Dementia care's bottleneck is measurement, not medicine

Diagnosis depends on where you live

A Bayesian spatial analysis of dementia diagnosis in England, published as a preprint, finds that where you live still decides whether your dementia gets diagnosed. Geographic variation in case ascertainment persisted even after the researchers adjusted for differences in underlying disease prevalence — in other words, the gaps are not explained by some areas simply having more dementia. Urbanicity, deprivation, ethnicity and healthcare provision were the factors associated with the variation, and areas with lower ascertainment had higher proportions of undifferentiated dementia diagnoses, cases recorded without a specific subtype. A diagnosis without a subtype is a weaker basis for treatment and planning, so the gap compounds. m:…6110

A national survey of dementia care provision in England, covering home care agencies and care homes, finds the same strain from the provider side. Care homes mostly do the coordination work — 72.5% hold regular multidisciplinary meetings — but providers in both settings named increasing care complexity and behavioural challenges as their primary operational barriers, with workforce recruitment, retention and financial constraints as the systemic barriers underneath. The survey is a baseline, not a verdict, but it frames the problem as capacity rather than knowledge: the system knows what good care looks like and cannot staff or fund it. m:…8002

Measuring decline at home, and in blood

Two more preprints address the measurement problem directly, one in the home and one in blood. The home study tested a room-aware Transformer — a model that processes sensor data room by room — for recognising activities of daily living in adults with mild neurocognitive disorder, using sensor data from five homes. It achieved the highest or tied accuracy in 9 of 15 home-window comparisons against global Transformers and LSTMs, but the global Transformer had marginally higher mean accuracy overall, and sleep and rest were the most consistently recognised activity across all models. The authors' conclusion is deliberately unheroic: no single architecture was uniformly best, and personalised models are probably necessary for clinical use. m:…9361

The blood study is smaller and more striking. In the study, serum carboxymethyllysine — a glycoxidative marker — was significantly lower in people with clinically diagnosed Alzheimer's disease than in healthy controls, and each 1-ng/mL increase in serum CML was associated with lower odds of Alzheimer's, with an adjusted odds ratio of 0.727. On its own the marker discriminated Alzheimer's well; combined with p-tau217, an established marker, discrimination rose to an AUC of 0.966. The authors note that longitudinal validation is still required — a single cross-sectional measurement is a candidate, not a test. m:…0977

The bottleneck is measurement

Read together, the four preprints describe one bottleneck from four directions: diagnosis is uneven, care is under-resourced, home monitoring is promising but not settled, and blood biomarkers are advancing faster than the services that would use them. A marker that discriminates Alzheimer's at an AUC of 0.966 matters little if the person carrying it lives in an area where cases go undifferentiated, or if the care system that would act on the result cannot recruit staff. The technology is not the constraint. The constraint is everything around it. m:…6110 m:…8002 m:…0977

That has a practical implication for anyone building in this space. Home monitoring and blood biomarkers are attractive because they scale — sensors in five homes, a blood draw in a clinic — but both preprints are careful about their limits: the monitoring study says personalisation is necessary, and the biomarker study says longitudinal validation is required. The most useful near-term work may be the least glamorous: closing the ascertainment gap in the places the spatial analysis flags, and funding the workforce the survey says is the binding constraint. A better test only helps if someone is there to give it and act on it. m:…9361 m:…0977 m:…6110