Technology for Longer Lives

Q: Do companion robots help people with dementia?

Not on validated instruments. The decisive trial randomised 415 people with dementia across 28 facilities to the PARO robotic seal, an identical look-alike plush toy with the robotics switched off, or usual care. PARO beat the plush toy on verbal and visual engagement only; on the Cohen-Mansfield Agitation Inventory — Short Form there was no difference between groups. The trial's own economic analysis found the plush toy offered marginally greater value for money per point of agitation averted, though both were cheap relative to staff-delivered psychosocial programmes. A 2024 umbrella review of 35 systematic reviews found no pooled effect on quality of life, anxiety or depression, and no effect on agitation, neuropsychiatric symptoms or medication use. If a robot brings someone pleasure, that is a real and reasonable thing to buy. It is not a clinical intervention.

Q: Will a fall-detection watch or pendant keep my parent safer?

It may shorten a long lie — the time spent on the floor before help arrives — and that is genuinely worth something. There is no randomised evidence that it prevents falls, prevents injury, reduces hospitalisation or preserves independence. The published literature on fall-detection wearables is overwhelmingly algorithm-validation work: how well the device spots a fall, not whether wearing one changes anything. Targeted searches for randomised trials with health outcomes returned nothing. The first adequately powered trial, INES in Germany, is randomising roughly 1,876 people aged 70+ living alone with days in hospital as the primary outcome. Until it reports, buy a detector for peace of mind and reduced time on the floor — and spend the rest of the budget on balance training.

Q: Do hearing aids prevent dementia?

The randomised trial says no, with an important caveat. ACHIEVE followed 977 adults aged 70–84 for three years; the primary cognition analysis was null, with a difference of 0.002 standard deviations and a p-value of 0.96. A prespecified analysis then found the effect differed by recruitment cohort (p=0.010), and a follow-up analysis showed 61.6% slower decline among participants in the top quartile of predicted risk. That is a subgroup, not a trial result, and the widely circulated percentage describes the subgroup. Separately, a much-cited 2023 UK Biobank paper reporting a hearing-aid benefit was retracted in December 2023 after a coding error swapped the exposure groups. The best current observational evidence — 61,089 adults across 33 countries — finds a modest association (HR 0.91) that is confined to people reporting effective hearing improvement. Get hearing aids for hearing, communication and falls. Do not buy them as a dementia drug.

Q: Does cataract surgery prevent falls?

It restores vision, function, confidence, mood and activity — that part is not in dispute and it is why cataract surgery sits second on our list. The falls claim is genuinely contested. One randomised trial of 306 women over 70 found expedited first-eye surgery cut the rate of falling by 34% using fall diaries. The equivalent trial in second eyes found a 32% reduction that was not statistically significant. And a February 2026 propensity-weighted analysis of 940,233 Medicare enrollees found no significant difference in one-year falls or fractures. The claims analysis may miss self-reported falls that never reach the health system, which is exactly what the diaries captured. Honest wording: restores vision and function, with a plausible but contested effect on falls.

Q: Do brain-training apps work?

They reliably make you better at brain-training games, and produce small improvements on similar laboratory tasks. Evidence that they preserve everyday function or prevent dementia is weak. Three independent meta-analyses spanning 2014 to 2020 show effects shrinking steadily as you move from the trained task, to similar tasks, to different domains, to real life — and after correcting for publication bias, the training effect on fluid intelligence was non-significant. The strongest-looking results come from trials whose comparator was doing nothing. The largest and longest randomised trial, ACTIVE, found reasoning and speed training maintained effects on the specific ability trained over ten years, but its everyday-function benefit was self-reported against a no-contact control, which cannot separate a real effect from expectancy.

Q: Is there an FDA-approved app or device for memory loss in older adults?

No. Searches of the FDA 510(k) and De Novo databases return zero records for dementia and zero for Alzheimer's at any date. EndeavorRx, the device most often cited in this context, is a Class II digital therapeutic cleared for attention deficit hyperactivity disorder in children and adolescents — not for older adults and not for cognitive decline. There is also no modern FDA-cleared fall-risk device: a device-name search for “fall” returns three records, all from 1979 to 1984, so consumer fall detection is not cleared as a fall-risk device. A cleared prescription digital therapeutic has a specific indication, an age range and a public regulatory pathway. A general-wellness brain-training app has none of these and has never been assessed by FDA for efficacy. For cognition in older adults, that second category is currently the only category that exists.

Q: Didn't Japan solve its care shortage with robots?

No, and the real story is more interesting. The peer-reviewed August 2026 analysis of roughly 860 Japanese nursing homes, using prefecture-level subsidy variation as an instrument, found robot adoption associated with 28% more care workers, 39% more nurses and about 26% higher total employment — growth occurring entirely among nonregular staff — alongside fewer retention problems. An earlier panel of 265 homes gave a smaller estimate of roughly 5.8%. Both are observational and their authors say so; report both and average neither. What Japan actually deploys is monitoring: 30.0% penetration for monitoring and communication against 1.2% for mobility support and 0.5% for toileting support, and task-automation indexes of only 1.2 to 1.7 — humans still do the bulk of every task. Also note that 26.7% of 4,688 providers had adopted at least one care robot, a figure the authors caution measures purchase, not sustained use.

Q: Do smart pill bottles and medication reminders improve health?

The reminder hardware question is settled and the answer is no. REMIND randomised 53,480 already-nonadherent people to a pill bottle strip, a digital timer cap, a standard pillbox or nothing, and found odds ratios versus control of 1.03, 1.00 and 0.94 — all null. Electronic pill bottles do better on the thing they measure and no better on the thing that matters: in a randomised trial of patients with uncontrolled hypertension, they produced good measured adherence and left systolic blood pressure indistinguishable from usual care. Cochrane's review of 182 trials found that what worked, weakly, was people: complex, tailored, ongoing support from allied health professionals, most often pharmacists. There is also a real risk in the other direction — for an over-prescribed older person, the highest-value medication intervention is often deprescribing, not tighter adherence to an over-long list.

Q: Does a smart home with sensors help someone stay independent longer?

There is currently no randomised evidence that it does. A systematic review of in-home localisation technologies screened 1,935 articles and found just 10 that reported any patient outcome at all — the field is overwhelmingly engineering, not clinical research. Those studies showed that in-home sensors could not differentiate healthy cognition from mild cognitive impairment, though frailty was detectable. The largest randomised test of the deployed version of this idea — telecare across 2,600 people and 217 English general practices — found no reduction in hospital admission (OR 0.90, p=0.211) and nothing on any secondary service-use metric, and the parallel telehealth arm cost about £92,000 per quality-adjusted life year. There is also a structural trade-off: the systems with the best acceptability are the least informative, and the most informative are the most invasive.

Q: If I can only do one thing, what should it be?

Balance and functional exercise, delivered or supervised by a physiotherapist. It is the only intervention on this page with high-certainty evidence pooled across 108 randomised trials and 23,407 participants, and delivery by a health professional produced larger effects than self-directed programmes. After that, in order: get cataracts assessed, get hearing tested and get aids fitted properly and actually worn, have a pharmacist review the medicines list with deprescribing on the table, and — only if the person is already at elevated fall risk — get a home hazard assessment, because in unselected people it does nothing. If you are choosing between a £1,000 gadget and a course of supervised exercise plus a hearing test, the evidence is not close.

What robots, sensors and apps have actually been shown to do for independence in later life — and what still works better.

Reviewed by CureMed LabsUpdated September 2026

An older woman holding a stair handrail during a supervised balance exercise session, a physiotherapist standing beside her. — The best-evidenced independence technology of the last thirty years is a handrail, a hearing aid and a physiotherapist. Balance and functional exercise has high-certainty evidence from 108 randomised trials; no consumer device in this section comes close.

Simply put

The things that best help older people stay in their own homes are old and unglamorous: strength and balance exercise, cataract surgery, hearing aids that are actually worn, and safer homes for people already at risk of falling. The robots, alarms, pill dispensers and monitoring gadgets sold for ageing have mostly never been shown to change whether anyone stays independent.

The short answer

Across every domain of ageing technology, the same pattern repeats: the device performs well on its own instrument and disappears when a validated clinical endpoint is used. The interventions with high-certainty randomised evidence for preserving independence are balance and functional exercise (23% fewer falls, 108 trials, 23,407 participants), cataract surgery, properly fitted hearing aids (27% fewer falls over three years), and home hazard reduction targeted at people already at elevated risk. The interventions with the most marketing — companion robots, fall-detection wearables, smart-home sensing, brain-training apps and reminder hardware — have either null results in the largest trials ever run on them, or no outcome trials at all. In the single cleanest demonstration, a robotic seal was tested against a look-alike plush toy with the robotics switched off; it beat the toy only on engagement, showed no difference on the validated agitation instrument, and the toy offered marginally better value for money.

