Dementia prediction models need broader testing beyond their development data
Across studies involving more than 1 million participants, artificial intelligence models predicted dementia onset with generally good accuracy. Testing in separate populations was limited, leaving clinical usefulness uncertain.
Based on the published abstract. The full paper may contain additional methods, results and limitations.
The 30-second takeaway
The reviewed models generally performed well at predicting who would develop dementia, and some machine-learning approaches outperformed traditional methods. However, success within the data used to develop a model does not establish that it will work reliably elsewhere. Limited testing in separate populations leaves uncertainty about using these models in routine clinical care.
The models differed substantially, external testing was limited, and accurate risk estimates remained a challenge. The abstract does not establish whether using these predictions improves diagnosis, care decisions, or patient outcomes.
What the models used
The review examined computer-based models designed to predict dementia onset. Clinical information and cognitive assessments were their most common inputs, while biological measurements were used less often. The models varied considerably in their methods.
Why testing elsewhere matters
Every study tested performance internally, within its own study data. Testing on separate populations was less common. The authors also identified calibration—whether predicted risks match observed outcomes—and applicability to other populations as unresolved challenges.
The original publication
Clinical prediction models using artificial intelligence approaches in dementia.
Veronese N, Bolzetta F, Gallo L et al.
Aging Clin Exp Res · 2025
- PubMed ID
- 40715606
- Record checked
AI-assisted research and writing. This explains one selected publication; it is not a complete review of everything known. Our approach.
One more question, understood.
Keep track of the research you’ve explored.