# Critical Evaluation of CSF p-tau217 Surrogate Endpoint Hypotheses
## Executive Summary
The presented hypotheses collectively address a clinically important gap—establishing evidence-based stopping rules for donanemab. However, the theoretical scaffolding supporting these hypotheses contains several logical gaps, confounds, and unverified assumptions that require rigorous falsification before translation to clinical practice.
---
## Hypothesis 1: Amyloid → Tau Cascade Mechanism
### Weak Links
**Temporal Causality Not Established**: The cited mechanism assumes amyloid clearance *causes* reduced neuronal injury, which *causes* p-tau217 decline. However, the supporting evidence (TRAILBLAZER-ALZ) demonstrates mere temporal precedence—amyloid clears before clinical benefits—without establishing p-tau217 as the mediating biomarker. Correlation of sequence ≠ mechanistic causation.
**p-tau217 Production Source Ambiguity**: The hypothesis implicitly assumes p-tau217 exclusively reflects *ongoing* neuronal injury. However, p-tau217 may predominantly derive from *existing* neurofibrillary tangle burden rather than active pathology. If so, amyloid clearance would not predictably reduce CSF p-tau217 levels.
**Specificity vs. Mechanism**: The PyClarity study establishes p-tau217 specificity for amyloid *status*, not for amyloid *driving* tau pathology. This is a category error—detecting amyloid association ≠ proving mechanistic dependence.
### Counter-Evidence
- Primary age-related tauopathy (PART) demonstrates tau pathology can propagate independently of amyloid
- Autopsy studies show substantial NFT burden in amyloid-negative elderly subjects
- The mechanistic pathway "amyloid → neuronal injury → p-tau217 release" lacks direct neurophysiological validation
### Falsifying Experiment
**Test**: Administer donanemab to patients with cleared amyloid (PET-negative) but elevated p-tau217. If p-tau217 remains elevated despite amyloid absence, the amyloid-driven cascade model fails. This would require a cohort with amyloid clearance from natural resolution or prior treatment, not just donanemab-induced clearance.
**Alternative falsification**: Show that p-tau217 declines in patients with *persistent* amyloid but reduced microglial activation (pharmacologically induced), disconnecting amyloid status from p-tau217 trajectory.
### Revised Confidence: **0.58** (down from 0.78)
The mechanistic pathway is biologically plausible but unproven. The cited evidence supports biomarker correlation, not causal mechanism. Substantial risk of conflating association with causation.
---
## Hypothesis 2: Specific Threshold (<0.15 pg/mL)
### Weak Links
**No Direct Threshold Evidence**: The specific value of <0.15 pg/mL or "80% reduction" appears without citation. This threshold likely derives from:
- Population-level statistical cutoffs for amyloid status classification (not treatment cessation)
- Post-hoc subgroup analyses from TRAILBLAZER-ALZ
- Analogous thresholds from other biomarkers without validation transfer
**"Disease Stasis" Lacks Operational Definition**: The proposed mechanism describes a conceptual window of safety, but:
- No biological threshold defines this "stasis" point
- It assumes p-tau217 level directly reflects pathological activity rather than cumulative burden
- The therapeutic index between "stop here" and "stop too early" remains unquantified
**Cognitive Correlation vs. Causal Prediction**: Even if p-tau217 decline correlates with slowed CDR-SB decline, this does not validate p-tau217 as a *stopping* criterion. The relevant question is: do patients who stop at this threshold maintain their benefit, not just that decline was slower during treatment?
### Counter-Evidence
- TRAILBLAZER-EXT trajectory analysis explicitly states amyloid normalization precedes p-tau217 decline in *some patients*—this asymmetry undermines a single threshold rule
- No prospective trial has tested stopping at specific p-tau217 cutoffs and demonstrated non-inferiority to continued treatment
- The proposed pragmatic trial (different thresholds) has not been conducted
### Falsifying Experiment
**Test**: Identify patients who achieve p-tau217 <0.15 pg/mL but show continued clinical decline (CDR-SB worsening >0.5 points) post-cessation. This would demonstrate the threshold fails to predict durable benefit.
**Alternative**: Show that patients who *never* achieve this threshold but continue treatment for 36+ months have better outcomes than those who stop at threshold—establishing "threshold achievement" as neither necessary nor sufficient for optimal outcomes.
### Revised Confidence: **0.42** (down from 0.65)
The threshold is numerically specific but mechanistically unsupported. The proposed pragmatic trial would directly test this hypothesis and should be prioritized. Current confidence inappropriately exceeds the evidence base.
---
## Hypothesis 3: Temporal Decoupling / Dual Threshold
### Weak Links
**Differential Compartment Kinetics: Unsupported Mechanism**: The stated mechanism assumes vascular vs. neuronal compartment kinetics explain differential clearance rates. However:
- Amyloid PET measures plaque burden; p-tau217 measures neuronal proteins—this conflation of biophysical compartments does not directly map onto the stated mechanism
- The "temporal lag" is asserted but not quantified with inter-patient variance data
- Preclinical models (PMID: 30895603) may not translate to human treatment kinetics
**Clinical Outcome of Premature Stopping: Untested**: The hypothesis assumes stopping at amyloid PET negativity is *harmful* (before tau resolution). However:
- No data demonstrates worse outcomes in patients who stopped at amyloid PET negativity
- The alternative (continuing until p-tau217 normalization) has not been tested
- This is a theoretical risk, not an empirically observed harm
**Biomarker Stability Assumption**: "Stable negative status" implies thresholds exist for both biomarkers, but the variability and measurement error of both PET SUVr and CSF p-tau217 are not accounted for. Centiloid scales have ±5-10 unit variability; CSF p-tau217 assays have significant inter-laboratory variation.
