoside treatment for multi-drug-resistant tuberculosis is documented [95], and N-acetylcysteine ameliorated gentamicin ototoxicity in haemodialysis patients [96]. Candidate protective agents include a peptide inhibitor of c-Jun N-terminal kinase, which protected against both aminoglycoside and acoustic trauma [97], though caspase inhibitors rather than JNK inhibition prevented cisplatin-induced loss in one comparison [98] — the pathways are not interchangeable.
Ototoxicity remains a challenge in diagnosis and treatment [99] [100] [101], and the practical point for patients is narrow and useful: when these drugs are unavoidable, baseline and serial audiometry make the damage visible while there is still a choice about dose.
Noise: the only genuinely preventable share
Occupational noise accounts for an estimated 16 percent of disabling adult hearing loss, and the authors are explicit that the majority of that burden can be minimised by engineering controls that reduce noise at its source [4] — controls at the source, ahead of protective equipment worn by individuals. The effects of noise on speech recognition have been characterised in service members [102], and the animal evidence that early exposure accelerates later age-related loss [5] gives protection in youth a value that is not realised for decades.
Hearing loss and cognition: what is actually established
This is the claim doing the most work in public messaging, so it deserves care.
The association is robust. A systematic review of 17 studies found all of them reporting an association between hearing loss and dementia or cognitive decline [103]. A meta-analysis of 36 unique studies with about 20,264 participants found a small but significant association across all cognitive domains, with pooled cross-sectional odds ratios of 2.00 for cognitive impairment and 2.42 for dementia [104]. In the Health ABC cohort, moderate or severe audiometric impairment was associated with a hazard ratio of 1.55 for incident dementia over 9 years after adjustment [16]. The association also appears in nationally representative data [105], in the English Longitudinal Study of Ageing [106], and with brain atrophy and tau pathology [107]. Combined vision and hearing impairment carries its own association with dementia [108].
The mechanisms proposed are plausible and multiple: the cognitive cost of effortful listening diverting resources from other processing [23], withdrawal from difficult listening situations producing social isolation and depression [8] [109], and deafferentation-driven changes in central auditory pathways and frontal atrophy [110] [111] [112]. Cortical neuroplasticity has been documented in early-stage mild-to-moderate loss [113], and central auditory dysfunction may precede clinical dementia in probable Alzheimer's disease [114] — which raises the reverse direction, that early neurodegeneration impairs central auditory processing rather than the other way round. Executive dysfunction and presbycusis co-occur in older people with and without memory loss [115]. That both peripheral and central hearing impairment relate to cognition [116], and that cognitive and health factors predict subsequent decline in hearing acuity [117], means the causal arrow is not simply established by direction of measurement.
Whether treating hearing loss changes cognitive outcomes is a different question, and the evidence is weaker. Observational work associates hearing-aid use with better cognition independently of social isolation and depression [109], and an 18-month before-and-after study found improved executive function with stability across the rest of the battery [13] — but that design has no control group. Against it, a cohort followed over 11 years found the long-term picture for hearing-aid use across mental health, social engagement, cognition, physical health and mortality considerably more mixed [118]. The Health ABC authors said what was needed explicitly: randomised trials to determine whether treating hearing loss postpones dementia onset [16].
That trial exists. ACHIEVE — hearing intervention versus health education control to reduce cognitive decline in older adults with hearing loss — is in this substrate [119]. Its abstract is not available in OpenAlex, only its title. This review will therefore not quote its results, and readers should go to the trial report itself. What the title alone establishes is that the right study was done: a randomised comparison of hearing intervention against an active control, with cognitive decline as the outcome. Until its findings are read directly, the honest summary is that hearing loss is a well-established risk marker for dementia, the mechanisms are plausible, and the strength of the treatment effect is the open question.
The broader life-course framing is available [17] [120] [121] [122].
Tinnitus
Tinnitus commonly accompanies hearing loss and often accompanies a normal audiogram [21], and hyperactivity in the auditory brainstem after acoustic trauma has been observed directly [123] [124]. Amplification helps where hearing loss co-exists [125]; cognitive behavioural therapy has been assessed in a Cochrane review [126].
Pillar 3: progress, and what would change the disease
Gene therapy has moved into patients
This is the development that has changed most recently, and it is genuine. Deafness caused by OTOF mutations — DFNB9, the otoferlin form of auditory neuropathy [48] [49] — is an unusually good first target: the hair cells are intact and the defect is in a single protein required for synaptic transmission. Because the OTOF coding sequence is too large for one adeno-associated virus, a dual-AAV approach splitting the cDNA across two vectors was developed and restored fast exocytosis and partially rescued auditory function in knock-out mice [127]. Improved vectors for the inner ear followed [128] [129] [130], and TMC1 gene therapy restored hearing and balance in mice with a different genetic disorder [131].
It then went into people. AAV-mediated OTOF gene therapy restored hearing in patients with DFNB9: after validation in mice and non-human primates to set dose, volume and route, the vector was delivered into one cochlea of a 5-year-old and both cochleae of an 8-year-old, with clear improvement on auditory brainstem response and pure-tone audiometry — the 5-year-old's injected ear reaching the normal range at one month [132]. A subsequent single-arm trial of bilateral administration in five paediatric patients reported no dose-limiting toxicity and no serious adverse events across 36 adverse events (most commonly raised lymphocyte counts and cholesterol), with hearing restored bilaterally in all patients from average auditory brainstem response thresholds above 95 dB at baseline [133].