Exercise is the only intervention here with high-certainty Cochrane evidence: 23% fewer falls across 108 randomised trials and 23,407 community-dwelling participants (PMID 30703272), with larger effects when a health professional delivers it.
The ACHIEVE hearing trial's primary cognition result was null — a difference of 0.002 SD, p=0.96 — with a prespecified cohort interaction (p=0.010) and 61.6% slower decline confined to the highest-risk quartile; the widely quoted benefit describes a subgroup, not the trial.
A 415-participant cluster-randomised trial compared the PARO robotic seal with a switched-off look-alike plush toy: the robot won only on engagement, showed no difference on the validated CMAI-SF, and the toy was marginally better value per point of agitation averted.
No FDA-cleared digital therapeutic exists for cognition in older adults, and no FDA-cleared fall-risk device exists; EndeavorRx is cleared for ADHD in children and adolescents, and consumer fall detection is not cleared as a fall-risk device.
Japan did not replace carers with robots: the peer-reviewed August 2026 analysis found robot adoption associated with 28% more care workers, and Japan's own deployment automates only a task index of 1.2–1.7 — humans still do the bulk of every task.

This section grades technology for ageing by what it has been shown to change — independence, hospitalisation, falls, mortality — rather than by what it can measure or how well it demos. That distinction turns out to do almost all the work.

The reader arriving here expects robots. The honest answer is that the interventions with the strongest randomised evidence for keeping older people in their own homes are decades old, cheap relative to their effect, thoroughly boring, and badly under-delivered. Meanwhile the US age-adjusted fall death rate rose 21% between 2018 and 2024, from 64.7 to 78.4 per 100,000 older adults, in the years after Cochrane published high-certainty evidence on exactly what prevents falls. The knowledge exists. The delivery does not. That gap — not a shortage of gadgets — is the subject.

Every figure below is attached to a trial, a registry entry or a dated agency page. Where a widely repeated number could not be verified against a primary source, it is not here, and in several places we say which number we deliberately left out and why.

The highest-value, lowest-glamour list

Ranked by strength of evidence for preserving independence, not by novelty. Nothing on this list is new, and nothing on it is a robot.

What actually keeps people independent

Balance and functional exercise, ideally delivered or supervised by a physiotherapist. Cochrane, 108 RCTs, 23,407 community-dwelling participants: 23% fewer falls (RaR 0.77, 95% CI 0.71–0.83, high-certainty), 15% fewer people experiencing at least one fall, 24% for balance and functional programmes specifically, 34% when combined with resistance training. Nothing else on this page has this evidence tier. (PMID 30703272)
Cataract surgery. In a randomised trial of 306 women over 70, expedited first-eye surgery cut the rate of falling by 34% and improved activity, anxiety, depression, confidence, visual disability and handicap. In a cohort with a well-chosen negative control — glaucoma surgery, which does not restore vision — cataract extraction was associated with a hazard ratio of 0.71 for incident dementia. (PMIDs 15615747, 34870676)
Hearing aids, fitted properly and actually worn. Self-reported communicative function improved within 6 months and held at 3 years, and hearing intervention cut the mean number of falls over 3 years by 27% — consistently across both study populations. (PMIDs 39266468, 40441816)
Home hazard reduction, but only for people already at elevated fall risk. 38% fewer falls in people selected for higher risk (high-certainty). No effect whatsoever in unselected populations (also high-certainty). The targeting is the intervention. (PMID 36893804)
A pharmacist. Across 182 randomised trials of medication adherence, the only interventions that improved both adherence and a clinical outcome were complex, tailored, human-delivered programmes involving allied health professionals — most often pharmacists doing education, counselling and ongoing treatment support. (PMID 25412402)
Structured clinical response, wherever remote monitoring is used at all. TIM-HF2 cut all-cause mortality (HR 0.70) with a defined heart-failure population and a 24/7 telemedicine centre that could change treatment. BEAT-HF, with similar sensors and no such structure, changed nothing (HR 1.03). The differentiator was never the device. (PMIDs 30153985, 26857383)

The evidence ledger: does it change outcomes?

The tiers below are the site's standard evidence tiers, which are drug-shaped. Devices, services and exercise programmes do not have Phase 1–3 labels, so they are mapped to the nearest honest equivalent and the mapping is stated in each row: a Cochrane high-certainty synthesis of large randomised trials is treated as Phase-3-equivalent, single-session crossovers of 8–20 people as Phase-1-equivalent, and anything sold to consumers with no controlled outcome evidence as marketed-unproven. Being tested at Phase-3 scale is not the same as working: several rows below are the largest trials ever run in their field and are null.

Technology by tier and by outcome

Ranked on: Highest tier that genuinely exists for each intervention as of 31 August 2026. Regulatory status is from FDA public records; trial status from ClinicalTrials.gov and PubMed. Non-drug interventions are mapped to the nearest tier and the mapping is named in the status line.

Balance and functional exercise (Otago-type)
Phase 3 trials
Supervised or professionally prescribed strength, balance and functional training for community-dwelling older adults.
23% fewer falls (RaR 0.77, 95% CI 0.71–0.83) and 15% fewer people falling at all. Balance and functional programmes 24%; multiple exercise types 34%. Downstream outcomes are weaker — fractures RR 0.73 and falls requiring medical attention RR 0.61 are low-certainty, hospital admission is unclear, and quality of life shows little or no important difference. Larger effects where a health professional, usually a physiotherapist, delivered the programme.
Phase-3-equivalent: Cochrane high-certainty evidence, 108 RCTs, 23,407 participants. Not a regulated product.PMID 30703272
Cataract surgery
Approved
Replacement of the clouded natural lens with an intraocular lens.
Randomised evidence for first-eye surgery is strong on vision, function, confidence and mood, and showed a 34% lower rate of falling. Second-eye surgery showed a 32% reduction that was not statistically significant. A 2026 propensity-weighted analysis of 940,233 Medicare enrollees found no significant difference in one-year falls or fractures. Say “restores vision and function, with a plausible but contested effect on falls” — not “prevents falls”.
Standard of care worldwide; the most-performed surgical procedure in older adults.PMIDs 15615747, 16364936, 41557859
Hearing aids, fitted and worn
Approved
Amplification devices, prescription or over-the-counter, for age-related hearing loss.
ACHIEVE improved communicative function at 6 months and held it at 3 years, and cut the mean number of falls over 3 years by 27%, consistently across both study populations. Cognition was null overall (see below). Observational pooling across 33 countries finds the dementia association tracks effective hearing improvement (HR 0.86) and vanishes where improvement was poor (HR 0.98) — it is corrected hearing, not device ownership.
FDA OTC final rule effective 17 October 2022 for adults 18+ with perceived mild-to-moderate loss. Apple's Hearing Aid Feature authorised via De Novo DEN230081 on 12 September 2024 — a software authorisation, not a clearance of the earbuds as a general medical device.PMIDs 39266468, 40441816, 42127901
Home hazard reduction, targeted
Phase 3 trials
Occupational-therapy-led assessment and removal of fall hazards in the home.
38% fewer falls in people selected for elevated fall risk (RaR 0.62, 95% CI 0.56–0.70). No effect at all in unselected people (RaR 1.05, 95% CI 0.96–1.16). Little or no difference in fractures, hospitalisation or quality of life. The same intervention is highly effective or completely useless depending on who receives it.
Phase-3-equivalent: Cochrane, 22 studies, 8,463 participants; high-certainty on both arms of the targeting finding.PMID 36893804
Structured remote patient management (heart failure)
Phase 3 trials
Daily physiological transmission into a staffed telemedicine centre with 24/7 authority to change treatment.
All-cause death 7.86 vs 11.34 per 100 person-years (HR 0.70, 95% CI 0.50–0.96, p=0.028). The primary endpoint was marginal (ratio 0.80, upper CI bound 1.00, p=0.046) and the population was tightly selected, excluding major depression. This is a care-delivery model, not a device.
Phase-3-equivalent: randomised, n=1,571 (TIM-HF2). Medicare pays for remote patient monitoring; there is no FDA approval of the care model itself.NCT01878630 / PMID 30153985
Generic telemonitoring bolted onto usual care
Phase 3 trials
Post-discharge transmission of blood pressure, weight, heart rate and symptoms reviewed by centralised nurses.
180-day all-cause readmission 50.8% vs 49.2% (adjusted HR 1.03, 95% CI 0.88–1.20, p=.74); no difference in 30-day readmission or 180-day mortality; quality of life did favour the intervention. A 2025 overview of 84 systematic reviews commissioned for an actual procurement decision found only 15% met the quality threshold and seven of sixteen patient groups had no adequate evidence at all.
Phase-3-equivalent and tested at scale: randomised, n=1,437 (BEAT-HF). Null.NCT01360203 / PMIDs 26857383, 39849519
Multifactorial fall-risk assessment programmes
Phase 3 trials
Nurse-delivered individualised risk assessment and management plans in primary care.
First adjudicated serious fall injury: 4.9 vs 5.3 per 100 person-years, HR 0.92 (95% CI 0.80–1.06), p=0.25. Self-reported fall injury was modestly lower (HR 0.90) — the classic pattern of an unblinded soft endpoint moving while the adjudicated one does not. Hospitalisation and death were similar; well-being outcomes showed no clinically meaningful improvement.
Phase-3-equivalent: pragmatic cluster-RCT, 86 practices, 5,451 adults ≥70 at increased risk (STRIDE). Null on the adjudicated endpoint.NCT02475850 / PMIDs 32640131, 33037632
Telecare and personal alarm systems
Phase 3 trials
Pendant alarms, automatic sensors and monitored response for people with social care needs.
Hospital admission within 12 months 46.8% vs 49.2%, OR 0.90 (95% CI 0.75–1.07), p=0.211. Mortality, seven secondary-care metrics, GP and nurse contacts, permanent residential admission and notional costs were all non-significant. The parallel telehealth arm came out at an incremental cost of about £92,000 per QALY with an 11% probability of being cost-effective at £30,000.
Phase-3-equivalent and deployed at national scale: cluster-RCT, 2,600 people, 217 English general practices (Whole Systems Demonstrator). Null.PMIDs 23443509, 23520339
Companion and socially assistive robots
Sold · unproven
PARO the robotic seal, robotic pets, humanoid companions and virtual coaches in care settings.
Against a switched-off look-alike plush toy, PARO won on verbal and visual engagement only; on the validated CMAI-SF there was no difference between any group. A 2024 umbrella review of 35 systematic reviews found no pooled effect on quality of life, anxiety or depression, and no effect on agitation, neuropsychiatric symptoms or medication use. A 2022 meta-analysis concluded there is no clear evidence people with dementia derive benefit for cognition, neuropsychiatric symptoms or quality of life.
Sold and deployed with no regulatory efficacy review. The largest randomised trial is n=415; the largest registered trial found is n=240.PMIDs 28780395, 35462001, 38430662
Wearable fall detection
Sold · unproven
Watches, pendants and phone features that detect a fall and raise an alert.
Targeted searches for randomised trials of wearable fall detection with health outcomes returned nothing. The logic matters: detection addresses the long lie — time on the floor before help arrives — not the fall. The first adequately powered outcome trial is running now: INES (DRKS00031408) is randomising roughly 1,876 people aged 70+ living alone, with days spent in hospital after emergency admission as the primary outcome.
No FDA-cleared fall-risk device exists. Consumer fall detection is not cleared as one. The published literature is algorithm-validation work, not outcome trials.DRKS00031408
Smart-home and ambient sensing
Sold · unproven
Motion, door, bed and localisation sensors inferring activity, cognition or deterioration.
A PRISMA systematic review screened 1,935 articles and found 36 technology-focused papers and 10 that reported any patient outcome. In-home sensors could not differentiate healthy cognition from mild cognitive impairment; frailty was detectable. Room-detection accuracy reaches 92%. There is no randomised evidence that a smart-home sensing system keeps anyone independent for longer, reduces hospitalisation or reduces mortality.
Sold as ageing-in-place infrastructure. Overwhelmingly an engineering literature.PMID 39622026
Exoskeletons for community-dwelling older adults
Phase 1 · safety only
Powered and passive lower-limb devices for gait assistance and augmentation.
A representative study is a randomised crossover treadmill session in 20 healthy adults averaging 63 years, reporting Physiological Cost Index and Borg exertion — acute surrogates, not falls or function. There is no RCT evidence of any patient-important outcome. Post-hip-fracture rehabilitation is the most plausible near-term indication; the first adequately designed multicentre trial (n=86) only began enrolling in February 2026.
Phase-1-equivalent: acute-effects and feasibility stage. Typical published studies are single-session crossovers of 8–20 people. The genuine RCT base is stroke and spinal-cord-injury rehabilitation and does not transfer.NCT07323147 / PMID 42372016
Medication reminder devices and electronic pill bottles
Phase 3 trials
Pillboxes, digital timer caps, bottle strips, Bluetooth caps and adherence feedback.
Optimal adherence was 15.5% with a standard pillbox, 15.1% with a digital timer cap, 16.3% with a bottle strip and 15.1% with no device; odds ratios versus control 1.03, 1.00 and 0.94. A separate randomised trial found electronic pill bottles produced good measured adherence and left systolic blood pressure exactly where bidirectional texting and usual care left it (−4.3, −4.6, −4.7 mmHg).
Phase-3-equivalent and settled: randomised, 53,480 people (REMIND). Null on all three devices.NCT02015806 / PMIDs 28241271, 31396815
Consumer brain-training apps
Sold · unproven
Commercial cognitive-training games marketed for memory and “brain health” in later life.
Three independent meta-analyses spanning 2014–2020 show effects shrinking monotonically from the trained task, to similar tasks, to different domains, to real life. After adjusting for publication bias, the training effect on fluid intelligence was non-significant. Brain-training games reliably make you better at brain-training games.
General-wellness software. No FDA-cleared digital therapeutic for cognition in older adults exists; openFDA returns zero 510(k) records for dementia or Alzheimer's. EndeavorRx is cleared for ADHD in children and adolescents.PMID 24417410

