# Feasibility Assessment: CSF p-tau217 Surrogate Endpoint for Donanemab Cessation
## Preliminary Note on Confidence Recalibration
The skeptic's downward revisions reflect a fundamental problem: several hypotheses conflate biomarker *association* with mechanistic *causation*, and some propose thresholds without empirical derivation. I will integrate these critiques and assess each surviving hypothesis across the five dimensions.
---
## Portfolio-Level Strategic Assessment
**Core feasibility question:** Can p-tau217 normalization serve as a reliable surrogate for durable clinical benefit sufficient to define a treatment cessation threshold?
**Threshold to pass:** For regulatory acceptance as a stopping rule surrogate, p-tau217 must demonstrate: (1) treatment effect on biomarker reliably predicts treatment effect on clinical outcome (modlational surrogate), and (2) the threshold has prospective validation showing non-inferiority of stopping at threshold versus continuing treatment.
**Current position:** No single hypothesis meets this bar. The portfolio is best understood as a **sequential validation pipeline** rather than a set of independent investigations.
---
## Hypothesis 1: Amyloid → Tau Cascade Mechanism
### Druggability
**Not directly applicable** — this hypothesis is about biomarker validation, not therapeutic target identification. However, the mechanistic pathway implies that *treatment duration* is the modifiable variable. If validated, the druggable question becomes: what biomarker determines optimal treatment duration?
### Biomarkers/Model Systems
| Component | Current Status | Gap |
|-----------|---------------|-----|
| CSF p-tau217 | ALZPATH Roche assay available; mass spec reference method in development | Inter-lab reproducibility remains problematic; AAAQC program in early stages |
| Amyloid PET | Centiloid scale established; FDA-qualified for enrollment | Longitudinal PET cost and patient burden |
| Mechanistic validation models | iPSC neurons from AD patients; acute slice preparations | No human model recapitulates trans-synaptic tau propagation in vivo |
**Critical issue:** The mechanism requires proving that amyloid clearance *causally reduces* p-tau217 production, not merely correlates with it. This demands either:
- An experimental medicine study (e.g., a BACE inhibitor that clears amyloid without directly affecting tau) to test causality, or
- Mendelian randomization using genetic variants affecting amyloid vs. tau pathways
**Model system limitation:** Mouse models of amyloid-tau interaction have limited translatability. The 3xTg and App/NL-G-F knock-in models show amyloid-tau spatial correlation but do not faithfully model human trans-synaptic propagation kinetics.
### Clinical Development Constraints
**Design:** Longitudinal biomarker trajectories in ongoing TRAILBLAZER-EXT/2 patients (estimated n≈400 with CSF at multiple timepoints) could address the correlation question relatively quickly. However, causation requires a prospective interventional design.
**Primary endpoint structure:**
- Surrogate validation requires demonstrating that p-tau217 treatment effect predicts CDR-SB treatment effect
- This demands a minimum 18-month post-cessation observation period
- Composite endpoint of amyloid rebound + cognitive trajectory is the realistic regulatory target
**Regulatory pathway:** FDA Biomarker Qualification Program (BQP) — the Level 2 validation study (quantitative >0.7) is feasible within 3 years if multi-site harmonization is prioritized from the outset.
### Safety
**Low inherent risk** — CSF sampling and PET imaging are standard-of-care in trial populations. The safety concern is indirect: if the causal mechanism is wrong, patients stopping at p-tau217 thresholds may experience amyloid rebound without tau pathology benefit. ARIA risk is already addressed in the donanemab label; biomarker-driven stopping does not introduce novel safety signals beyond the stopping decision itself.
### Timeline/Cost
| Phase | Duration | Estimated Cost |
|-------|----------|----------------|
| Assay harmonization (AAAQC program) | 12-18 months | $2-3M |
| Retrospective biomarker analysis (existing TRAILBLAZER samples) | 6-12 months | $1.5-2M |
| Prospective validation cohort | 30-36 months | $10-15M |
| **Total to BQP Level 2** | **4-5 years** | **$13-20M** |
**Risk:** The mechanistic causation question may not be resolvable without a second mechanism-disrupting agent (e.g., a selective amyloid clearer without tau effects). Without this, the pathway remains associative.
**Post-skeptic revised confidence: 0.58** — The mechanistic pathway is biologically plausible but unproven. This hypothesis is best positioned as **foundational/background** rather than a standalone stopping rule.
---
## Hypothesis 2: <0.15 pg/mL Threshold Validation
### Druggability
**Indirect relevance** — if validated, this threshold becomes the basis for a **companion diagnostic (CDx)** indication. This is a regulatory and commercial opportunity, not a druggability question per se.
**CDx development pathway:**
- The threshold value requires independent validation as a medical device (in vitro diagnostic)
- Requires FDA Center for Devices and Radiological Health (CDRH) engagement
- Would follow the CDx pathway similar to HER2/neu for trastuzumab
### Biomarkers/Model Systems
**Major obstacle: the threshold is operationally undefined.**
The stated <0.15 pg/mL value has three possible origins, none validated for cessation:
1. Cross-sectional amyloid status cutoffs from population studies (not treatment cessation)
2. Post-hoc subgroup data from TRAILBLAZER-ALZ that was never prospectively validated
3. Analogous to p-tau181/p-tau231 thresholds from different assays and populations
**Assay-specific problem:** p-tau217 values are highly assay-dependent. The Roche Lumipulse values, Quanterix Simoa values, and mass spec values do not have established equivalence at the <0.15 pg/mL range. This threshold cannot be treated as universal without a reference standard.
