A machine learning enrichment strategy for presymptomatic cohorts in Alzheimer's disease clinical trials. (31st December 2021)

• Record Type: Journal Article
• Authors: Tam, Angela; Laurent, César; Noriega de la Colina, Adrián; Gauthier, Serge; Dansereau, Christian
• Is Part Of: Alzheimer's & dementia. Volume 17:(2021)Supplement 1
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-12-31
• Subjects: Alzheimer's disease -- Periodicals; Alzheimer Disease -- Periodicals; Dementia -- Periodicals; Démence; Maladie d'Alzheimer; Périodique électronique (Descripteur de forme); Ressource Internet (Descripteur de forme); 616.83
• DOI: 10.1002/alz.054702
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: It is challenging for Alzheimer's disease trials to enroll presymptomatic individuals who are likely to decline cognitively. We propose using a multimodal machine learning approach to discriminate between cognitively normal individuals who progressed to mild cognitive impairment (MCI) and those who remained stable over 48 months. Method: Patterns of gray matter distribution, derived from structural MRI, associated with either healthy aging or Alzheimer's disease were extracted from AIBL [1] (n=401). We processed images from 823 individuals deemed cognitively normal at baseline from NACC [2] (n=433), ADNI [3] (n=260), and OASIS‐3 [4] (n=130) and generated scores representing the spatial similarities between the gray matter distribution of each individual to the patterns produced from AIBL. Gray matter scores, MMSE, CDR‐SB, FAQ, APOE4 status, education, age, and sex were used to train a machine learning prognostic pipeline using the Foresight platform (Perceiv Research Inc.) to distinguish individuals who received a MCI diagnosis within 48 months from baseline (progressors) from those who remained stable. The model was trained and tested with 5‐fold inner‐loop cross‐validation. Result: The model achieved a mean (± std) AUC of 0.752±0.045, accuracy of 70.72±2.44, sensitivity of 62.51±12.12, and specificity of 71.82±3.87. The predicted progressors experienced a steeper cognitive decline, tended to be older, and contained a higher proportion APOE4 carriers and amyloid positive cases, compared to stable individuals. Of the predicted progressors, 33.3% were true progressors. This represents a nearly two‐fold enrichment of progressors over the prevalence of the entire sample where only 17.7% were true progressors. Conclusion: An automated algorithm can successfully identify presymptomatic individuals with impending cognitive impairment. This tool can enhance trial enrollment by targeting individuals who are at the highest risk of cognitive decline and potentially reduce trial costs by minimizing the need for prohibitively large sample sizes. [1] Australian Imaging Biomarkers and Lifestyle Study of Ageing (aibl.csiro.au). [2] National Alzheimer's Coordinating Center (naccdata.org). [3] Alzheimer's Disease Neuroimaging Initiative (adni.loni.usc.edu). [4] Open Access Series of Imaging Studies (oasis‐brains.org).