A simple blood draw for Alzheimer’s disease is only as useful as its ability to give the same answer twice. A new study from researchers at University College London has put that question to one of the most rigorous tests yet, measuring the same biomarkers in the same people at two closely spaced time points and reporting that the core markers of Alzheimer’s pathology hold up remarkably well, while many broader neurological proteins show far more wobble than clinicians might expect.
The study, published in Alzheimer’s Research & Therapy, recruited 92 participants from observational research cohorts at the Dementia Research Centre at UCL, including 22 people with a clinical diagnosis of Alzheimer’s disease, 8 with other neurodegenerative diseases, 27 asymptomatic carriers of mutations causing autosomal dominant Alzheimer’s disease, and 35 controls. Each participant provided two serial plasma samples, taken either on the same day or up to 92 days apart, with a median interval of just 85 minutes between draws. The participants had a median age of 52 years, younger than a typical memory clinic population, and 46 percent were female.
The researchers measured 131 biomarkers across three immunoassay platforms: the Lumipulse G plasma p-tau217 assay from Fujirebio, the NULISAseq CNS Disease Panel and the NULISAqpcr AD 5-Plex assay, both from Alamar Biosciences. The high-plex NULISAseq panel, which quantifies 131 neuro-specific and inflammatory proteins simultaneously using a proximity extension assay approach with attomolar sensitivity, had never before been evaluated for short-term test-retest performance, making this the first study of its kind on a high-plex platform.
The headline result concerns the core Alzheimer’s biomarkers: amyloid beta 42, brain-derived p-tau181, brain-derived p-tau217, brain-derived p-tau231, p-tau181, p-tau217, p-tau231, glial fibrillary acidic protein (GFAP) and neurofilament light chain (NfL). Across the assays providing absolute quantification, all of these showed strong correlations between the paired measurements, with Spearman rho values of at least 0.80 and intraclass correlation coefficients of at least 0.92. Within-subject variability, expressed as the coefficient of variation, ranged from 9.7 percent for NfL to 22.0 percent for brain-derived p-tau217 measured on the NULISAqpcr AD 5-Plex assay.
Plasma p-tau217, widely regarded as the leading candidate blood-based diagnostic marker for Alzheimer’s disease, performed particularly well. Measured on the Lumipulse G platform, it achieved a Spearman rho of 0.95 with a 95 percent confidence interval of 0.92 to 0.97, and an intraclass correlation coefficient of 0.94. Its within-subject coefficient of variation was 11.2 percent, notably lower than the 20 percent test-retest variability reported previously for p-tau217 by an earlier study. Brain-derived p-tau217, measured on the NULISAqpcr platform, matched the same correlation and reliability figures but showed roughly double the within-subject variability, a difference the authors attribute at least in part to differences in how the two platforms report their measurements.
Perhaps the most clinically consequential finding involves diagnostic classification. The team applied a validated dual cut-point approach for plasma p-tau217, using thresholds derived from an independent cohort: concentrations above 0.422 picograms per millilitre indicate high Alzheimer’s disease probability, values between 0.153 and 0.422 picograms per millilitre fall in an indeterminate zone, and values below 0.153 picograms per millilitre indicate low probability. When participants were categorised at both timepoints, 92 percent remained in the same category, and no individual moved across two boundaries, meaning nobody flipped from high to low probability or vice versa. Only 8 percent, seven of the 92 participants, crossed a single cut-point boundary, and none of those carried a clinical diagnosis of Alzheimer’s disease.
The researchers also tested whether averaging the two p-tau217 measurements could shrink the indeterminate grey zone, which typically captures around 20 percent of people and triggers referral for confirmatory cerebrospinal fluid or PET testing. It did not. Averaging paired values failed to reduce the proportion of individuals classified as indeterminate, leading the authors to conclude that short-interval repeat blood sampling offers limited benefit for interpreting plasma p-tau217 with these cut-points. For people whose results sit near a threshold, however, the finding carries a practical warning: even small measurement fluctuations can alter classification, so results in that range should be interpreted within the broader clinical context.
