Three times a day for fifteen days, a text message arrived. Participants had half an hour to open it and play what looked like a game: match digits to symbols as fast as you can, press for city scenes and withhold for mountains. On their arm or abdomen, a continuous glucose monitor kept its own running record. That design, published by Zoe Hawks and colleagues in npj Digital Medicine on 18 March 2024, captured 200 adults with type 1 diabetes going about ordinary life, and it confirmed in the wild what laboratories had long shown in controlled conditions: reaction time slows when glucose runs low and when it runs high. Plot one person's performance against their own glucose and you get a shallow U. Plot another person's and the U is steep. The steepness is the point. The researchers named it cognitive vulnerability to glucose, or CVG: a number for how much your thinking wobbles when your sugar moves.
The effects at stake are small in absolute terms. In that first paper, optimal processing speed was only 5.30 milliseconds faster than a person's own average, a difference the authors compared to the effect sizes seen in sleep restriction studies. No one notices five milliseconds. But the question that followed was not whether the wobble is felt. It was whether the wobble is a signal of something happening in the brain tissue underneath.
That is what Luciana Mascarenhas Fonseca and colleagues went looking for in Alzheimer's & Dementia, published online on 24 February 2025. They took 114 of the same adults, those who had also given a blood sample, and measured five plasma markers: the amyloid beta 42/40 ratio, tau phosphorylated at threonine 181 and at 217, neurofilament light chain (a marker of injured neurons), and glial fibrillary acidic protein (a marker of reactive astrocytes). Mean age was 48.9 years, with a range of 19 to 84; 60 of the 114 were female; 102 were White. Nobody with a dementia diagnosis in their medical record was included. This was a cross-sectional study: one set of blood draws, one window of cognitive testing, no follow-up over years.
The pattern that emerged was not uniform, and the differences between the two tasks matter more than the headline. Vulnerability measured on the digit symbol task was associated with GFAP (B = 0.104, 95 percent CI 0.062 to 0.145, explaining 18.6 percent of the variance) and with p-tau181 (B = 0.012, CI 0.005 to 0.019, 10.6 percent), but that version of CVG was also strongly tied to age itself (Spearman rho = 0.72), which makes it hard to separate the wobble from the simple fact of getting older. Vulnerability measured on the sustained attention task was not correlated with age at all (rho = -0.01), and it was the one that held: associations with p-tau181 and p-tau217 persisted across models adjusted for demographics, mean glucose, mean cognition, and the presence of nephropathy. For p-tau217 the p values hovered at the edge, from 0.049 to 0.101 across specifications. The amyloid ratio showed nothing, which is consistent with it also being nearly uncorrelated with the other four markers in this sample (rho between -0.08 and 0.09).
What the authors will not claim, and neither will we, is direction. Their own discussion puts it flatly: it remains unclear whether cognitive vulnerability to glucose contributes to neurodegeneration or arises from it. A brain already accumulating tau may simply have less reserve to absorb a glucose swing, in which case the wobble is a symptom and not a cause. The authors also list their own constraints: a modest sample, predominantly non-Hispanic White participants, only two cognitive domains tested, nephropathy established by chart review rather than kidney function tests, and no formal assessment for mild cognitive impairment.
For a family living with type 1 diabetes, the practical meaning today is close to zero and the research meaning is considerable. There is no test to ask for, no threshold to cross, no number on a CGM app that corresponds to any of this. What there is, for the first time in the same people, is a method: passive glucose data, cognitive testing that happens on a phone in a kitchen rather than in a clinic room, and blood markers that until recently required a lumbar puncture or a scan. If the sustained attention finding survives replication in a larger and more diverse group, and if anyone follows these participants forward in time, the wobble could become a cheap, repeatable metric where none currently exists. That is a long way from a risk score. It is not nothing.