**Validation requirements:**
- Lock the assay platform and reference standard first
- Establish the threshold in a training cohort (n≥150)
- Validate in an independent test cohort
- Confirm against post-cessation clinical outcomes
### Clinical Development Constraints
**Proposed design:** Pragmatic RCT randomizing to different thresholds (0.10, 0.15, 0.20 pg/mL) with 18-month post-cessation CDR-SB primary endpoint.
**Critical flaws:**
- The 18-month timeframe is likely insufficient to detect harm from premature stopping — amyloid rebound and tau pathology resurgence may manifest at 24-36 months
- A non-inferiority design requires pre-specified non-inferiority margins that have not been established for CDR-SB in this context
- Sample size to detect clinically meaningful differences across threshold arms: n≈100/arm minimum, total n≥300
**Regulatory challenge:** A pragmatic trial comparing stopping at different thresholds faces an ethical problem — continuing treatment beyond amyloid PET negativity has been considered acceptable practice. Randomizing to "stop too early" requires an IRBs and may face FDA concerns about trial ethics.
### Safety
**This hypothesis carries the highest safety risk of the portfolio.** If the threshold is set too high (premature cessation), patients may experience:
- Amyloid rebound with renewed inflammatory activity
- Re-acceleration of tau pathology
- Clinical decline that was previously slowed
Unlike biomarker-based enrollment criteria (which expose participants to a treatment), this hypothesis exposes participants to a *treatment cessation decision* that may be harmful. The risk-benefit calculation requires careful consideration.
**Mitigation:** A Data Safety Monitoring Board (DSMB) with pre-specified stopping rules for the early cessation arms if amyloid rebound exceeds a safety threshold (e.g., >10 Centiloid units above baseline at 12 months).
### Timeline/Cost
**The proposed trial has never been conducted, and the investment risk is high given low initial confidence.**
| Activity | Duration | Cost |
|----------|----------|------|
| Assay lock and reference standardization | 18-24 months | $3-5M |
| Training set threshold derivation | 12 months | $2-3M |
| Pivotal non-inferiority RCT (n=300, 24-month f/u) | 36-48 months | $35-45M |
| CDx regulatory submission | 12 months | $2-3M |
| **Total to potential CDx approval** | **6-7 years** | **$42-56M** |
**Cost-risk ratio assessment:** Given the revised confidence of 0.42, this represents high investment risk. The threshold should be de-risked through retrospective analysis of TRAILBLAZER-EXT data *before* committing to the pivotal RCT. A negative result at the retrospective stage would save $35-45M.
**Post-skeptic revised confidence: 0.42** — The numerical specificity is not matched by mechanistic support. This is the highest-priority hypothesis for falsification, not the highest-priority for confirmation.
---
## Hypothesis 3: Temporal Decoupling / Dual Threshold
### Druggability
**Low direct druggability relevance.** The therapeutic question is about *when* to stop, not *what* to drug.
### Biomarkers/Model Systems
**Strength:** Both amyloid PET and CSF p-tau217 are established, well-characterized biomarkers with published longitudinal trajectories.
**Key limitation:** The "temporal lag" is asserted without quantification of inter-patient variance. The stated evidence (TRAILBLAZER-ALZ/EXT) shows:
- Median time to amyloid PET negativity: 12-18 months
- Median time to p-tau217 normalization: ongoing decline beyond 24 months
But the **range** of these values is critical. If some patients achieve p-tau217 normalization at 12 months while amyloid PET normalizes at 24 months, a dual threshold may be appropriate. If the vast majority of patients show p-tau217 normalization within 6 months of amyloid PET negativity, the dual threshold adds complexity without clinical benefit.
**Analytical gap:** The published data does not report the *distribution* of lag times, only medians. A post-hoc analysis of existing TRAILBLAZER data could address this within 6 months at low cost.
**Model systems:** Not particularly relevant — this is a pharmacokinetic/pharmacodynamic (PK/PD) question about biomarker compartment kinetics. Mixed-effects modeling of existing longitudinal data is the appropriate method.
### Clinical Development Constraints
**Key clinical question:** Does stopping at amyloid PET negativity lead to worse outcomes than stopping at dual negativity?
**Critical evidence gap:** There is no prospective data showing that stopping at amyloid PET negativity is harmful. The dual threshold is justified by theoretical risk, not observed harm.
**Trial design options:**
1. **Superiority design (preferred):** Demonstrate that dual-threshold stopping produces superior 24-month cognitive outcomes versus amyloid-PET-only stopping. Requires n≈400 (200/arm) for 80% power at Δ=0.5 CDR-SB points. Cost: $40-55M.
2. **Non-inferiority design:** Demonstrate that dual-threshold stopping is not worse than amyloid-PET-only stopping. Smaller sample (n≈250), but requires pre-specified NI margin.
3. **Observational cohort:**