Falls: the trend is the story, not the total

A fall is the single most common event that ends independent living. Every competing page opens with “one in four older adults falls each year.” The more useful and more damning fact is what happened to the death rate after the evidence base was settled.

+21%

Rise in the US age-adjusted fall death rate among older adults, 64.7 to 78.4 per 100,000, between 2018 and 2024

CDC WONDER, released 2026

23%

Fewer falls with exercise — high-certainty, pooled across 108 randomised trials and 23,407 community-dwelling participants

Cochrane 2019, PMID 30703272

38%

Fewer falls from home hazard reduction in people selected for elevated risk — and no effect at all in unselected people

Cochrane 2023, PMID 36893804

27%

Fewer falls over 3 years with hearing intervention, consistent across both ACHIEVE study populations

Lancet Public Health 2025, PMID 40441816

US$80.0bn

US healthcare spending on non-fatal falls among older adults in 2020 (95% CI $32.1–128.0bn), versus $49.5bn in 2015

Injury Prevention, July 2024, PMC11445707

Randomised trials of wearable fall-detection devices reporting a health outcome; the first adequately powered one is recruiting

INES trial, DRKS00031408

The largest test of putting the effective programme into an app is the Safe Step trial: 1,628 adults aged 70+, twelve months, self-managed and unsupervised. The fall rate was unchanged (IRR 0.92, 95% CI 0.76–1.11, p=.37), the risk of experiencing at least one fall was 11% lower (RR 0.89, p=.03), and injurious falls were similar. The story is adherence: the proportion meeting the 90-minute weekly exercise goal was 12.7%, 13.4%, 8.6% and 9.1% at 3, 6, 9 and 12 months, and 20% of the exercise group formally withdrew against 8% of controls. Unsupervised digital delivery preserves a fraction of the supervised effect, and the loss is almost entirely an adherence problem — which is exactly what Cochrane's health-professional-delivery finding predicts.

Note the scale asymmetry running through this whole field. The trials that changed practice enrolled 1,628 and 5,451 people. The technology trials currently running in fall prevention enrol 40 to 320.

Hearing and vision: reported exactly, including the retraction

Hearing and vision are the two highest-value technologies in ageing, and they are also the two most frequently mis-reported. Getting them right is the test of whether a source is worth reading.

ACHIEVE randomised 977 adults aged 70–84 with untreated hearing loss and no substantial cognitive impairment to a best-practice hearing intervention or a health-education control, and followed them for three years. The primary analysis, combining both recruitment cohorts, was null: three-year cognitive change was −0.200 SD in the hearing arm and −0.202 in the control arm, a difference of 0.002 SD (95% CI −0.077 to 0.081), p=0.96. A p-value of 0.96 on a point estimate of 0.002 SD is as null as a result gets, and it must be stated first.

A prespecified sensitivity analysis then found the effect differed significantly between the two recruitment populations (p=0.010). The trial drew 238 participants (24%) from the ARIC cardiovascular cohort — older, more risk factors, lower baseline cognition — and 739 (76%) from healthy de novo community volunteers. A subsequent secondary analysis quantified it: among participants in the top quartile of predicted risk, three-year cognitive decline in the hearing arm was 61.6% slower (95% CI 33.7%–94.1%) than control. That is hypothesis-generating and biologically plausible. It is not a positive trial, and the headline figure circulating elsewhere describes a subgroup.

What ACHIEVE did show unambiguously is easier to overlook. Self-reported communicative function improved within six months (HHIE-S difference −8.9 points) and was sustained at three years (−9.5). And in a prespecified exploratory analysis published in June 2025, the intervention group had a 27% reduction in the mean number of falls over three years (1.45 vs 1.98; mean difference −0.54, 95% CI −0.77 to −0.31) — and critically, this effect was consistent across both the ARIC and de novo populations, unlike the cognition signal. Hearing aids may be a falls intervention that nobody markets as one.

The most-cited hearing-and-dementia paper was retracted

A 2023 UK Biobank analysis reporting that hearing loss without hearing aids carried a hazard ratio of 1.42 for all-cause dementia while hearing-aid users showed no excess risk is flagged by PubMed as a Retracted Publication (original PMID 37062296; retraction notice PMID 38101424, published 12 December 2023). Per the retraction, an error in the authors' SAS output format setting switched the data for hearing-aid users and non-users, producing false and misleading findings; it was caught by an external researcher attempting replication. We do not cite it, and a great deal of currently published ageing-technology content still does. The 2026 replacement evidence is better and more modest: pooling 61,089 hearing-impaired adults aged 55+ across seven harmonised cohorts in 33 countries, hearing aid use carried HR 0.91 for probable dementia — HR 0.86 among those reporting effective hearing improvement, and HR 0.98, no benefit, among those reporting poor improvement.

