Re‐calibration and performance of a proteomic prognostic test for 5‐year dementia risk
Background Proteomics‐driven health and wellness tests can provide personalized insight into individual risk for disease that is sensitive to change across time. The SomaSignalTM Mid‐Life Dementia Risk Test is a blood‐based, protein‐only prognostic model developed using SomaScan® Assay v4.0 (measuri...
Gespeichert in:
Veröffentlicht in: | Alzheimer's & dementia 2023-12, Vol.19 (S14), p.n/a |
---|---|
Hauptverfasser: | , , , , , , , |
Format: | Artikel |
Sprache: | eng |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | Background
Proteomics‐driven health and wellness tests can provide personalized insight into individual risk for disease that is sensitive to change across time. The SomaSignalTM Mid‐Life Dementia Risk Test is a blood‐based, protein‐only prognostic model developed using SomaScan® Assay v4.0 (measuring ∼5,000 proteins) combined with machine learning methods to predict 20‐year risk for dementia in individuals aged 49 years and older. However, shorter term risk stratification for dementia may better enable early detection of emerging changes in cognition, enrich clinical trial enrollment for novel interventions, and eventually improve precision medicine prescribing of therapeutics. Thus, the 20‐year dementia risk test was re‐calibrated to maximize performance of 5‐year prediction in an older population.
Method
The 25‐protein Mid‐Life Dementia Risk model was re‐calibrated to predict 5‐year risk for a dementia diagnosis in adults aged 65 and older. We utilized plasma samples from ARIC visit 5 (n = 4985; median age 75 years) for the re‐calibration, which included a total of 391 total incident dementia diagnoses occurring within 5 years of blood draw. Re‐calibrated models were assessed in a training dataset consisting of 70% of the ARIC visit 5 data to identify the optimal 5‐year risk model. The final model was tested for robustness in a 15% holdout sample and validated independently in the remaining 15% samples.
Result
The AUC of the re‐calibrated 5‐Year Dementia Risk model was equal to 0.778, 0.735, and 0.776, in training, verification, and validation datasets, respectively. Individuals were stratified by relative risk into Low‐, Medium‐Low‐, Medium‐High‐, and High‐risk bins, with a 5‐year dementia diagnosis rate of 1.510% in the Low‐risk bin vs 20.337% in the High‐risk bin. The performance of the proteomic model was superior to that of APOE genotype or age in predicting 5‐year dementia risk as evaluated by AUC and dynamic range.
Conclusion
The 20‐Year Dementia Risk Test was successfully re‐calibrated to predict 5‐year dementia risk in individuals aged 65 years and older with more than a 13‐fold dynamic range for risk stratification. The proteomics‐based 5‐Year Dementia Risk Test outperforms leading risk factors for dementia, including APOE genotype and age. |
---|---|
ISSN: | 1552-5260 1552-5279 |
DOI: | 10.1002/alz.074901 |