ONKI shows older adults how their own typing rhythm is changing, relative to their peers. It makes no diagnosis and sets no threshold. We are testing whether that alone is enough to prompt someone to go for screening.
Keyboard on. Memory on. Onki (온기) means warmth in Korean — the kind a neighbour once offered by noticing something and saying so.
Korea already has the infrastructure. Community dementia centres are in place and CIST — the national cognitive screening test — is free. Participation among older adults is still low.
The barrier is not access or cost. What is missing is the prompt: the moment someone thinks, maybe I should go and get checked.
Before treatment comes diagnosis. Before diagnosis comes screening. And before either, awareness. ONKI works only at that last step. It collects keystroke timing passively during ordinary phone use, and shows a person how their own rate of change compares with others in their age group. No risk score, no classification, no recommendation — the inference is left entirely to the individual.
The comparison logic is borrowed from audit: not an absolute threshold, but a deviation from what was expected of this particular case.
Jung Woo Hong (presenting), Okran Yim, Chang Nam Lee, Yong Jeong Yi
Background. Before treatment comes diagnosis, before diagnosis screening, and before either, awareness. In super-aged Seoul, dementia screening reaches only a fraction of older adults, and most early mild cognitive impairment (about 28% prevalence) goes undetected. A recent cluster-RCT showed nudges improved dementia-prevention behaviours (Aravena et al., JAGS 2026), but delivered them non-specifically; targeting whom to prompt remains unaddressed, and daily-test designs lose the highest-risk elderly to attrition. Crucially, the intervention neither diagnoses nor screens: absolute typing speed is irrelevant, being inherently slow is fine, and only those slowing faster than their age peers are gently prompted. The aim is motivational, not diagnostic.
Methods. Stratified, block-randomised 2×2 factorial RCT crossing a screening nudge with peer-referenced social comparison; 500 adults aged 67 to 72, stratified by caregiver-alert consent. A wellness keyboard passively records typing timing only (hold time, flight time). Unlike large device-based detection studies pursuing classification, the endpoint is screening behaviour: the model detects within-cohort relative deceleration versus same-age peers, so habitually slow users are never labelled. Recruitment runs through municipal senior digital-education programmes; referrals go to district dementia-safety centres, layering onto existing infrastructure without new facilities. Primary outcome: screening uptake (CIST, intention-to-treat); exploratory secondary: cognitive-test (MoCA-K) enrichment in the top passive-index group.
Anticipated results. We hypothesise the combined nudge-plus-comparison arm yields the highest uptake, the design isolating each component and their interaction. IRB-approved (SKKU 2026-04-078); prospectively registered (CRIS, a WHO ICTRP Primary Registry, Korea; KCT0011992); recruitment from 2026.
Conclusion. If effective, this offers a non-stigmatising, infrastructure-light model to raise screening uptake: digital neighbourly care reconnecting isolated elders to peers and public services, transferable across ageing Asian cities.
Relevance. It applies behavioural science (nudging, social comparison) to a high-burden screening behaviour and tests an implementation strategy: a fully passive design embedded in existing infrastructure for equitable reach, addressing the know–do gap.
Abstract to appear in an Implementation Science supplement.
Conflict of interest. The presenting author holds the patent on the tool under study (KR 10-2025-0049943; PCT/KR2026/004436), declared in the IRB submission and the trial registration.
Protocol approved and registered. Recruitment has not yet begun.
The app has been brought in-house, the collection pipeline rebuilt for timing data only, and the analysis plan is being finalised before enrolment locks the protocol. Updates will be posted here.
Academic and clinical collaborators will be listed here as the trial proceeds.