Vision and hearing: the verified scale figures

Figure	What it means	Source and page date
Over 25% of people aged 60+ have disabling hearing loss	Disabling means worse than 35 dB in the better ear. This is the number that matters for older adults, not the global total.	WHO deafness and hearing loss fact sheet, 3 March 2026
430 million people need rehabilitation for disabling hearing loss; 2.5 billion projected to have some degree of hearing loss by 2050	The current WHO hearing sheet leads with the projection, not with a present-day total-prevalence figure.	WHO deafness and hearing loss fact sheet, 3 March 2026
Hearing aid production meets less than 10% of global demand	Access, not efficacy, is the binding constraint on the best-evidenced device in this section.	WHO assistive technology fact sheet, 2 January 2024
94 million people have unaddressed vision impairment from cataract; 826 million from presbyopia	Cataract is the largest single cause of unaddressed distance vision impairment and blindness. Presbyopia — reading glasses — is the largest category of all.	WHO blindness and vision impairment fact sheet, 10 February 2026
1 in 2 people globally who need cataract surgery do not have access to it	The intervention ranked second on our low-glamour list is unavailable to half the people who need it.	WHO blindness and vision impairment fact sheet, 10 February 2026

WHO fact sheets, with the page date attached. The hearing and vision sheets were both refreshed in early 2026; the falls sheet was not.

On vision, the honest verdict has two halves. Cataract surgery is very strongly supported for restoring acuity, function, confidence, mood and independence-relevant activity — that is not in dispute. The falls claim is weaker than it is usually presented: supported by one randomised trial in first eyes, not replicated in second eyes (rate ratio 0.68, 95% CI 0.39–1.19, p=0.18), and not detected in a February 2026 propensity-weighted analysis of 940,233 Medicare enrollees, which found a risk difference of −0.15 percentage points (95% CI −1.0 to 0.74). The authors of that analysis flag that one year may be too short and that claims data miss self-reported falls — precisely what the original trial's fall diaries captured. Say “restores vision and function, with a plausible but contested effect on falls”.

One further nuance from Cochrane: vision improvement interventions in general, as distinct from cataract surgery, may make little or no difference to fall rate (RaR 1.12, 95% CI 0.84–1.50, low-certainty), with an open question about whether changing someone's eye prescription warrants extra precautions.

Robots: what happens when you control against a plush toy

Blinding is essentially impossible in robot trials — everybody in the room knows there is a robot. So the best available control is a switched-off copy of the same device. One trial did exactly that, and it is the most informative study in this entire section.

Moyle and colleagues ran a three-arm cluster-randomised trial across 28 long-term care facilities in South-East Queensland: 415 participants aged 60 and over with documented dementia, assessors masked, intention-to-treat. Arm one received PARO, the robotic seal, in individual 15-minute sessions three times a week for ten weeks. Arm two received the identical PARO with the robotic features disabled — a look-alike plush toy. Arm three received usual care.

Against usual care, both PARO and the plush toy significantly reduced neutral affect, and PARO improved pleasure and video-observed agitation. Against the plush toy, PARO participants were more verbally engaged (P=.011) and more visually engaged (P<.0001). And on the Cohen-Mansfield Agitation Inventory — Short Form, the validated instrument the trial was actually measuring against, there was no difference between groups. The authors' conclusion, verbatim: “Although more effective than usual care in improving mood states and agitation, PARO was only more effective than a plush toy in encouraging engagement.”

The economic analysis from the same 415 residents is quietly decisive. PARO cost $50.47 more per resident than usual care; the plush toy cost $37.26 more. Incremental cost per CMAI-SF point averted was $13.01 for PARO and $12.85 for the plush toy, and the authors wrote that the plush toy “offered marginally greater value for money than PARO in improving agitation.” Quote the rest of their sentence too, because it is two-sided: both are much cheaper than staff-delivered psychosocial and sensory programmes, so both are reasonable options. PARO is not a rip-off. It is simply not doing anything a soft toy does not do.

The scale asymmetry

Study	What it tested	Participants	Result on the validated endpoint
Cochrane exercise review (PMID 30703272)	Exercise to prevent falls	23,407 pooled, 108 RCTs	23% fewer falls, high-certainty
STRIDE (NCT02475850)	Multifactorial fall-injury prevention	5,451	Null on adjudicated serious fall injury (HR 0.92, p=0.25)
Whole Systems Demonstrator telecare arm	Telecare and monitored alarms	2,600	Null on hospital admission (OR 0.90, p=0.211)
Moyle PARO cluster-RCT (PMID 28780395)	Robotic seal vs switched-off plush toy vs usual care	415	No difference on CMAI-SF; engagement only
NCT05835856 (e-VITA virtual coach)	European–Japanese virtual coach	240 — largest registered robot trial found	Status unknown
NCT05178992 (Vanderbilt)	Socially assistive robotic architecture in long-term care	142	Completed October 2025
NCT05884424 (IMSERSO, Spain)	PARO robot therapy, 15 sites	123	Completed

Registered robot trials versus the trials that actually changed practice. Twenty-one registered socially-assistive-robot trials matched a ClinicalTrials.gov search; the largest was 240 participants.

What the syntheses found

A 2024 umbrella review of 35 systematic reviews found no pooled effect of socially assistive robots on quality of life, anxiety or depression, and no effect on agitation, neuropsychiatric symptoms or medication use. Its suggested mechanism is worth keeping: the main effect within group settings appeared to be the stimulation of social interaction with other humans. The robot is a conversation piece; the humans are the intervention. (PMID 38430662)
A 2022 meta-analysis of 66 studies found PARO had no effect versus usual care or active control on agitation, cognition, overall neuropsychiatric symptoms, apathy, depression, anxiety or quality of life, and concluded there is no clear evidence people with dementia derive benefit for cognition, neuropsychiatric symptoms or quality of life. (PMID 35462001)
A 2023 meta-analysis of 12 articles and 1,461 participants did find small-to-moderate effects on medication use, anxiety, agitation and depression — but graded the quality of evidence for all outcomes as low, citing methodological limitations, small samples and wide confidence intervals. (PMID 37348392)
In a head-to-head across four nursing homes, live dog visits beat the robot seal on week-three sleep duration (610 vs 498 minutes), and the robot scored below an inert soft toy cat (540 minutes). (PMID 26510632)
An accelerometry analysis of the same 415-resident trial found PARO produced greater reductions in daytime step count than both usual care and the plush toy. In a frail population at risk of deconditioning, less movement is not self-evidently a benefit. (PMID 29563027)
Exoskeletons for community-dwelling older adults have no RCT evidence of any patient-important outcome — not falls, not activities of daily living, not institutionalisation. The published work is single-session crossovers of 8 to 20 people and engineering studies; the real randomised base sits in stroke and spinal-cord-injury rehabilitation and does not transfer.

Japan did not replace carers with robots

The peer-reviewed study published on 6 August 2026 used roughly 860 Japanese nursing homes with prefecture-level care-robot subsidy variation as an instrumental variable, and found robot adoption associated with 28% more care workers, 39% more nurses and about 26% higher total facility employment — growth occurring entirely among nonregular employees, alongside a significant reduction in staffing retention problems. An earlier two-wave panel of 265 homes gave a smaller estimate, roughly 5.8% more caregiver staffing for monitoring robots. Different datasets, different years, different identification: report both, average neither, and do not state either as settled causal fact. What Japan actually bought is sensors — the average adopting home held 10.3 monitoring robots, 1.3 mobility robots and 1.0 transfer robot — and the task-automation indexes were only 1.2 to 1.7, which the authors gloss as humans still performing the bulk of each task with a little bit of support.

What Japan actually deploys

Priority field	Penetration
Monitoring and communication	30.0%
Bathing assistance	11.2%
Care work support	10.2%
Transfer assistance	9.7%
Mobility support	1.2%
Excretion and toileting support	0.5%

Penetration rates from Japan's FY2021 national survey, published by METI alongside its 28 June 2024 revision of the Priority Fields in the Use of Technologies for Long-term Care. The distribution is the argument.

Two further Japanese findings belong here. A 2026 analysis of three years of national survey data found that of 4,688 residential long-term care providers, 1,250 — 26.7% — had adopted at least one care robot, with the authors explicitly cautioning that adoption was defined as implementing at least one robot and “does not necessarily reflect sustained or intensive use.” Every Japanese care-robot adoption statistic in circulation measures purchase, not use. And a well-designed ABAB study of a nonverbal AI communication robot in disaster-affected Japanese care facilities was null on all three staff well-being outcomes, with the authors stating efficacy remains unproven.

Two claims about Japan we do not make in either direction: there is no verifiable quantitative abandonment or discontinuation rate for care robots in the peer-reviewed literature, and no study located measured staff workload as distinct from employment. The workload claim is the one most often asserted and the one least supported.

Sensors, pill bottles and brain games: surrogate creep

Medication management is the cleanest example, and it is settled. REMIND randomised 53,480 CVS Caremark enrollees who were already suboptimally adherent to a pill bottle strip, a digital timer cap, a standard pillbox, or no device, with twelve months of pharmacy-claims follow-up. Optimal adherence was 15.5%, 15.1%, 16.3% and 15.1% respectively; odds ratios versus control were 1.03, 1.00 and 0.94. The authors: low-cost reminder devices did not improve adherence. Separately, a pragmatic trial of 149 primary care patients with uncontrolled hypertension found that electronic pill bottles and bidirectional texting produced good measured adherence and left systolic blood pressure indistinguishable from usual care (−4.3, −4.6, −4.7 mmHg; all p ≥ 0.93).

