Consensus Risk Modeling and Uncertainty Quantification of Alzheimers Disease Using 5ADCSI Plasma Biomarkers and Multiple External Machine-Learning Frameworks

The study examined whether plasma biomarkers measured using the 5ADCSI platform preserve biologically meaningful information about Alzheimer's disease when using three independently developed machine-learning models. The research involved 472 participants from the Louisville Twins Study and compared three different models: one from the A4 study with the Lilly p217tau assay and two from the ADNI study with Quanterix Simoa p217tau assays using different antibodies. All three models recognized biologically coherent signals related to Alzheimer's disease despite differences in training cohorts and assay methodology. The strongest agreement was between the A4-MSD and ADNI-AlzPath models, while agreement with the ADNI-Janssen model was weaker. Consensus risk modeling identified a high-risk subgroup with low prediction uncertainty and low rank instability. Participants with high risk according to the consensus framework had elevated levels of APOE ε4, p-tau217, and GFAP.