The scan comes back and the ventricles are large. The gait has gone shuffling and magnetic, the memory is patchy, there have been accidents on the way to the bathroom. Somewhere in that room is a question with an operation attached to it: is this idiopathic normal pressure hydrocephalus, where draining cerebrospinal fluid through a shunt can give some people their walking back, or is this Alzheimer's disease, where it cannot? For twenty years the tiebreaker has meant a needle in the back. Now there is a blood test that memory clinics trust for Alzheimer's, plasma phosphorylated tau 217, and the obvious temptation is to run it here. Two papers published in Fluids and Barriers of the CNS on 5 March 2026 are the first serious attempt to ask what that tube of blood is actually worth when the ventricles are big.
The first, by Horikoshi, Nakajima, Miyajima and colleagues, is a cross-sectional comparison of 121 people: 22 healthy controls, 76 people with iNPH, and 23 with Alzheimer's disease, with plasma p-tau217 and amyloid-beta oligomers measured in each. For the job clinicians most want done, telling iNPH apart from Alzheimer's, plasma p-tau217 did respectable work, with an area under the curve of 0.79 (95% confidence interval 0.700 to 0.882). Within the iNPH group, plasma and CSF p-tau217 tracked each other closely (rho = 0.736), which is reassuring about the assay itself. This is a single-centre, cross-sectional comparison, not a trial, and the authors call it the first investigation of these two markers in iNPH.
Then the same test fell over. Asked to separate people with iNPH from healthy controls, plasma p-tau217 returned an AUC of 0.48, which is to say a coin toss, and a coin toss slightly on the wrong side. Amyloid-beta oligomers did the opposite job, separating controls from iNPH only moderately (AUC 0.71, CI 0.567 to 0.848), with a confidence interval that brushes uncomfortably close to chance at its lower end. And for the question that matters most to a surgeon weighing whether Alzheimer's pathology is riding along with the hydrocephalus and blunting the benefit of surgery, all three measures were modest: 0.64 for plasma p-tau217, 0.63 for oligomers, 0.66 for the ratio of the two. Coexisting Alzheimer's pathology here was defined by a CSF p-tau181 threshold above 30 pg/mL, a stand-in for pathology rather than pathology itself, which is worth holding in mind before treating those numbers as harsher or kinder than they are.
The longitudinal arm is the part that has not been done before. Seventy-five patients from the iNPH group had lumbo-peritoneal shunt surgery with both CSF and plasma sampled before and afterwards, over a follow-up ranging from 1 to 42 months (mean 10.0, standard deviation 8.4). Clinically they improved: the iNPH grading scale, the Mini-Mental State Examination and the Frontal Assessment Battery all moved significantly in the right direction. The fluids moved too, in different directions. Amyloid-beta oligomers fell in both CSF and plasma, CSF p-tau217 rose, and plasma p-tau217 sat still. A blood marker that does not budge while a person walks better is a blood marker that is not, on this evidence, measuring the thing the surgery fixed.
The companion paper, from Bermudez, Hofrenning, Fought and colleagues at the Mayo Clinic, shows the confound running back the other way. In 509 participants of the Mayo Clinic Study of Aging (average age 72.2, 54 per cent male, 30 per cent APOE4 carriers) who had MRI, Pittsburgh compound B amyloid PET and mass spectrometry plasma testing, 32 people (6.3 per cent) were classified by a validated automated algorithm as having disproportionately enlarged subarachnoid space hydrocephalus, the DESH pattern that helps select shunt candidates. In age-adjusted logistic regression, abnormally high plasma p-tau217 was associated with DESH (odds ratio 2.74, CI 1.09 to 7.85), as was abnormally low plasma Aβ42 (OR 2.56, CI 1.19 to 5.74). The ratio versions of the same markers were not significantly associated (Aβ42/40 OR 1.95, %p-tau OR 1.82), which is the practical hint buried in the paper. This is a retrospective, cross-sectional analysis of a research cohort, the DESH group numbers 32 people, and the confidence intervals are wide enough to see through.