Cochrane's definitive review of 182 randomised trials — all required to measure both adherence and a clinical outcome — found that only 17 had the lowest risk of bias, and only five of those reported improvements in both. The reviewers stopped classifying by intervention type because of heterogeneity, which means “smart pill bottles work” is not a claim this evidence base supports. What the lowest-bias trials had in common was not gadgets: they were complex, multi-component interventions with tailored ongoing support from allied health professionals such as pharmacists, often delivering intense education, counselling or daily treatment support. For a pharmacist-run site the honest answer is uncomfortable and clear: the effective ingredient is a pharmacist, not a dispenser.

There is also an under-discussed harm. In English community pharmacy, an estimated 273,529 medication compliance aids are filled monthly, a figure that more than doubled in a decade, on an evidence base its own researchers describe as lacking cost-effectiveness. In a survey of pharmacists, 74.2% said their decision to start a compliance aid was never or rarely affected by risk of adverse events. Suddenly increasing adherence in an over-prescribed older patient can be dangerous. For many older people the highest-value medication intervention is deprescribing, not better adherence to an over-long list.

Smart-home sensing tells the same story from the research end. A PRISMA systematic review of in-home localisation technologies screened 1,935 articles and found 36 technology-focused papers and 10 that reported any patient outcome. Of those, in-home sensors could not differentiate older adults with healthy cognition from those with mild cognitive impairment; frailty was detectable. The best observational sensor cohorts are sophisticated and tiny — 85 older adults living alone in the Oregon work, and a loneliness model built in 16 people explaining about a third of the variance. That is proof of concept, not a screening tool. And there is a structural privacy trade-off underneath all of it: the systems with the best acceptability are the least informative, and the most informative are the most invasive.

On cognitive tools, the plain version: brain-training games reliably make you better at brain-training games. ACTIVE, the largest and longest randomised trial in healthy older adults, trained 2,832 people and followed them ten years. Reasoning and speed training maintained effects on the specific ability trained, memory training did not, and all three groups reported less difficulty with instrumental activities of daily living. But that everyday-function benefit is self-reported against a no-contact control in an unblinded trial — which cannot separate a real functional effect from expectancy. Three independent meta-analyses spanning 2014 to 2020 show effects shrinking monotonically from trained task to near transfer to far transfer to real life, and after adjusting for publication bias the training effect on fluid intelligence was non-significant. We do not state a dementia-prevention claim for cognitive training, and we do not cite the widely publicised secondary analysis that is used to support one.

The regulatory vacuum readers assume does not exist

Searches of the FDA 510(k) and De Novo databases return a clear negative result. There is no FDA-cleared digital therapeutic for cognition or dementia in older adults — zero records for “dementia” and zero for “alzheimer” at any date. EndeavorRx is a Class II digital therapeutic cleared for attention deficit hyperactivity disorder, originally for ages 8–12 with a later adolescent expansion; any article listing it as evidence for cognitive technology in ageing is simply wrong. And there is no modern FDA-cleared fall-risk device at all: a device-name search for “fall” returns three records, all from 1979 to 1984. Consumer fall detection, including the watch feature, is not FDA-cleared as a fall-risk device. A cleared prescription digital therapeutic has a specific indication, an age range and a regulatory pathway on the public record. A general-wellness brain-training app has none of these and has never been evaluated by FDA for efficacy. For cognition in older adults, as of August 2026, the second category is the only category that exists.

Delivery, equity, and what changed in 2025–2026

This undercuts every claim in every section above, and it is documented rather than speculative. A retrospective cohort using the nationally representative National Health and Aging Trends Study examined 5,274 home or community rehabilitation users aged 70+, representing a weighted 33,576,313 US older adults. Unreadiness was defined as lacking an internet-enabled device, having limited proficiency, or living with severe cognitive, visual or hearing impairment. Approximately two in three older rehabilitation users were ready to participate in video-based rehabilitation.

The disparities are not gaps; they are different worlds. Odds of readiness were 0.75 for rural residents, 0.37 for the financially strained, 0.23 for non-Hispanic Black older adults and 0.17 for Hispanic older adults, each against the relevant reference group. The authors' policy conclusion matters as much as the numbers: efforts should focus not only on broadband access but on technology ownership and training. Broadband alone does not close this.

Note the compounding effect built into that definition. Sensory and cognitive impairment themselves cause digital unreadiness. So the older adult with untreated hearing loss and cataract — the person for whom the two best-evidenced interventions on this page would do the most good — is also the person least able to reach a telehealth pathway to them. That is the strongest available argument for the ordering we recommend: fix hearing and vision first, because they are prerequisites for everything digital.

The evidence base itself is drawn from a healthier, younger, whiter and more digitally capable population than the one the technology is sold to. A heart-failure remote-monitoring analysis explicitly noted a systematic bias toward recruiting individuals younger than the epidemiological average. ACHIEVE's cohort was 88% White.

Globally the picture is starker still. More than 2.5 billion people need one or more assistive products, projected to reach 3.5 billion by 2050. As few as 3% of people who need assistive products have access in low-income countries, against 90% in some wealthy ones. Only 5–35% of the roughly 80 million people who need a wheelchair have one. Hearing aid production meets less than 10% of global demand.

And the workforce that would deliver any of this has not grown. OECD's Health at a Glance 2025 records the average number of long-term care workers holding stable at 5 per 100 people aged 65 and over between 2013 and 2023 across 31 countries — ranging from 13.0 in Norway to 0.2 in Greece — with more than one in three in part-time employment, one in six on fixed-term contracts, and no qualification requirement at all in Greece, Iceland, the United Kingdom and the United States. Meanwhile the population aged 60 and over climbs toward 1.4 billion by 2030. That flat ratio is the real reason technology is being pushed into this space, and it is precisely why overstating what the technology does is harmful rather than merely inaccurate.

Against that delivery backdrop, here is what actually moved in the last eighteen months — and where the money moved ahead of the evidence.

Developments that actually move the evidence — or move the money ahead of it

March 2025
Safe Step publishes: the largest test of app-delivered fall prevention
1,628 adults aged 70+, twelve months, self-managed. Fall rate unchanged; risk of at least one fall 11% lower; adherence to the weekly exercise goal collapsed to 9–13%.
PMID 40163860 / NCT03963570
May 2025
ACHIEVE risk-stratified analysis
Among participants in the top quartile of predicted risk, three-year cognitive decline in the hearing arm was 61.6% slower than control — the number behind the headlines, and a subgroup finding.
PMID 40369891
June 2025
Hearing intervention reframed as a falls intervention
ACHIEVE's prespecified exploratory analysis reports a 27% reduction in mean falls over three years, consistent across both study populations — unlike the cognition signal.
PMID 40441816
November 2025
OECD records a decade of flat long-term care staffing
Health at a Glance 2025: long-term care workers stable at 5 per 100 people aged 65+ from 2013 to 2023 across 31 countries. The demand curve did not stay flat.
OECD, DOI 10.1787/8f9e3f98-en
January 2026
Medicare starts paying for far less remote monitoring data
Under the CY 2026 Physician Fee Schedule final rule, new CPT codes 99445 and 99470 cover remote patient monitoring at 2–15 days of data per 30 days, down from a 16-day minimum. CMS records that none of the RPM codes met the RUC's minimum survey requirements and all are to be resurveyed by January 2028. Reimbursement is expanding ahead of the outcome evidence.
CMS-1832-F, Federal Register Vol. 90 No. 212
February 2026
The period's most important contrarian finding
A propensity-weighted analysis of 940,233 Medicare enrollees with untreated cataract finds no significant association between cataract surgery and one-year falls or fractures (risk difference −0.15 percentage points, 95% CI −1.0 to 0.74).
PMID 41557859
May 2026
Largest-ever pooled cohort on hearing aids and dementia
61,089 hearing-impaired adults aged 55+, seven harmonised cohorts, 33 countries: HR 0.91 overall, HR 0.86 among those reporting effective hearing improvement, HR 0.98 — no benefit — among those reporting poor improvement. It is corrected hearing, not aid ownership.
PMID 42127901
28 May 2026
ACHIEVE Brain Health Follow-Up reaches primary completion
n=629, six-year change in global cognition and time to adjudicated MCI or dementia. Control-arm participants were offered the intervention after Year 3, making this an early-versus-delayed comparison. This is the most consequential pending result in the field; no publication was located as of 31 August 2026.
NCT05532657
22 July 2026
FDA names first participants in the TEMPO digital health pilot
Tied to the CMS Innovation Center ACCESS model, under which FDA intends to exercise enforcement discretion for certain requirements including premarket authorisation. FDA's own caveat, verbatim: the effectiveness of the selected devices has not yet been evaluated by the FDA. A pathway that waives premarket authorisation is a regulatory development, not evidence of efficacy.
FDA Digital Health Center of Excellence
6 August 2026
The Japan story is corrected in peer review
An instrumental-variable analysis of roughly 860 Japanese nursing homes finds robot adoption associated with 28% more care workers, 39% more nurses and about 26% higher total employment, entirely among nonregular staff, plus reduced retention problems. Robots complement rather than replace caregivers.
Health Affairs Review, inaugural issue

Consumer wearables: what's actually validated versus marketed

Whoop, Oura, Apple Watch and Garmin are the four devices that dominate longevity-adjacent self-tracking, each selling some version of "recovery," "readiness" or "strain" scores built on proprietary algorithms. The devices themselves are consumer electronics, not medical devices — none is FDA-cleared for diagnosing anything — so the only honest way to compare them is against independent validation studies that check their raw sensor readings (heart rate, heart rate variability, sleep staging) against a clinical-grade reference, rather than against each brand's own marketing claims about its algorithm.

