Diagnosis & Biomarkers

1. The Diagnostic Revolution

For most of the twentieth century, Alzheimer’s disease was a clinico-pathological diagnosis: a clinical syndrome of progressive amnestic dementia, confirmed only at autopsy by counting plaques and tangles. The NINCDS-ADRDA criteria (McKhann et al., 1984) formalised this approach — clinical “probable AD” was the best you could do in life, with sensitivity ~80% but specificity only ~70% against autopsy gold-standard. A meaningful fraction of clinically diagnosed AD turned out, at autopsy, to be DLB, FTD, vascular disease, or hippocampal sclerosis (LATE).

Three developments overturned this paradigm:

• CSF biomarkers (1990s–2000s) — Aβ42 falls and tau rises in AD CSF, allowing antemortem confirmation of pathology.
• Amyloid PET (2004) — Pittsburgh Compound B (PiB; Klunk & Mathis) imaged plaque burden directly in living brain.
• The pre-symptomatic recognition — biomarker-positive cognitively-unimpaired adults (DIAN, A4) revealed that pathology accumulates 15–25 years before symptoms.

The NIA-AA Research Framework(Jack et al., Alz Dem 2018) cut the Gordian knot by redefining AD as a biological entity: present whenever amyloid and tau pathology are demonstrable, regardless of clinical state. The clinical syndrome (cognitively unimpaired, MCI, dementia) is layered on top of, but no longer constitutive of, the disease.

2. The ATN Framework

The 2018 NIA-AA framework reduces AD pathobiology to three orthogonal binary axes:

The decisive rule: A+T+ = AD (regardless of N). A+T− is “Alzheimer’s pathologic change” (early or atypical). A−T+ is not AD — it is some other tauopathy (PSP, CBD, primary age-related tauopathy / PART, or chronic traumatic encephalopathy). The eight ATN combinations:

The 2024 revision (Jack et al., Alz Dem 2024) added an “I” (inflammation) axis and reorganised core biomarkers as Core 1 (early plaque-related, including plasma p-tau217 and Aβ42/Aβ40) versus Core 2 (mature tangle pathology, e.g. tau PET). Core 1 alone is now sufficient for diagnosis — effectively endorsing plasma p-tau217 as a stand-alone biological diagnostic.

3. Cognitive Assessment

Despite the biomarker era, bedside cognitive testing remains the entry point. Three instruments dominate clinical use:

Cutoffs are population-dependent. A common MoCA pattern in early AD: deficits in delayed recall and visuospatial (cube, clock), with relatively preserved orientation and language. A patient with MoCA 24 and a strong amnestic profile has a higher pre-test probability of AD than one with MoCA 24 and a dysexecutive profile (in whom vascular or FTD becomes more likely).

Neuropsychological batteries (WMS, RAVLT, BVMT-R, Trails, BNT) refine the cognitive phenotype but are not necessary for routine diagnosis when biomarkers are available. They remain critical for detecting MCI in highly educated patients in whom a normal MMSE may mask substantial decline from baseline.

4. CSF Biomarkers — flagship

For two decades CSF analysis was the most accurate antemortem AD test. The classical “CSF triad” reflects the biology directly:

• Aβ42 ↓ — soluble Aβ42 is sequestered in plaques, lowering CSF concentration.
• p-tau (181, 217, 231) ↑ — phosphorylated tau released from neurons during AD pathology.
• Total tau (t-tau) ↑ — non-specific marker of neuronal injury (also raised in CJD, stroke).

Cutoffs depend on assay. For the Roche Elecsys (the most commonly used electrochemiluminescence platform) on lumbar CSF, typical cutoffs from Hansson et al. (Alz Dem 2018) and the BioFINDER cohorts:

The single most informative number is the p-tau / Aβ42 ratio: it cancels out pre-analytic Aβ losses (peptide adsorption to plastic surfaces), normalises across labs, and gives near-PET-equivalent accuracy:

The Aβ42/Aβ40 ratio works because CSF Aβ40 is far more abundant (~10×) than Aβ42 and serves as a per-patient normaliser of total Aβ production:

Assay history. The first CSF Aβ42 ELISA came from Innogenetics (Innotest) in the late 1990s — manual, labour-intensive, with substantial inter-lab variability that undermined adoption of universal cutoffs. The Fujirebio Lumipulse chemiluminescent enzyme immunoassay (CLEIA) automated the workflow with much tighter reproducibility; FDA cleared the Lumipulse G β-Amyloid Ratio (1-42/1-40) test in 2022 for AD diagnosis in adults aged 55+ being evaluated for cognitive decline. The Roche Elecsys platform provides comparable automated quantification for Aβ42, Aβ40, p-tau181 and t-tau and has dominated trial cohorts (ADNI, BioFINDER, A4).