### Counter-Evidence
- The stated evidence (PMID: 37340743, 38042029) describes differential *rates*, not differential *clinical outcomes* of stopping at different times
- No trial has demonstrated harm from stopping at amyloid PET negativity alone
- The "dual threshold" rule increases treatment duration and associated risks (ARIA) without proven benefit
### Falsifying Experiment
**Test**: Randomized controlled trial comparing stopping at amyloid PET negativity vs. stopping at dual-threshold (both PET and p-tau217 negative). If no significant difference in 24-month cognitive outcomes exists, the temporal decoupling hypothesis lacks clinical relevance.
**Alternative**: Demonstrate that patients who stopped at amyloid PET negativity had equivalent p-tau217 normalization at 12 months post-cessation compared to those who continued—showing p-tau217 catches up regardless.
### Revised Confidence: **0.58** (down from 0.72)
While the differential kinetics observation is biologically plausible, the clinical inference (dual threshold required) is not supported by outcome data. The added complexity requires higher evidentiary bar.
---
## Hypothesis 4: Neurodegeneration Floor Effect
### Weak Links
**Floor Effect Mechanism Unverified**: The hypothesis asserts severe neurodegeneration causes "spurious" p-tau217 normalization through neuron loss, but:
- No threshold defines "severe" neurodegeneration for this purpose
- The relationship between structural atrophy and neuronal capacity to produce p-tau217 is not established
- Alternative explanations for p-tau217 normalization in advanced AD exist (see below)
**Alternative Floor Effect Mechanism Not Considered**: The hypothesis focuses on "spurious normalization" via neuron loss, but an equally plausible mechanism is *assay insensitivity*—p-tau217 at very low concentrations may be below quantification limits, producing "normalization" artifacts.
**NfL Evidence Misapplied**: The cited NfL evidence (PMID: 37616244) shows NfL trajectories diverge by baseline atrophy, but this does not validate p-tau217 floor effects. NfL and p-tau217 may have different production and clearance dynamics.
### Counter-Evidence
- No published study explicitly tests p-tau217 validity across neurodegeneration severity strata
- The "floor effect" concept is plausible but lacks direct experimental validation
- Advanced AD patients were largely excluded from TRAILBLAZER-ALZ, limiting empirical data
### Falsifying Experiment
**Test**: In patients with severe hippocampal atrophy (MTA score ≥3), measure p-tau217 alongside synaptic biomarkers (SV2A PET, CSF Ng) and post-cessation clinical trajectories. If preserved synaptic integrity predicts better cognitive outcomes despite advanced atrophy, the floor effect may be neuronal-population-specific, not global.
**Alternative**: Demonstrate that patients with "spurious" p-tau217 normalization (by this hypothesis) have equivalent treatment benefit to those with "true" normalization—challenging the clinical relevance of the floor effect.
### Revised Confidence: **0.48** (down from 0.58)
Mechanistically coherent but empirically unsupported. The hypothesis identifies a potentially important confound but does not establish operational criteria for identifying affected patients.
---
## Hypothesis 5: Combined p-tau217 + p-tau231 Panel
### Weak Links
**"Treatment Response Score" Not Defined**: The proposed composite algorithm lacks operationalization:
- What weighting scheme combines slopes and trajectories?
- What defines "adequate" vs. "inadequate" combined response?
- The score is described but not quantified
**Temporal Window Evidence May Not Transfer**: p-tau231 increases earlier in the *Alzheimer's disease course*—this refers to preclinical/early symptomatic disease natural history. The biomarker kinetics during *treatment-induced* amyloid clearance may differ fundamentally from natural disease progression kinetics.
**Specificity Paradox**: If p-tau231 is sensitive to earlier amyloid changes, it may be *more* confounded by non-AD amyloid pathology (cerebral amyloid angiopathy, Lewy body disease comorbidity) than p-tau217 alone.
### Counter-Evidence
- Combination of p-tau217/p-tau231 ratio improves amyloid *status* discrimination (PMID: 36566449), not treatment *cessation* outcomes
- No study has validated a combined panel for clinical decision-making in donanemab treatment
- Dual biomarkers increase assay complexity, cost, and harmonization challenges
### Falsifying Experiment
**Test**: Develop the composite score and test whether it outperforms p-tau217 alone in predicting post-cessation cognitive stability. If the added information from p-tau231 does not improve prediction (area under curve improvement <0.05), the combination lacks clinical utility.
**Alternative**: Test whether p-tau231 provides *independent* information about treatment response not captured by p-tau217, rather than merely corroborating the same signal.
### Revised Confidence: **0.50** (down from 0.62)
While multi-marker approaches are conceptually sound, this hypothesis lacks operational specificity and empirical validation for the specific clinical application.
---
## Hypothesis 6: Microglial Phenotype Modulation
### Weak Links
**TSPO-PET M1/M2 Specificity in Humans is Questionable**: TSPO binds translocator protein expressed in microglia, but:
- TSPO-PET signal is non-specific for activation state (M1 vs. M2)
- Human microglia do not conform cleanly to the M1/M2 dichotomy established in rodent models
- The cited TREM2 evidence shows microglial involvement in clearance, not specific phenotype determination
**Acute vs. Chronic Activation: Unresolved**: Donanemab triggers acute microgl