Listen

Description

Send us Fan Mail

Machine learning can forecast how fast thinking skills will slip — but how well do those models work on a new group of patients? Researchers in Zurich trained models on ADNI (1,237 people) and OASIS-3 (662 people) to predict yearly change in CDR-SOB and MMSE scores. Adding structural MRI to clinical and demographic data improved predictions inside each dataset (R² up to about 0.41–0.42 for CDR-SOB). When the same models were tested on the other cohort, performance dropped (R² roughly 0.18–0.35). Slimming the models to only the top predictors did about as well externally, which points to redundant features rather than a free lunch. The paper argues that measuring this generalizability gap is essential before trusting these tools for individual planning.

Source: Hüppi RM, Langer N, Vieira BH, Quantifying generalization error in machine learning prediction of cognitive decline, J Prev Alzheimers Dis 2026. https://doi.org/10.1016/j.tjpad.2026.100646 (CC BY 4.0)

This podcast is created by Ai for educational and entertainment purposes only and does not constitute professional medical or health advice. Please talk to your healthcare team for medical advice. 

Never miss an episode—subscribe on your favorite podcast app!