Performance boundaries of plasma-centred prediction across PET- and MRI-defined AT(N): QC-first internal validation in ADNI with separate axis-specific assessments
- 1The University of Manchester
Abstract
BackgroundPlasma p-tau217 identifies Alzheimers disease pathology, but performance in mixed cohorts may not extend to conditional tau-positivity dis-crimination within amyloid-positive individuals or to MRI-defined hippocampal neurodegeneration. We compared plasma prediction across PET- and MRI-defined AT(N) outcomes and conducted separate axis-specific assessments.
MethodsThis retrospective secondary study used Alzheimers Disease Neuroimaging Initiative (ADNI) data. Tasks were amyloid positivity (A; 1,097 participants, 496 positive), tau positivity (T; 929, 195), T within A+ (361, 161), study-defined hippocampal neurodegeneration (N; 1,303, 248), and joint A+T+ (852, 161). Repeated nested cross-validation used fold-internal preprocessing and calibration. Post-hoc assessments comprised a Bio-Hermes A-axis staged random-forest analysis using Lilly p-tau217 and C2N A{beta}40/A{beta}42 (908 participants; 328 and 324 positive under the primary and alternative PET classifications), and an A4 ordinal p-tau217 analysis against an A4-specific medial-temporal tau-PET endpoint in A+ participants.
ResultsADNI random-forest AUROCs were 0.9425 (95% CI 0.9271-0.9562) for A, 0.9495 (0.9325-0.9645) for T, 0.8852 (0.8485-0.9163) for T within A+, 0.7865 (0.7553-0.8151) for N, and 0.9614 (0.9484-0.9726) for joint A+T+. At thresholds targeting 90% sensitivity, specificity was 87.5% for T overall and 60.7% within A+. In Bio-Hermes, adding A{beta}40 and A{beta}42 to p-tau217 increased AUROC from 0.894 to 0.916 (paired difference 0.022, 95% conditional interval 0.010- 0.035) and changed Brier score by -0.011 (-0.017 to -0.005); the alternative PET classification was concordant. In A4, ordinal p-tau217 had AUROC 0.7356 (95% CI 0.6775-0.7907), average precision 0.5657 (0.4732-0.6677), and Spearman{rho} = 0.4718 (0.3824-0.5528).
ConclusionsADNI identified task-dependent performance boundaries, particularly attenuated conditional T discrimination within A+. Separate-cohort evidence supported moderate single-marker conditional-T discrimination in A4 and modest complementary A-axis information from plasma A{beta} in Bio-Hermes, but only within those respective axes and assay implementations.
Related
- Comparison of Internal Validation Strategies for Estimating Machine Learning Model Performance in One-Year Mortality Prediction in Hepatocellular Carcinoma
- Development and internal validation of an exploratory clinical prediction model for MRI-defined white matter injury in moderate-to-late preterm infants: a single-center retrospective study
- Development and internal validation of a machine learning prediction model for postoperative hemorrhage following modified radical mastectomy for breast cancer
- Development and Internal Validation of a Prediction Model Incorporating Respiratory Muscle Function for CAT Score Deterioration in Stable COPD
- Development and internal validation of a prediction model for postpartum haemorrhage following caesarean delivery in a resource-limited setting in Ethiopia
BibTeX
@misc{xu2026performance,
title = {{Performance boundaries of plasma-centred prediction across PET- and MRI-defined AT(N): QC-first internal validation in ADNI with separate axis-specific assessments}},
author = {J. Xu and S. Michopoulou and F. Costen},
howpublished = {medRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.15.26365631},
url = {https://www.medrxiv.org/content/10.64898/2026.10.15.26365631v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Performance boundaries of plasma-centred prediction across PET- and MRI-defined AT(N): QC-first internal validation in ADNI with separate axis-specific assessments
AU - J. Xu
AU - S. Michopoulou
AU - F. Costen
PY - 2026
DA - 2026/10/15
PB - medRxiv
DO - 10.64898/2026.10.15.26365631
UR - https://www.medrxiv.org/content/10.64898/2026.10.15.26365631v1
AB - BackgroundPlasma p-tau217 identifies Alzheimers disease pathology, but performance in mixed cohorts may not extend to conditional tau-positivity dis-crimination within amyloid-positive individuals or to MRI-defined hippocampal neurodegeneration. We compared plasma prediction across PET- and MRI-defined AT(N) outcomes and conducted separate axis-specific assessments. MethodsThis retrospective secondary study used Alzheimers Disease Neuroimaging Initiative (ADNI) data. Tasks were amyloid positivity (A; 1,097 participants, 496 positive), tau positivity (T; 929, 195), T within A+ (361, 161), study-defined hippocampal neurodegeneration (N; 1,303, 248), and joint A+T+ (852, 161). Repeated nested cross-validation used fold-internal preprocessing and calibration. Post-hoc assessments comprised a Bio-Hermes A-axis staged random-forest analysis using Lilly p-tau217 and C2N A{beta}40/A{beta}42 (908 participants; 328 and 324 positive under the primary and alternative PET classifications), and an A4 ordinal p-tau217 analysis against an A4-specific medial-temporal tau-PET endpoint in A+ participants. ResultsADNI random-forest AUROCs were 0.9425 (95% CI 0.9271-0.9562) for A, 0.9495 (0.9325-0.9645) for T, 0.8852 (0.8485-0.9163) for T within A+, 0.7865 (0.7553-0.8151) for N, and 0.9614 (0.9484-0.9726) for joint A+T+. At thresholds targeting 90% sensitivity, specificity was 87.5% for T overall and 60.7% within A+. In Bio-Hermes, adding A{beta}40 and A{beta}42 to p-tau217 increased AUROC from 0.894 to 0.916 (paired difference 0.022, 95% conditional interval 0.010- 0.035) and changed Brier score by -0.011 (-0.017 to -0.005); the alternative PET classification was concordant. In A4, ordinal p-tau217 had AUROC 0.7356 (95% CI 0.6775-0.7907), average precision 0.5657 (0.4732-0.6677), and Spearman{rho} = 0.4718 (0.3824-0.5528). ConclusionsADNI identified task-dependent performance boundaries, particularly attenuated conditional T discrimination within A+. Separate-cohort evidence supported moderate single-marker conditional-T discrimination in A4 and modest complementary A-axis information from plasma A{beta} in Bio-Hermes, but only within those respective axes and assay implementations.
ER -