See also Part III for the underlying APP processing biology that makes Aβ42 the “sticky” isoform that ends up in plaques, and Part IV for the tau hyperphosphorylation biology that drives p-tau release.

5. Amyloid PET

Amyloid PET visualises fibrillar Aβ plaque burden directly. Klunk and Mathis (Ann Neurol 2004) introduced Pittsburgh Compound B (PiB), a thioflavin-T derivative labelled with carbon-11. PiB binding correlates tightly with neuropathological plaque counts and remains the research reference. Its 20-minute ¹¹C half-life confines it to centres with on-site cyclotrons, so three ¹⁸F-labelled tracers (110-min half-life, distributable) carry clinical practice:

Two readout strategies coexist:

• Visual read — certified reader scores positive/negative based on cortical-vs-white-matter contrast. Used clinically; concordance with autopsy ~96%.
• SUVR — standardised uptake value ratio (cortex / cerebellar grey reference). Used in research and increasingly in clinical trials.
• Centiloid scale — tracer-independent normalisation (Klunk et al., Alz Dem 2015): 0 = young controls, 100 = typical mild AD. Centiloid > 24 is a widely used positivity threshold; > 90 may indicate severe burden affecting anti-amyloid therapy decisions.

Amyloid-PET sensitivity is ~92%, specificity ~95% versus autopsy. The dominant clinical use case is now therapy eligibility: lecanemab and donanemab labels require demonstrated amyloid pathology (PET or CSF) before initiation. Cost (~$3000–5000) and limited availability in many health systems remain the primary barriers; this is precisely the gap that plasma p-tau217 is poised to fill.

6. Tau PET

Tau PET arrived ~10 years after amyloid PET and has fundamentally changed the picture: tau-PET signal correlates far better with cognitive decline and rate of progression than amyloid does. The first-generation tracer flortaucipir (T807, AV-1451) was approved by the FDA in 2020 as Tauvid. Second-generation tracers reduce off-target binding to MAO and choroid plexus:

Spatial pattern matters. Flortaucipir uptake follows the Braak staging known from neuropathology (Part IV): entorhinal cortex (Braak I/II) → limbic (III/IV) → temporoparietal neocortex (V) → primary sensory cortex (VI). The Mayo tau-PET staging system maps SUVR regions onto these stages and predicts cognitive trajectory: a patient with widespread Braak V pattern has a much steeper decline slope than one with isolated Braak I/II uptake.

Tau PET also stratifies eligibility for therapy. The donanemab TRAILBLAZER-ALZ-2 trial used flortaucipir to define low/medium tau (where the drug worked best) versus high tau populations. Patients with high tau-PET burden showed less benefit, suggesting a window of opportunity that closes once tangle-driven neurodegeneration is widespread.

7. Plasma Biomarkers — flagship: the 2023+ revolution

The single biggest practical change in AD diagnosis since amyloid PET is the arrival of plasma p-tau217 as a high-accuracy, scalable, low-cost biomarker. A simple venous draw, processed by mass spectrometry or immunoassay, now approaches CSF and PET accuracy for detecting amyloid pathology — opening AD diagnosis to primary care.

The plasma biomarker panel:

The leading mass-spec assay is C2N Diagnostics’ PrecivityAD2, which combines plasma Aβ42/Aβ40 and the p-tau217 / non-phosphorylated tau217 ratio (%p-tau217) into a single Amyloid Probability Score (APS2). In the SCREEN-D and validation cohorts (Hu et al., JAMA 2024), APS2 achieved AUC ~0.95 against amyloid-PET in symptomatic patients across 18 cohorts. Janelidze, Hansson and colleagues (Nat Med 2020; JAMA 2023) showed plasma p-tau217 alone outperformed all other plasma analytes for amyloid-PET prediction and increased ~8-fold from cognitively normal to AD dementia — a far larger dynamic range than plasma Aβ42/Aβ40 (~15%).