Beyond the core Alzheimer’s markers, the picture grew more complicated. On the NULISAseq CNS Disease Panel, 74 of the analytes showed strong correlations of 0.75 or above, 43 showed moderate correlations between 0.46 and 0.74, and 14 showed low correlations of 0.45 or below. Reliability followed a similar spread, with 78 analytes achieving high reliability, 37 moderate and 16 low. The majority, 101 of 132 analytes, had within-subject variability below 20 percent. Notably, both oligomeric and monomeric alpha-synuclein showed low correlation and reduced reliability, and analytes grouped under a synuclein and synaptic disorder ontology performed worse overall than other protein groups. TAR DNA-binding protein, which has been associated with Alzheimer’s disease in prior work, also showed low correlation and reduced reliability. The authors suggest these findings matter most for association studies, where unrecognised short-term variability could be mistaken for a disease signal.
The study also probed whether longer gaps between blood draws increase variability. After adjusting for age and sex and correcting for multiple comparisons, 30 biomarkers on the high-plex panel and one on the 5-plex assay showed significant associations between longer inter-sample intervals and greater within-subject variability. Among them was NfL, a core Alzheimer’s biomarker that tracks neurodegeneration dynamically over time, making some drift over weeks biologically plausible. The authors note that the wide range of intervals, from 30 minutes to 92 days, means same-day pairs likely capture very short-term influences such as time of day and food intake, while longer intervals may reflect gradual physiological change, and further studies are needed to disentangle the two.
Ratio-based measures, often proposed as a way to sharpen diagnostic performance, delivered mixed results. The amyloid beta 42/40 ratio on the high-plex panel was more reliable than amyloid beta 42 alone, with a Spearman rho of 0.85 versus 0.79. But the brain-derived p-tau217/amyloid beta 42 ratio on the 5-plex assay did not improve reliability relative to brain-derived p-tau217 alone, and only modestly reduced variability, from 22.0 percent to 21.1 percent. Given that the first FDA-approved plasma biomarker test for Alzheimer’s disease incorporates a p-tau217/amyloid beta 42 ratio, the authors argue that ratios should be adopted clinically only with clear evidence of added value, particularly since pre-analytical variability affects several of the constituent analytes.
Cross-platform agreement was strong throughout. P-tau217 measured on the Lumipulse G correlated at 0.88 with p-tau217 on the NULISAseq panel, and brain-derived p-tau217 correlated at 0.90 between the two NULISA platforms, consistent with previously reported cross-platform concordance ranging from 0.78 to 0.94. The study has limitations the authors acknowledge: the cohort was younger than a typical memory clinic population and included relatively few people with symptomatic neurodegenerative disease, which may underestimate the variability seen in older patients with co-pathologies. Participants were not required to fast, and factors such as body mass index and chronic kidney disease were not accounted for, though these were unlikely to change substantially over the short sampling intervals. The authors call for replication in larger, biomarker-defined cohorts that control for fasting status and sample timing. For now, the message is one of reassurance with a caveat: the blood tests anchoring Alzheimer’s diagnosis give stable answers over days to weeks, but the wider universe of CNS proteins they can now measure in a single panel demands more careful interpretation.
Subject of Research: Short-term variability and test-retest reliability of Alzheimer's disease plasma biomarkers measured on single-, low- and high-plex immunoassays
Article Title: Short-interval variability and test-retest reliability of Alzheimer’s disease plasma biomarkers in a clinically heterogenous cohort
Article References: Beament, M., Wiltshire, K., Belder, C. R. S., Abel, E., Hart, M., Heslegrave, A. J., Lunn, M. P., Rosa-Grilo, M., Swann, O., Schott, J. M., Zetterberg, H., Fox, N. C., & Keshavan, A. (2026). Short-interval variability and test-retest reliability of Alzheimer’s disease plasma biomarkers in a clinically heterogenous cohort. Alzheimer's Research & Therapy, 18(1), Article 238. https://doi.org/10.1186/s13195-026-02180-8
Image Credits: AI Generated
DOI: 10.1186/s13195-026-02180-8
Keywords: Alzheimer's disease, plasma biomarkers, p-tau217, test-retest reliability, NULISA, blood-based biomarkers, neurofilament light chain, GFAP, diagnostic cut-points, within-subject variability, Lumipulse, proteomics
News Source: Diana Fleming. (October 28, 2026). Research report: Short-interval variability and test-retest reliability of Alzheimer’s disease plasma biomarkers in a clinically heterogenous cohort. Scienmag.
