The scene is familiar to anyone who has sat in on a memory appointment. A daughter has been keeping a list on her phone: the missed turn on a road driven for thirty years, the same story twice in an afternoon, the electricity bill paid twice. The doctor has a clock face on a sheet of paper and a stopwatch, and about as long as it takes to boil a kettle. Increasingly there is a third party in the room, a tablet or a wearable or an app that has been quietly counting steps, typing speed and time out of the house since spring. Everybody in that room wants the same thing, which is an answer. A review published on 7 July 2026 in Frontiers in Neurology argues that the paper test, the app and the daughter's list are not competing versions of the same answer at all, and that clinics get into trouble when they pretend otherwise.
The review is by Miren Altuna, whose listed affiliations include the Department of Neurology at Bioaraba Health Research Institute, Araba University Hospital-Txagorritxu in Vitoria-Gasteiz, and the Faculty of Health Sciences at the University of Deusto in Bilbao, with correspondence through the CITA-Alzheimer Foundation. It is a narrative review, which means it is a structured argument built from the published literature rather than a new experiment with patients in it, and that distinction matters for how much weight a family should put on it. What it offers is a map of the toolkit. It starts with the instruments most people will actually meet: the Mini-Mental State Examination, the Montreal Cognitive Assessment, the Mini-Cog, the Clock Drawing Test, the Fototest, the Memory Alteration Test, the Eurotest and the informant questionnaire AD8. These persist, the review notes, because they are cheap, interpretable and realistic inside a routine appointment. Their limitation is baked into their birth: most were built to identify cognitive impairment that has already established itself, not the subtler territory of mild cognitive impairment or Alzheimer's disease defined biologically, before the syndrome is obvious.
And they are not interchangeable with each other, which is the first conflation the review pulls apart. Some brief tools sweep several cognitive domains; others lean deliberately on memory, because episodic memory failure is a common early feature of typical Alzheimer's disease. The Memory Alteration Test was designed that way, for amnestic mild cognitive impairment, and the review is blunt that its strength and its weakness are the same trait: a memory-centred instrument can miss presentations that do not begin with memory. Education cuts across all of it. High educational attainment and cognitive reserve can mask real impairment on a brief test, which is why a professor who scores 28 out of 30 and a retired farm worker who scores 24 are not telling you the same thing.
The digital half of the argument is where the review is most useful to a section like this one. It draws a line between digital phenotyping, which is profiling across cognitive, functional, behavioural, linguistic, motor and interaction domains using both active tasks and passive sensing, and digital biomarkers, which are the individual measurable signals inside that profile: pauses in speech, response times, pen movements, gaze patterns, mobility features. The promise sits in four words: standardisation, granularity, scalability and longitudinal measurement. A screen taken once in a fluorescent-lit room is a snapshot; a stream of data is a slope. But the review is unsparing about what can go wrong, listing performance figures inflated by selected samples, poor generalisation to new settings, systematic errors in underrepresented groups, exclusion of people with limited digital access or sensory impairment, misuse of sensitive behavioural data, and premature reliance on automated outputs without adequate clinical oversight. A digitised version of a biased task, it notes, inherits the bias.
So what are the digital signals for? Triage, and tracking. The review's position is that these signals may sharpen risk stratification and monitoring over time, yet they do not resolve diagnostic uncertainty or establish the cause of a problem by themselves, and it argues digital screening belongs in practice as a "governed triage and phenotyping layer rather than a stand-alone diagnostic label". What follows an abnormal screen, digital or paper, is a staged pathway: informant and functional measures, neuropsychiatric assessment, then biological triage with plasma biomarkers and MRI to decide whether specialist referral, cerebrospinal fluid testing or amyloid and tau PET are warranted. The review also sets an evidence bar for the tools themselves, insisting a high area under the curve in a selected sample is not sufficient, and that validation should match intended use across technical, psychometric, construct, diagnostic, biomarker and implementation domains, with equity, governance and post-market monitoring running through all of them. It flags the ethics too, in language families will recognise instantly: consent granularity, household privacy, commercial access to the data, and the prospect of cognitive surveillance inferred from ordinary daily behaviour.
For a family, the practical translation is small but freeing. If a relative brings six months of app data to an appointment, that data is a reason to look harder, not a verdict, and it is not failing anyone if the clinician still wants the paper test, the conversation with the person who knows the patient best, and, where appropriate, blood work or imaging. If a brief screen comes back reassuring in someone whose family is quietly certain something has changed, the review is on the family's side about why that can happen. The eleven-minute visit does not get longer because the data got bigger. What changes is which question each tool is being asked to answer.