A leakage-controlled multi-compartment benchmark of fluid biomarkers for predicting mild cognitive impairment to Alzheimer's disease conversion: plasma, CSF, and high-plex omics in a single cohort.

Approved disease-modifying therapies have made it a practical priority to predict, early and at low cost, who among people with mild cognitive impairment (MCI) will progress to Alzheimer's disease (AD) dementia. The candidate biomarkers sit across biological compartments that differ sharply in invasiveness and cost: blood plasma, cerebrospinal fluid (CSF, via lumbar puncture), and high-plex "omics" assays. Once data leakage in feature selection is controlled, how much each compartment truly contributes to predicting MCI-to-AD conversion is an open question, and answering it is the point of this benchmark. Using the Alzheimer's Disease Neuroimaging Initiative (ADNI), we benchmarked prediction of MCI to AD conversion across four clinical compartments (demographics, classic plasma, plasma mass-spectrometry, classic CSF) and four omics compartments (plasma metabolomics, plasma lipidomics, CSF metabolomics, CSF SOMAscan proteomics). Validation used 5-fold cross-validation with bootstrap 95% confidence intervals and nested, in-fold feature selection to keep leakage out. The best single compartment was classic plasma (p-tau217/NfL/GFAP), which reached an AUC of 0.785 (0.71-0.85); CSF proteomics (SOMAscan) came close behind at 0.758 (0.71-0.80), and classic CSF followed at 0.747 (0.71-0.79), whereas plasma metabolomics and lipidomics sat near chance at 0.51-0.60. When the sample is narrowed to the 159 subjects who had complete data on all three, the comparison gets sharper still. Plasma, in other words, is where most of the usable information sits: put it on top of demographics and the AUC climbs from 0.578 to 0.769, an NRI of +0.79. CSF, on top of all that, is worth only another +0.03 of AUC, a gain too small to lean on. APOE4 keeps predicting conversion on its own, at an HR of 1.83 per allele (p=0.008). So, to say it plainly: a blood-based panel already carries most of what is needed to predict MCI to AD conversion, and CSF adds only a little on top of it. Read for practice, the results support a staged, blood-first screening strategy, and they put a concrete number on what each additional compartment contributes once overoptimism is accounted for.