Experiment Proposal (crux): Multi-Biomarker Composite Index Surpassing Amyloid PET for Treatment Response Prediction — Composite index performance superiority over individual biomarkers not demonstra

AIMS

HYPOTHESES

PROTOCOL SUMMARY

Phase 1 - Retrospective Validation: Retrieve archived plasma samples from ADNI, TRIAD, and EMIF-AD cohorts (n=800 total; 400 treatment-responders defined by 12-month cognitive stabilization or improvement on CDRsb/ADAS-Cog13, 400 non-responders matched by age/sex/baseline severity). Measure biomarkers in duplicate using Lumipulse (Fujirebio) for Aβ42/40, p-tau181, p-tau217, and Simoa for NfL/GFAP. Calculate MBI using pre-specified algorithm: MBI = 0.35×(Aβ42/40 z-score) + 0.25×(p-tau217 z-score) + 0.20×(p-tau181 z-score) + 0.10×(NfL z-score) + 0.10×(GFAP z-score). Acquire amyloid PET SUVR values (Centiloid conversion) from existing datasets. Phase 2 - Head-to-Head Classification: Conduct matched cross-validation (5-fold, 10-repeat) comparing AUC for response prediction: MBI vs each individual biomarker vs amyloid PET SUVR alone. Calculate Delong test for AUC comparison with Bonferroni correction. Phase 3 - Non-Inferiority Test: Apply pre-specified non-inferiority margin (δ = 0.10). If lower bound of 95% CI for AUC difference (MBI - PET) exceeds -δ, non-inferiority established. Phase 4 - Minimal Panel Analysis: Recursively eliminate markers to identify reduced panel maintaining ≥90% full MBI performance. All biomarker assays performed in ISO 17025 accredited facility with blinded clinical outcome assessment.

PREDICTED OBSERVATIONS

If the hypothesis is true: MBI will show statistically significant AUC improvement over individual biomarkers (estimated ΔAUC 0.08-0.12) and meet non-inferiority threshold against amyloid PET (potentially exceeding PET AUC by 0.05-0.08 given mechanistically broader pathology capture). Reduced three-marker panel will retain ≥90% predictive power, indicating robust signal across mechanistically distinct pathways.

FALSIFICATION CRITERIA

H1 falsified: No significant AUC difference between best single biomarker (predicted: p-tau217) and MBI (DeLong test p > 0.05 after correction), indicating composite provides no additional predictive value beyond validated single markers. H2 falsified: Lower 95% CI bound for AUC(MBI) - AUC(PET) falls below -0.10 (non-inferiority margin), establishing MBI is inferior to amyloid PET for treatment response. H3 falsified: Reduced panel retains <90% of full MBI performance, indicating all five markers are necessary and redundancy does not exist. Grand falsification: Even the full MBI fails to exceed chance-level prediction (AUC CI < 0.55), indicating biomarkers have no predictive utility for treatment response in this cohort.

DATASET DEPENDENCIES