Two-cutoff strategy. Because plasma p-tau217 has so much dynamic range, a two-threshold approach delivers near-100% accuracy on the “clear” cases and triages only an intermediate band for confirmatory CSF or PET:

• Below low cutoff — ~95% NPV: amyloid-negative; do not refer.
• Above high cutoff — ~95% PPV: amyloid-positive; proceed to therapy evaluation.
• Intermediate (“grey zone”) — ~15–20% of patients: confirm with CSF or amyloid PET.

Palmqvist, Hansson et al. (JAMA 2024) showed that this strategy applied in primary care reduced misdiagnosis from ~30% (clinician judgement alone) to ~10%, with similar accuracy in primary as in specialist memory clinics. That is the practical payoff: AD diagnosis no longer requires a lumbar puncture or a $5000 PET scan for the majority of patients.

Caveats.

• Renal function — p-tau217 is partly cleared renally; severe CKD raises plasma levels and can produce false positives.
• Comorbid neurodegeneration — co-existing FTD or DLB can produce intermediate values without amyloid.
• Ethnicity / pre-analytics — cutoffs derived in Swedish and US cohorts; validation in diverse populations is ongoing. Tube type (EDTA vs P-100) and processing time matter.

8. MRI Volumetry

Structural MRI is the “N” of the ATN framework most accessible in practice. Beyond exclusion of vascular disease, tumour, NPH and subdural haematoma, volumetric quantitation contributes prognostic and supportive information.

Key AD-typical features:

• Hippocampal atrophy — the canonical sign. Visual rating: Scheltens medial-temporal-atrophy (MTA) score 0–4. Volumetric: hippocampal volume < 5–10th percentile for age, normalised to intracranial volume.
• Entorhinal cortex thinning — the earliest detectable atrophy in AD; precedes hippocampal volume loss.
• Posterior cingulate / precuneus — default-mode network hub.
• Temporoparietal cortical thinning — the “AD signature” pattern (Dickerson et al., Cereb Cortex 2009): inferior parietal, supramarginal, superior temporal, precuneus, inferior frontal.
• Whole-brain atrophy rate — serial MRI gives ~2.5%/yr in AD versus ~0.5%/yr in healthy aging.

The boundary score (Fox / Freeborough boundary-shift integral) and registration tools (FreeSurfer, FSL-FIRST, NeuroQuant, Brainreader, Combinostics cNeuro) automate these measurements clinically. NIA-AA does not specify a single cutoff for “N positivity” on MRI, but a hippocampal volume Z-score < −1.5 vs age-matched controls is a common research definition.

9. FDG-PET

[¹⁸F]fluorodeoxyglucose PET measures cerebral glucose metabolism — a sensitive, if non-specific, “N” biomarker that often reveals functional change before structural atrophy. The AD-typical pattern is bilateral temporoparietal and posterior cingulate / precuneus hypometabolism, with relative sparing of primary sensorimotor and visual cortex.

Differential signatures:

FDG-PET is a strong tool for differentiating AD from FTD when clinical features overlap (~96% sensitivity, ~85% specificity in Foster et al., Brain 2007). It is reimbursed for this indication in many systems (CPT 78608 with FTD workup criteria in the US). Tau PET, where available, has largely supplanted FDG-PET for AD-specific staging because it directly visualises the disease protein.

10. Putting It Together — A Modern Diagnostic Algorithm

In 2026, a typical workup for cognitive complaint > age 60 unfolds along three tiers:

Cutoff cheat-sheet:

Take-home rules of thumb:

• If plasma p-tau217 / APS2 is firmly negative, AD is essentially excluded as the cause of cognitive symptoms; pursue alternative diagnoses.
• If clearly positive in an MCI or mild-dementia patient, proceed to therapy work-up; CSF/PET is now confirmatory rather than mandatory.
• Grey-zone p-tau217 → CSF Aβ42/Aβ40 + p-tau181 (or amyloid PET).
• Tau PET is reserved for prognostic stratification and therapy selection (especially donanemab eligibility).
• FDG-PET is the workhorse when the differential includes FTD or DLB.
• MRI is mandatory in every patient — both for excluding mimics and for ARIA-baseline before anti-amyloid therapy (Part VII).