That distinction matters because a device can be excellent at collecting a number and still questionable in how it interprets that number into a "score." The validation research below covers the sensor accuracy question, which is the part that's actually been tested independently. Whether acting on a daily "readiness" score changes any health outcome is a separate, much less studied question — no wearable's proprietary score has outcome-trial evidence behind it.

Sensor accuracy in independent validation research

Device (validation studies used earlier generations)	Resting heart rate / HRV	Sleep staging	Notes
Oura Ring — current: Gen 4 (Oct 2024)	Strongest agreement of the group — Gen 3 MAPE 1.67%, Gen 4 MAPE 1.94% for resting heart rate, in a 2025 study covering 500+ nights of real-world data	About 5% more accurate than Apple Watch and 10% more accurate than Fitbit for four-stage sleep classification in comparative testing	Ring form factor; consistently the top performer for nighttime physiological signals across independent studies. Gen 4 added an 18-path PPG sensor array (up from 6), Cardiovascular Age and a VO2 max estimate over Gen 3 — none of those additions have independent validation data yet
Whoop — current: Whoop 5.0 and Whoop MG (May 2025)	Moderate agreement — MAPE around 3.00% for resting heart rate in the 2025 study, using the prior-generation device; behind Oura	Reported roughly 75-86% accuracy depending on the specific sleep metric in its own validation research	Subscription-only, no purchase price for the hardware itself. Whoop MG adds an FDA-cleared ECG and blood-pressure estimates, available only on its top-tier annual plan — that ECG feature has FDA clearance, but is separate from the resting-HR/sleep-staging figures in this row, which predate MG
Apple Watch — current: Series 11, Ultra 3, SE 3 (Sept 2025)	Not the top performer for overnight HRV/resting heart rate in comparative validation work	Sharp at detecting when someone is genuinely awake; weaker specifically at identifying deep sleep stages	Broadest software ecosystem and the only lineup with an FDA-cleared feature (ECG app, AFib notifications) on its mainstream models — those two cleared features are the exception, not the rule, for the rest of its health metrics
Garmin — current: Fenix 9/9 Pro, Venu 4/X1, Forerunner 570/970 (2026)	Included in the same 2025 multi-device validation study; performance fell behind Oura for nighttime HR/HRV agreement	Less independent sleep-staging validation data published specifically for Garmin than for the other three	Widest range of hardware form factors and price points of any brand here — Fenix for rugged/outdoor, Venu for everyday wellness, Forerunner for running; strongest for exercise/GPS tracking specifically, which is a separate, well-validated use case

Drawn from controlled validation studies comparing device output against clinical-grade reference measurements (polysomnography for sleep, ECG for heart metrics) — not from manufacturer-reported figures.

Current-generation models, official product pages

Oura Ring 4 (2024) — oura.com/product/oura-ring-4
Whoop 5.0 and Whoop MG (2025) — whoop.com
Apple Watch Series 11, Ultra 3, SE 3 (2025) — apple.com/apple-watch
Garmin Fenix 9/9 Pro, Venu 4, Forerunner 570/970 (2026) — garmin.com

Where to buy

Oura Ring 4 — Check price on Amazon Whoop 5.0/MG — Check price on Amazon Apple Watch Series 11 — Check price on Amazon Garmin Venu 4 — Check price on Amazon

Frequently asked questions

Do companion robots help people with dementia?

Not on validated instruments. The decisive trial randomised 415 people with dementia across 28 facilities to the PARO robotic seal, an identical look-alike plush toy with the robotics switched off, or usual care. PARO beat the plush toy on verbal and visual engagement only; on the Cohen-Mansfield Agitation Inventory — Short Form there was no difference between groups. The trial's own economic analysis found the plush toy offered marginally greater value for money per point of agitation averted, though both were cheap relative to staff-delivered psychosocial programmes. A 2024 umbrella review of 35 systematic reviews found no pooled effect on quality of life, anxiety or depression, and no effect on agitation, neuropsychiatric symptoms or medication use. If a robot brings someone pleasure, that is a real and reasonable thing to buy. It is not a clinical intervention.

Will a fall-detection watch or pendant keep my parent safer?

It may shorten a long lie — the time spent on the floor before help arrives — and that is genuinely worth something. There is no randomised evidence that it prevents falls, prevents injury, reduces hospitalisation or preserves independence. The published literature on fall-detection wearables is overwhelmingly algorithm-validation work: how well the device spots a fall, not whether wearing one changes anything. Targeted searches for randomised trials with health outcomes returned nothing. The first adequately powered trial, INES in Germany, is randomising roughly 1,876 people aged 70+ living alone with days in hospital as the primary outcome. Until it reports, buy a detector for peace of mind and reduced time on the floor — and spend the rest of the budget on balance training.

Do hearing aids prevent dementia?

The randomised trial says no, with an important caveat. ACHIEVE followed 977 adults aged 70–84 for three years; the primary cognition analysis was null, with a difference of 0.002 standard deviations and a p-value of 0.96. A prespecified analysis then found the effect differed by recruitment cohort (p=0.010), and a follow-up analysis showed 61.6% slower decline among participants in the top quartile of predicted risk. That is a subgroup, not a trial result, and the widely circulated percentage describes the subgroup. Separately, a much-cited 2023 UK Biobank paper reporting a hearing-aid benefit was retracted in December 2023 after a coding error swapped the exposure groups. The best current observational evidence — 61,089 adults across 33 countries — finds a modest association (HR 0.91) that is confined to people reporting effective hearing improvement. Get hearing aids for hearing, communication and falls. Do not buy them as a dementia drug.

Does cataract surgery prevent falls?

It restores vision, function, confidence, mood and activity — that part is not in dispute and it is why cataract surgery sits second on our list. The falls claim is genuinely contested. One randomised trial of 306 women over 70 found expedited first-eye surgery cut the rate of falling by 34% using fall diaries. The equivalent trial in second eyes found a 32% reduction that was not statistically significant. And a February 2026 propensity-weighted analysis of 940,233 Medicare enrollees found no significant difference in one-year falls or fractures. The claims analysis may miss self-reported falls that never reach the health system, which is exactly what the diaries captured. Honest wording: restores vision and function, with a plausible but contested effect on falls.

Do brain-training apps work?

They reliably make you better at brain-training games, and produce small improvements on similar laboratory tasks. Evidence that they preserve everyday function or prevent dementia is weak. Three independent meta-analyses spanning 2014 to 2020 show effects shrinking steadily as you move from the trained task, to similar tasks, to different domains, to real life — and after correcting for publication bias, the training effect on fluid intelligence was non-significant. The strongest-looking results come from trials whose comparator was doing nothing. The largest and longest randomised trial, ACTIVE, found reasoning and speed training maintained effects on the specific ability trained over ten years, but its everyday-function benefit was self-reported against a no-contact control, which cannot separate a real effect from expectancy.

Is there an FDA-approved app or device for memory loss in older adults?

No. Searches of the FDA 510(k) and De Novo databases return zero records for dementia and zero for Alzheimer's at any date. EndeavorRx, the device most often cited in this context, is a Class II digital therapeutic cleared for attention deficit hyperactivity disorder in children and adolescents — not for older adults and not for cognitive decline. There is also no modern FDA-cleared fall-risk device: a device-name search for “fall” returns three records, all from 1979 to 1984, so consumer fall detection is not cleared as a fall-risk device. A cleared prescription digital therapeutic has a specific indication, an age range and a public regulatory pathway. A general-wellness brain-training app has none of these and has never been assessed by FDA for efficacy. For cognition in older adults, that second category is currently the only category that exists.

Didn't Japan solve its care shortage with robots?

No, and the real story is more interesting. The peer-reviewed August 2026 analysis of roughly 860 Japanese nursing homes, using prefecture-level subsidy variation as an instrument, found robot adoption associated with 28% more care workers, 39% more nurses and about 26% higher total employment — growth occurring entirely among nonregular staff — alongside fewer retention problems. An earlier panel of 265 homes gave a smaller estimate of roughly 5.8%. Both are observational and their authors say so; report both and average neither. What Japan actually deploys is monitoring: 30.0% penetration for monitoring and communication against 1.2% for mobility support and 0.5% for toileting support, and task-automation indexes of only 1.2 to 1.7 — humans still do the bulk of every task. Also note that 26.7% of 4,688 providers had adopted at least one care robot, a figure the authors caution measures purchase, not sustained use.

Do smart pill bottles and medication reminders improve health?

The reminder hardware question is settled and the answer is no. REMIND randomised 53,480 already-nonadherent people to a pill bottle strip, a digital timer cap, a standard pillbox or nothing, and found odds ratios versus control of 1.03, 1.00 and 0.94 — all null. Electronic pill bottles do better on the thing they measure and no better on the thing that matters: in a randomised trial of patients with uncontrolled hypertension, they produced good measured adherence and left systolic blood pressure indistinguishable from usual care. Cochrane's review of 182 trials found that what worked, weakly, was people: complex, tailored, ongoing support from allied health professionals, most often pharmacists. There is also a real risk in the other direction — for an over-prescribed older person, the highest-value medication intervention is often deprescribing, not tighter adherence to an over-long list.

