AI Science Tracker
EN

External dementia-risk evaluation shows cohort matching changes apparent performance

Fraunhofer Institute for Algorithms and Scientific Computing; University of Bonn; Erasmus MC; University of Oxford; UCB Biosciences · 22 Jun 2020 · biology

Discovery

External dementia-risk evaluation shows cohort matching changes apparent performance

Source check: Source checked. Not independent scientific confirmation.

Finding: A revised gradient-boosted dementia-risk model trained on ADNI was evaluated in 244 AddNeuroMed participants, including thirty later dementia diagnoses. Performance differed when evaluation was restricted to demographically matched participants, demonstrating how cohort selection changes apparent validation results. This is retrospective clinical model validation, without proven preventive benefit or prospective clinical deployment.

Technology: Stochastic gradient-boosted decision trees for time-dependent dementia-risk prediction

Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.

AI contribution
The trained GBM combined multimodal baseline features into time-dependent risk estimates; its concrete external clinical evaluation exposed the effect of cohort composition on reported performance. Propensity-score matching alone is not treated as the enabling AI contribution.
Test and result
A feature-restricted GBM was retrained on ADNI to match available AddNeuroMed measurements, retaining thirty-two clinical, imaging and genetic features. It was then evaluated in 244 initially normal or mildly impaired AddNeuroMed participants, thirty of whom later received a dementia diagnosis. The reported external concordance index was approximately 0.81. Across one hundred propensity-score matchings, the average matched set contained 164 participants and twenty conversions, with concordance approximately 0.88.
Limitation / still unknown
External evaluation was retrospective; prospective cohort follow-up does not establish predictor locking before clinical outcomes were known. Only thirty external events and fewer at long horizons limit precision. Matching changes the selected population and can improve apparent performance without improving the model or eliminating all bias. The revised model differs from the earlier full-feature model. No clinical prevention, treatment selection, economic benefit or general-population diagnostic utility was demonstrated.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://link.springer.com/article/10.1007/s13167-020-00216-z

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.

Review date: 21 Nov 2026. Review scope: Read actual primary cohort comparison/matching, revised GBM feature/training and external evaluation Methods, unmatched/matched dementia Results and limitations. Supplement implementation was not independently read. Specific clinical external evaluation is distinguished from the earlier model publication and generic six-year-prediction headlines. Exact original article front DOI/title and fresh PubMed primary relationship metadata checked..

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open in the full catalog for filters, charts and related discoveries. Browse biology discoveries.