The Lumipulse assay measures the ratio of phosphorylated tau at threonine 217 (p-tau217) to amyloid-beta 1−42 in blood plasma, and was cleared as the first blood test to help evaluate Alzheimer’s disease in symptomatic adults aged 55 and older. Its approval coincides with the arrival of anti-amyloid therapies such as lecanemab and donanemab, making timely evaluation increasingly important. Yet the Barcelona team emphasizes that the clearance was based largely on performance in enriched cohorts of carefully pre-screened patients, and that real-world accuracy may be substantially lower. At a disease prevalence of 10 percent, typical of some memory clinics, even a test with 90 percent sensitivity and 90 percent specificity would yield a positive predictive value of only 50 percent, meaning half of positive results could be false. In primary care settings, where prevalence may be as low as 1 to 5 percent, the positive predictive value falls to roughly 8 to 32 percent, rendering the test unsuitable for screening.

The molecular critique centers on what p-tau217 actually represents. Tau, the neuronal protein whose abnormal aggregation characterizes Alzheimer’s disease, carries at least 85 distinct phosphorylation sites identified by mass spectrometry, and phosphorylation at threonine 217 is not confined to diseased brains. It is detectable in cognitively normal individuals, and studies of tau phosphorylation have found that the patterns of physiological and pathological tau are, in the authors’ words, surprisingly similar and heterogeneous, making it difficult to identify specific modifications as reliable biomarkers. Compounding this, plasma tau exists as a mixture of full-length protein and truncated fragments generated by calpain- and caspase-mediated cleavage, each with different phosphorylation patterns, stabilities and blood-brain barrier permeability. Current immunoassays cannot distinguish among these species, so the measured p-tau217 signal is a composite of uncertain molecular composition rather than a single defined entity.

The denominator of the ratio fares no better in their assessment. Amyloid-beta 1−42 in blood is an intrinsically disordered peptide that rapidly binds carrier proteins such as albumin and lipoproteins, masking epitopes and making measured concentrations highly dependent on assay format and sample handling. Reported plasma concentrations vary enormously across platforms, from 2.72 to 11.09 pg/mL on one ultrasensitive platform to 8.12 to 29.00 pg/mL on another, a fourfold spread even among healthy controls. Critically, a large fraction of circulating amyloid-beta appears to be produced outside the brain: platelets contain more than 90 percent of the blood’s amyloid precursor protein and release amyloid-beta upon activation. Estimates of the brain-derived share of plasma amyloid-beta 1−42 range from under 10 percent to about 30 percent, depending on blood-brain barrier integrity, weakening the peptide’s linkage to actual brain amyloid burden.

Pre-analytical variables compound the problem. Hemolysis can raise p-tau217 while lowering amyloid-beta 1−42, platelet contamination has a high to very high impact on both analytes, and the timing of centrifugation, storage temperature and repeated freeze-thaw cycles all introduce systematic variability that may approach or exceed the biological differences between diagnostic groups. Even under standardized protocols using preferred EDTA tubes, the coefficient of variation for amyloid-beta 1−42 can reach 15 to 25 percent.

The ratio itself introduces a further layer of statistical fragility. Standard error-propagation theory dictates that the coefficient of variation of a quotient approximates the square root of the sum of the squared coefficients of its numerator and denominator. Two measurements each carrying 10 percent variability yield a ratio with roughly 14 percent variability, before any biological signal is considered. Moreover, because the ratio is hyperbolically sensitive to its denominator, small fluctuations in amyloid-beta 1−42 produce disproportionately large shifts: a 10 percent drop in the denominator inflates the ratio by about 11 percent, a 20 percent drop by about 25 percent. The authors illustrate the stakes with concrete numbers. Holding p-tau217 constant at 0.26 pg/mL, the reported healthy-range amyloid-beta values yield ratios spanning 0.023 to 0.096, a roughly fourfold spread driven entirely by the denominator. In the BALTAZAR study, with p-tau217 fixed at 1.45 pg/mL, the confidence interval for amyloid-beta 1−42 in patients produces ratios spanning more than a fourfold range. Two individuals with identical p-tau217 values could therefore receive different biomarker classifications solely because of denominator variation.

There is also a deeper chemical objection. The Lumipulse assay treats the quotient as a mass-per-mass, or pg/pg, ratio. Such a ratio is only meaningfully interpretable if both components represent single, chemically well-defined species with constant molar mass. Because plasma p-tau217 is a mixture of tau fragments, the effective molar mass is not fixed, and the mass ratio cannot be converted into a molar ratio. Apparent algebraic cancellation of units, the authors argue, does not by itself confer biological interpretability. The FDA clearance, they stress, validates one platform under defined conditions, not the ratio as an assay-independent diagnostic standard.

The review also mounts a methodological critique of how Alzheimer’s biomarkers are validated more broadly. A recent meta-analysis found that approximately 90 percent of studies of plasma p-tau217 were judged at high risk of bias because diagnostic thresholds were optimized after the fact rather than prespecified or externally validated. Validation studies routinely preselect participants in well-defined categories, excluding diagnostically ambiguous individuals with mixed pathologies, atypical presentations and comorbidities, so reported area-under-the-curve values are best viewed as upper-range estimates from favorable conditions. More troubling still is what the authors call the circular validation trap: plasma assays are validated against amyloid PET, PET against cerebrospinal fluid markers, CSF against clinical diagnosis, and clinical thresholds are then refined using neuropathology, even though clinical diagnosis shows 30 to 60 percent discordance with autopsy findings. Each imperfect method supports the next, creating an appearance of reliability without fully independent verification, a self-reinforcing loop in which internal coherence is mistaken for correctness.

The consequences extend beyond misdiagnosis of individuals. If trial eligibility is determined by ratio-driven thresholds, denominator-driven shifts could move biologically similar individuals across enrollment cutoffs, lowering the proportion of truly target-positive participants, diluting observed treatment effects and distorting subgroup analyses in trials of new Alzheimer’s therapies. Misclassification propagates into treatment-effect estimates and downstream validation, amplifying the stakes of the analytical weaknesses described.

The authors do not dismiss blood-based biomarkers, which remain promising. Instead they propose more statistically principled alternatives: modeling p-tau217 and amyloid-beta 1−42 jointly as separate but correlated variables with distinct error structures, within multivariable or Bayesian frameworks that propagate measurement uncertainty rather than hiding it inside a quotient. Future assays could become fragment-aware and proteoform-aware, combining immunocapture with targeted mass spectrometry to resolve the composite signal. In the meantime, they argue the ratio’s most defensible use lies in longitudinal monitoring within the same individual, where directional change over time may be more informative than a single threshold-based result, and as a contextual triage tool rather than a standalone diagnostic benchmark, so that precision medicine does not rest on associations whose causal basis remains insufficiently established.