A Norwegian memory clinic cohort study combining MRI-derived imaging hazard scores, polygenic risk scores, cognitive assessments (MMSE-NR3 or CERAD 10-word recall), and demographics into a single Multimodal Hazard Score for Real-World Data (MHS-RWD) achieved AUCs of 0.89 (female) and 0.84 (male) for early dementia detection across 1,100 patients, and 0.91 (female) and 0.83 (male) for distinguishing Alzheimer's from non-AD dementias in 788 patients — consistently surpassing any individual predictor alone.
The sex-stratified performance gap is immediately notable: female patients show roughly 6–8 percentage points higher discrimination across both tasks, a pattern consistent with growing evidence that AD biomarker profiles differ meaningfully by sex, possibly tied to hormonal influences on tau pathology and APOE-ε4 penetrance. Clinically, an AUC above 0.90 for differential diagnosis approaches the threshold where tools can meaningfully redirect treatment pathways — particularly relevant as anti-amyloid therapies like lecanemab demand accurate AD confirmation before prescription.
Key caveats temper enthusiasm. The cohort is drawn exclusively from Norwegian memory clinics, limiting generalizability across ethnic and socioeconomic populations where polygenic risk scores built on European GWAS data perform less reliably. The three-year diagnostic follow-up window introduces label uncertainty. Importantly, this is a preprint posted on medRxiv and has not yet undergone peer review — methodology, effect sizes, and conclusions may change substantially. Still, the real-world clinical data design and individualized risk report output mark this as a practically oriented, potentially impactful contribution to precision dementia care.
