Alzheimer's disease is typically confirmed through expensive PET imaging or invasive cerebrospinal fluid sampling — barriers that delay diagnosis for millions. A framework that achieves meaningful risk stratification from a standard blood draw could fundamentally shift how clinicians identify high-risk individuals before symptoms consolidate. That possibility gains traction from new evidence pairing two genomic layers: fixed inherited variants and the more fluid landscape of gene expression.
The analysis drew on two distinct cohorts — a North American sample of 313 participants from the Alzheimer's Disease Neuroimaging Initiative and a Korean sample of 173 from Seoul National University Bundang Hospital — making cross-ancestry replication possible within a single study. Researchers derived polygenic risk scores (PRS) from genome-wide genotyping data and transcriptional risk scores (TRS) from blood-based RNA expression profiles, then combined them using logistic regression and machine learning classifiers. Compared with individuals carrying low values on both measures, those in the high-PRS / high-TRS stratum faced odds of Alzheimer's diagnosis that were 2.5 times greater in the American cohort and 3.4 times greater in the Korean cohort. Critically, the integrated model reached an AUC of 0.705, outperforming the PRS-alone model's AUC of 0.635 — an 11-point gain that reflects a meaningful improvement in discriminative power.
The result is conceptually important because PRS captures static inherited susceptibility while TRS reflects dynamic, context-sensitive gene regulation — the two signals are at least partially orthogonal, which explains why their combination adds predictive value. This mirrors a growing trend in precision medicine toward multi-omic integration. Nevertheless, an AUC of 0.705, while encouraging, remains well below clinical deployment thresholds typically set around 0.85–0.90, and the relatively small cohort sizes limit confidence in the machine learning estimates. The study is also cross-sectional, meaning it cannot confirm whether baseline scores predict conversion over time — arguably the more pressing clinical question. Still, the successful replication across ancestrally distinct populations is a genuine strength that many single-cohort genomic studies lack, and the blood-only input requirement positions this framework as a realistic screening tool deserving of prospective longitudinal validation.