Does a smart home with sensors help someone stay independent longer?

There is currently no randomised evidence that it does. A systematic review of in-home localisation technologies screened 1,935 articles and found just 10 that reported any patient outcome at all — the field is overwhelmingly engineering, not clinical research. Those studies showed that in-home sensors could not differentiate healthy cognition from mild cognitive impairment, though frailty was detectable. The largest randomised test of the deployed version of this idea — telecare across 2,600 people and 217 English general practices — found no reduction in hospital admission (OR 0.90, p=0.211) and nothing on any secondary service-use metric, and the parallel telehealth arm cost about £92,000 per quality-adjusted life year. There is also a structural trade-off: the systems with the best acceptability are the least informative, and the most informative are the most invasive.

If I can only do one thing, what should it be?

Balance and functional exercise, delivered or supervised by a physiotherapist. It is the only intervention on this page with high-certainty evidence pooled across 108 randomised trials and 23,407 participants, and delivery by a health professional produced larger effects than self-directed programmes. After that, in order: get cataracts assessed, get hearing tested and get aids fitted properly and actually worn, have a pharmacist review the medicines list with deprescribing on the table, and — only if the person is already at elevated fall risk — get a home hazard assessment, because in unselected people it does nothing. If you are choosing between a £1,000 gadget and a course of supervised exercise plus a hearing test, the evidence is not close.

Which is more accurate: Whoop, Oura, Apple Watch or Garmin?

For overnight resting heart rate and heart rate variability specifically, independent validation research (a 2025 study covering 500+ nights of real-world data) found Oura's Gen 3 and Gen 4 rings had the strongest agreement with reference measurements, with Whoop showing moderate accuracy and Apple Watch and Garmin not leading on this specific metric. For sleep staging, Oura outperformed Apple Watch by roughly 5% and Fitbit by roughly 10% in comparative testing, while Apple Watch was notably sharp at detecting wakefulness specifically. None of this is a verdict on which device is "best" overall — Garmin remains strongest for exercise and GPS tracking, and Apple Watch is the only one with an FDA-cleared feature (its ECG app) at all. "Most accurate sensor for a specific metric" and "best device for you" are different questions.

Does wearing a fitness tracker actually make you healthier, or just measure you more?

This is genuinely unresolved. Sensor accuracy (can the device correctly measure your heart rate or sleep stage) and outcome evidence (does using the device change your actual health) are separate questions, and only the first has decent independent research behind it. No wearable's proprietary "readiness," "recovery" or "strain" score has been tested in a randomised trial showing it improves a health outcome compared with not having one. What the device measures may be accurate; what you do in response to a daily score, and whether that response helps, is unstudied.

Is Apple Watch a medical device?

Only for two specific, FDA-cleared features: its ECG app (which can flag signs of atrial fibrillation) and its irregular rhythm notifications. Every other health metric it reports — sleep stages, "cardio fitness" estimates, blood oxygen readings — is a consumer wellness feature, not a cleared diagnostic one, the same regulatory status as its competitors' equivalent metrics.

Sources

Every figure and claim on this page traces to one of these. Where a source is a company announcement rather than peer-reviewed research or a regulator, it is labelled as such.

01Sherrington C, et al. Exercise for preventing falls in older people living in the community (Cochrane, 108 RCTs, 23,407 participants) — Cochrane Database of Systematic Reviews, 2019 · PMID 30703272
02Sherrington C, et al. Exercise for preventing falls: abridged Cochrane systematic review, with the health-professional delivery finding — British Journal of Sports Medicine, 2020 · PMID 31792067
03Clemson L, et al. Environmental interventions for preventing falls: 38% reduction in selected high-risk, no effect unselected — Cochrane Database of Systematic Reviews, 2023 · PMID 36893804
04Bhasin S, et al. A randomized trial of a multifactorial strategy to prevent serious fall injuries (STRIDE) — New England Journal of Medicine, 2020 · PMID 32640131 / NCT02475850
05Gill TM, et al. Effect of a multifactorial fall injury prevention intervention on patient well-being (STRIDE) — Journal of the American Geriatrics Society, 2021 · PMID 33037632
06Pettersson B, et al. Effectiveness of the Safe Step digital exercise program to prevent falls (n=1,628) — Journal of Medical Internet Research, 2025 · PMID 40163860 / NCT03963570
07Koehler F, et al. Efficacy of telemedical interventional management in patients with heart failure (TIM-HF2) — The Lancet, 2018 · PMID 30153985 / NCT01878630
08Ong MK, et al. Effectiveness of remote patient monitoring after discharge of hospitalized patients with heart failure (BEAT-HF) — JAMA Internal Medicine, 2016 · PMID 26857383 / NCT01360203
09Wartenberg C, et al. Clinical benefits and risks of remote patient monitoring: overview and assessment of methodological rigour — BMC Health Services Research, 2025 · PMID 39849519
10Lin FR, et al. Hearing intervention versus health education control to reduce cognitive decline (ACHIEVE) — primary result null, p=0.96 — The Lancet, 2023 · PMID 37478886 / NCT03243422
11Goman AM, et al. Effects of hearing intervention on falls in older adults: ACHIEVE secondary analysis (27% fewer falls) — The Lancet Public Health, 2025 · PMID 40441816
12Pike JR, et al. Cognitive benefits of hearing intervention vary by risk of cognitive decline (61.6% in highest-risk quartile) — Alzheimer's & Dementia, 2025 · PMID 40369891
13Sanchez VA, et al. Effect of hearing intervention on communicative function (ACHIEVE) — Journal of the American Geriatrics Society, 2024 · PMID 39266468
14Jiang F, et al. Hearing aid effectiveness and probable dementia risk across 33 countries (61,089 participants, seven cohorts) — Cell Reports Medicine, 2026 · PMID 42127901
15Retraction notice for the 2023 UK Biobank hearing aid and dementia analysis — exposure groups swapped by a SAS output format error (original PMID 37062296, do not cite) — The Lancet Public Health, 2024 · PMID 38101424
16ACHIEVE Brain Health Follow-Up Study — six-year cognition and incident MCI/dementia, n=629, primary completion 28 May 2026 — ClinicalTrials.gov / Johns Hopkins University, 2026 · NCT05532657
17Harwood RH, et al. Falls and health status in elderly women following first eye cataract surgery: randomised controlled trial — British Journal of Ophthalmology, 2005 · PMID 15615747
18Foss AJE, et al. Falls and health status following second eye cataract surgery: randomised controlled trial (not significant) — Age and Ageing, 2006 · PMID 16364936
19Lee CS, et al. Association between cataract extraction and development of dementia, with glaucoma surgery as a negative control — JAMA Internal Medicine, 2022 · PMID 34870676
20Thompson AC, et al. Cataract surgery and risk of falls and fractures among 940,233 Medicare enrollees — null — Journals of Gerontology Series A, 2026 · PMID 41557859
21Moyle W, et al. Use of a robotic seal as a therapeutic tool to improve dementia symptoms: cluster-randomised trial with a switched-off plush toy control (n=415) — Journal of the American Medical Directors Association, 2017 · PMID 28780395
22Cost-effectiveness analysis of the PARO cluster-randomised trial — the plush toy offered marginally greater value for money — Journal of the American Medical Directors Association, 2018 · PMID 29325922
23Motor activity and sleep analysis of the PARO cluster-randomised trial — greater reductions in daytime step count — Maturitas, 2018 · PMID 29563027
24Yu C, Sommerlad A, Sakure L, Livingston G. Socially assistive robots for people with dementia: systematic review and meta-analysis (66 studies) — Ageing Research Reviews, 2022 · PMID 35462001
25Nichol B, et al. Socially assistive robots: umbrella review of 35 systematic reviews — no pooled effect on quality of life, anxiety or depression — International Journal of Nursing Studies, 2024 · PMID 38430662 / PROSPERO CRD42023423862
26Rashid NLA, et al. Robot-assisted therapy for older adults: systematic review and meta-analysis — all outcomes graded low quality — International Journal of Nursing Studies, 2023 · PMID 37348392
27Live dog visits versus robot seal versus soft toy cat in nursing homes — the robot scored below the inert toy on sleep duration — Psychogeriatrics, 2015 · PMID 26510632
28Tosaka Y, Funada S, Goto R. Organization-level factors associated with the adoption of care robots in Japanese long-term care providers (4,688 providers, 26.7% adoption) — BMC Health Services Research, 2026 · PMID 41572284
29Nonverbal AI communication robot in disaster-affected Japanese care facilities, ABAB design — null on all three staff well-being outcomes — JMIR Formative Research, 2026 · PMID 42447466
30Randomised crossover, single-session powered exoskeleton in healthy older adults — acute surrogate outcomes only — Journal of Visualized Experiments, 2026 · PMID 42372016
31Exoskeleton-assisted rehabilitation after hip fracture in elderly patients — first multicentre RCT, enrolling from February 2026 — ClinicalTrials.gov, 2026 · NCT07323147
32Steventon A, et al. Effect of telecare on use of health and social care services: Whole Systems Demonstrator cluster-randomised trial (n=2,600) — Age and Ageing, 2013 · PMID 23443509
33Henderson C, et al. Cost effectiveness of telehealth for patients with long term conditions — £92,000 per QALY — BMJ, 2013 · PMID 23520339
34Chan A, et al. In-home positioning for remote home health monitoring in older adults: systematic review — 1,935 articles screened, 10 reported a patient outcome — JMIR Aging, 2024 · PMID 39622026 / PROSPERO CRD42022339845
35Rebok GW, et al. Ten-year effects of the ACTIVE cognitive training trial on cognition and everyday functioning (n=2,832) — Journal of the American Geriatrics Society, 2014 · PMID 24417410
36Nieuwlaat R, et al. Interventions for enhancing medication adherence — 182 RCTs; only 5 of the 17 lowest-bias trials improved both adherence and a clinical outcome — Cochrane Database of Systematic Reviews, 2014 · PMID 25412402
37Choudhry NK, et al. Effect of reminder devices on medication adherence: the REMIND randomized clinical trial (n=53,480) — all three devices null — JAMA Internal Medicine, 2017 · PMID 28241271 / NCT02015806
38Mehta SJ, et al. Electronic pill bottles or bidirectional text messaging to improve hypertension medication adherence — blood pressure unchanged — Journal of General Internal Medicine, 2019 · PMID 31396815 / NCT02778542
39Shenoy P, Scott S, Bhattacharya D. Medication compliance aids in English community pharmacy — 273,529 filled monthly on an evidence base lacking cost-effectiveness — Research in Social and Administrative Pharmacy, 2019 · PMID 31477529
40Falvey JR, et al. Demystifying the digital divide: disparities in telerehabilitation readiness among older adults in the United States — Archives of Physical Medicine and Rehabilitation, 2024 · PMID 38554795
41Leyland LA, et al. The effect of cycling on cognitive function and well-being in older adults (n=100, aged 50–83) — PLOS ONE, 2019 · PMID 30785893
42Livingston G, et al. Dementia prevention, intervention, and care: 2024 report of the Lancet standing Commission — The Lancet, 2024 · PMID 39096926
43Healthcare spending on non-fatal falls among older US adults, 2020: US$80.0 billion — Injury Prevention, 2024 · PMC11445707
44Deafness and hearing loss fact sheet — over 25% of people aged 60+ have disabling hearing loss (page dated 3 March 2026) — World Health Organization, 2026
45Blindness and vision impairment fact sheet — 94 million with unaddressed cataract; 1 in 2 lack access to surgery (page dated 10 February 2026) — World Health Organization, 2026
46Assistive technology fact sheet — 2.5 billion people need assistive products; 3% access in low-income countries versus 90% in some wealthy ones — World Health Organization, 2024
47Health at a Glance 2025 — long-term care workers stable at 5 per 100 people aged 65+ between 2013 and 2023 — OECD, 2025 · DOI 10.1787/8f9e3f98-en
48FDA authorizes first over-the-counter hearing aid software (Apple Hearing Aid Feature) — De Novo DEN230081, 21 CFR 874.3335, decision 12 September 2024 — US Food and Drug Administration, 2024 · DEN230081
49Digital Health Center of Excellence — TEMPO pilot first participants announced 22 July 2026, with enforcement discretion on premarket authorisation — US Food and Drug Administration, 2026
50CY 2026 Medicare Physician Fee Schedule final rule — new remote patient monitoring codes 99445 and 99470 for 2–15 days of data — Centers for Medicare & Medicaid Services / Federal Register Vol. 90 No. 212, 2025 · CMS-1832-F, RIN 0938-AV50
51Revision of Priority Fields in the Use of Technologies for Long-term Care, 28 June 2024, with FY2021 penetration rates (monitoring 30.0%, toileting 0.5%) — METI / MHLW, Japan, 2024
52Eggleston K, Iizuka T, Lee YS, Xi E. Robots and labor in the service sector: evidence from nursing homes — +28% care workers, instrumental-variable design — Health Affairs Review (inaugural issue) / Stanford summary, 2026
53Validation of nocturnal resting heart rate and heart rate variability in consumer wearables (500+ nights, Oura/Whoop/Garmin/Polar) — PMC / peer-reviewed validation study, 2025
54A Validation of Six Wearable Devices for Estimating Sleep, Heart Rate and Heart Rate Variability in Healthy Adults — PMC, 2022

