Design: Retrospective single-centre memory-clinic study with reference biomarkers
Population: 706 patients with plasma p-tau217 and amyloid-PET or CSF confirmation
Primary outcome: Factors associated with discordant plasma vs reference AD biomarkers
Effect: Discordance 8.5% (false negative 5.6%, false positive 2.8%); tied to age, renal function, BMI, stage

Plasma p-tau217 is spreading fast as an Alzheimer's test because it is cheap and accessible, but a Mayo Clinic study of 706 memory-clinic patients shows where it misleads. Against amyloid-PET or CSF, plasma p-tau217 was discordant in 8.5%, with false negatives (5.6%) more common than false positives (2.8%).

The discordance was not random. False negatives clustered in milder disease, higher body-mass index and better renal function; false positives in older age, thyroid disease, lower renal function and advanced disease. In other words, body size and kidney function — which change the concentration and clearance of the marker — and disease stage shift the result away from the truth.

The practice point is to read a plasma result in that context: in a patient who is older, has impaired renal function, thyroid disease, a high body-mass index or a mismatched clinical stage, confirm with PET or CSF before letting the plasma value drive a diagnosis or the decision to start a disease-modifying drug.

Plasma p-tau217 disagreed with PET or CSF in about 1 in 12 memory-clinic patients.
False negatives were commoner in milder disease, higher BMI and better renal function.
False positives were commoner with older age, thyroid disease, lower renal function and advanced disease.
Confirm a plasma result with PET or CSF in these groups before acting on it.

Why it matters

It marks out exactly which patients a cheap, popular blood test is most likely to mislead, before that error reaches a disease-modifying drug.

The statistics, in plain English

An 8.5% discordance rate means roughly one in twelve results is wrong against the reference standard; because the errors cluster in identifiable groups rather than scattering randomly, you can predict in advance who needs a confirmatory test.