That is a small number of children with one rare genetic form of deafness. It is also the first time hearing has been restored by correcting the molecular defect, and the field's own framing runs from genetic architecture to therapy [53].
Hair-cell regeneration: forty years of a solved problem in the wrong species
Birds regenerate hair cells. After gentamicin toxicity in chicks, hair-cell counts fell 36 percent and then progressively recovered [134]; after acoustic trauma, injured sensory cells in the mature chicken cochlea are replaced by mitosis of supporting cells that do not divide without injury [135]. Both findings are from 1987–1988. Mammals do not do this — with one exception: the neonatal mouse cochlea shows spontaneous regeneration, averaging over 120 regenerated hair cells per cochlea after ablation at birth, arising from neighbouring supporting cells by both mitotic regeneration and direct transdifferentiation, and that capacity is lost with maturity [6].
Two routes have been pursued to reopen it. Atoh1 (Math1) forced expression generates new hair cells in mature guinea pigs in vivo [136] and improved hearing in deaf mammals [137], and its role in sensory mosaic formation and proliferation is characterised [138] [139] [140]. Notch inhibition releases supporting cells from lateral inhibition: it induces hair-cell formation in neonate mouse cochleas [141], nonmitotic regeneration in adult mouse utricles [142], mitotically generated hair cells via Wnt activation [143], and — the result that matters most — cochlear hair cell regeneration and recovery of hearing after acoustic trauma [144]. The Notch ligands controlling hair-cell development are known [145], its role in regeneration is established in the zebrafish lateral line [146], and single-cell RNA sequencing has resolved distinct stem-cell populations driving regeneration [147]. Age-dependent conversion of cochlear pillar and Deiters' cells to immature hair cells shows the window closing with maturity [148], as does hair-cell replacement in adult mouse utricles after targeted ablation [149]. Spontaneous regeneration after gentamicin has been documented in the adult mouse utricle [150], and supporting cells were identified as the progenitors decades ago [151]. The zebrafish lateral line now serves as an ototoxicity screening platform [152].
The honest position: regeneration works in birds, in the neonatal mammalian cochlea, and in mammalian vestibular epithelium; it has been induced in the adult mammalian cochlea in the laboratory with recovery of hearing; and it has not been shown to restore hearing in a human being.
Hidden hearing loss needs a clinical test
The synaptopathy story [19] [20] [18] currently has no routine diagnostic. Reduced wave I amplitude is the leading candidate [21], but auditory brainstem response wave characteristics vary substantially between individuals [153], stimulus design affects the measurement [154], and central compensation can restore response magnitude in the young but not the old [155] — so the peripheral deficit can be masked by exactly the process it triggers. Until a test exists, an entire category of hearing damage has no diagnosis, which means no prevalence estimate, no trial endpoint and no treatment.
Extending amplification to what patients actually need
The unmet need is not loudness, it is intelligibility in noise. Signal-processing improvement toward that goal is under way [56] [55], and there are hints from an unexpected direction: musical experience is associated with better hearing of speech in noise in the ageing auditory system [156].
Access
The Global Burden of Disease analysis found that the Healthcare Access and Quality Index explained 65.8 percent of the variation in national age-standardised years lived with disability from hearing loss [1]. Two-thirds of the international variation in disability from this condition is explained by health-system quality, not by biology. Combined with the 6.3 percent aid-adjustment figure from the same paper and the 41 percent self-funding rate for adult bilateral implants [76], the pattern is consistent: the largest available gains in hearing health are in delivery. Smartphone-based screening in primary care [37] [36] [38] and tele-audiology [57] are attempts to close it. United States trends over 1990–2019, including hearing-aid utilisation, have been described within the same framework [157].
Dig deeper in lmmol
Alzheimer's disease is the destination of the risk-factor claim examined above, and reading the two together is instructive: hearing loss is one of the strongest and most consistently replicated modifiable associations in the dementia literature [104] [16], while the causal direction is genuinely unsettled, with central auditory dysfunction sometimes preceding clinical dementia [114]. Depression is the other end of the psychosocial pathway, and the mechanism proposed is behavioural before it is neural — withdrawal from situations where hearing is difficult, then isolation, then mood [8] [7]. Glaucoma and macular degeneration are the sensory analogues and share this review's structural problem exactly: a standard clinical measurement that misses early disease, and damage that has already occurred by the time the test detects it. Concurrent vision and hearing impairment carries its own dementia association [108]. Osteoarthritis shares a different structural problem — the dissociation between what the objective test grades and what the patient experiences, which here is the normal audiogram with real speech difficulty [22]. Obstructive sleep apnea is the other condition where a device that works is routinely abandoned by the people prescribed it, and the reasons are similarly practical [15]. Stroke is worth reading against sudden sensorineural hearing loss: both are acute neurological emergencies where treatment delay costs function permanently, and one of them is widely recognised as such [43]. The full collection is at health.