Articles on this topic

Which longevity tech robots offer the most accurate sensors?
Longevity and care robots ranked on sensor accuracy against clinical references: telepresence and monitoring robots using validated peripheral devices, social companion robots (ElliQ, PARO), mobile home robots with camera-based activity and fall detection, reminder and dispensing robots, and consumer 'health robots' with contactless vital-sign sensing — with what each sensor has actually been validated to do.
Best longevity robots with advanced biometric monitoring sensors.
Longevity robots with biometric monitoring ranked on the standard applied to any monitor — validated sensors, a staffed response and outcome evidence: care-programme telepresence robots, hospital and care-home monitoring robots, home companion robots with health check-ins, autonomous vitals robots, and consumer biometric robots.
How to choose longevity robots with reliable health sensors?
A ranked five-step method for choosing a longevity or care robot with reliable health sensors: decide the job first, verify each sensor's validation and clearance, check who responds to a reading, test the failure modes, and price against the non-robot alternative — with the questions that expose an unvalidated sensor.
High-end longevity tech robots featuring AI health sensors.
High-end longevity robots with AI health sensors ranked on what the premium actually buys: enterprise telepresence and care-programme robots, premium social companions (ElliQ, Lovot, Aibo class), research-grade assistive and mobility robots, luxury home robots with contactless vitals, and imported 'AI doctor' robots — with the evidence, the sensors and the honest price comparison.
What are the top-rated longevity robots for seniors?
Longevity robots for seniors ranked on trial evidence and sustained real-world use: proactive companion robots (ElliQ), therapeutic robots for dementia care (PARO), telepresence robots in care programmes, robotic pets, mobile home robots, and reminder and monitoring robots — with what each has shown and who it suits.
Premium longevity robots with continuous vital sign sensors.
Premium robots claiming continuous vital-sign sensing ranked against what actually delivers continuous vitals — wearables, patches and connected devices: care-programme robots relaying wearable data, companion robots with wearable integration, radar-based presence and breathing robots, camera-based vital-sign robots, and contactless blood-pressure claims — with what 'continuous' should mean.
Which robot sensors are best for long-term elderly care?
Robot and ambient sensors for long-term elderly care ranked on tolerance over years, validation and whether they change a care decision: attached validated devices, bed and presence sensors, ambient activity sensing for frailty, wearable relays, voice and conversation sensors, and camera-based monitoring — with the privacy and dignity trade-offs.
Smart longevity care robots with fall detection sensors.
Care robots with fall detection ranked against what the evidence supports: no fall-detection device is regulator-cleared, US fall deaths rose 21% while the prevention evidence sat unused, and the interventions that cut falls are exercise, hearing aids, cataract surgery and home hazard reduction — with how robot, wearable and pendant detection compare and what a family should actually buy.
Are there medical-grade longevity robots with wearable sensors?
Whether any longevity robot is medical-grade, answered by regulatory status: no consumer care robot is a cleared medical device, but cleared wearables (ECG, AF detection, patches, CGMs) and cleared attached devices exist and can be relayed by a robot — ranked configurations, what 'medical-grade' legally means, and how to check a claim.
Home-use longevity robots with integrated movement tracking sensors.
Home robots with movement tracking ranked on whether the tracking feeds anything with evidence: exercise-coaching robots and screens that deliver balance and strength programmes, ambient gait and activity sensing for frailty trends, wearable activity relays, mobile robots with camera activity logging, and robots that count steps and nothing else — with what movement data should change.
Which AI health smartwatch is best for longevity?
AI health smartwatches ranked for longevity purposes on regulated, validated features rather than marketing: Apple Watch and Samsung Galaxy Watch for cleared cardiac features, Garmin and Whoop for training and recovery data quality, and why no smartwatch AI currently changes a lifespan outcome on its own.
Are Apple Watch or Garmin better for health tracking?
Apple Watch versus Garmin for health tracking, compared feature by feature: cardiac monitoring (cleared ECG features), battery life and continuous tracking, training and recovery metrics, sleep tracking, and which platform is actually better suited to which health goal.

Wearables & devices · 30 →Genomics & DNA testing · 10 →

Explore this section

Which wearable devices support the best longevity tracking?
Wearable metrics graded on accuracy and outcome evidence; watch vs band vs ring.
Longevity clinics
Where devices, panels and monitoring get sold as a package — and how to judge what is inside it.
Public health and policy
Delivery, reimbursement and workforce: the reasons the effective interventions on this page are under-used.
AI and health forecasting
The same surrogate-creep question applied to predictive models: what do they change, not what do they measure.

The Longevity Brief

One evidence-graded email a week: what is new in longevity research, what is hype, and the one change actually worth making.

Free · one email a week · unsubscribe anytime.

Technology for Longer Lives

The highest-value, lowest-glamour list

What actually keeps people independent

The evidence ledger: does it change outcomes?

Technology by tier and by outcome

Balance and functional exercise (Otago-type)

Cataract surgery

Hearing aids, fitted and worn

Home hazard reduction, targeted

Structured remote patient management (heart failure)

Generic telemonitoring bolted onto usual care

Multifactorial fall-risk assessment programmes

Telecare and personal alarm systems

Companion and socially assistive robots

Wearable fall detection

Smart-home and ambient sensing

Exoskeletons for community-dwelling older adults

Medication reminder devices and electronic pill bottles

Consumer brain-training apps