Robust prediction of memory and neuroticism in men and women using connectome‐based predictive modeling. (20th December 2022)

• Record Type: Journal Article
• Authors: Ju, Suyeon; Horien, Corey; Constable, Todd; Fredericks, Carolyn A
• Is Part Of: Alzheimer's & dementia. Volume 18(2022)Supplement 5
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-12-20
• 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.063015
• Languages: English
• ISSNs: 1552-5260

Abstract

Abstract: Background: Alzheimer's disease (AD) disproportionately impacts women, yet we have little understanding of the differences in brain circuitry that underlie this vulnerability. We explored aspects of brain connectivity that robustly predict memory performance (measured by RAVLT performance) and neuroticism (measured by NEO‐N) in healthy women vs men by leveraging data‐driven machine learning to model the brain connectome in the Lifespan Human Connectome Project Aging (HCP‐A) dataset. Method: We used functional MRI scans to create whole‐brain connectivity matrices based on 268 regions‐of‐interest for 725 healthy subjects (319 men, 406 women) aged 36 to 100 enrolled in HCP‐A. Before and after separating the subjects by sex, connectome‐based predictive modeling (CPM) was used with a p‐value threshold of 0.01 and split‐half cross‐validation to identify and implement edges that significantly predicted the neurobehavioral scores to train a predictive model for each group. The models were applied to test sets within each group, outputting the predicted behavioral measures and Pearson correlations between predicted and observed scores. Result: As anticipated, overall performance in HCP‐A was higher for women than men in RAVLT (p < 1x10‐7) and NEO‐N (p = 0.12). Our model predicted the RAVLT measure (Fig. 1) with accuracies ranging from R = 0.2154 (p < 1x10‐4) to R = 0.4293 (p < 1x10‐4). Our sex‐based models also predicted a high variance in the RAVLT measure between sexes (σ 2 = 0.015, Fig. 2). While neuroticism predictors were not as robust for short‐term memory, the models were still able to predict a significant amount of variance (σ 2 = 0.023, Fig. 2). Conclusion: We successfully implemented CPM to derive robust brain‐based predictors of memory performance and neuroticism. Models derived separately for each sex differ in their ability to explain variance in these measures. Future analyses will pinpoint specific edges within the connectivity matrices that explain variance in memory performance or neuroticism in men vs women. We anticipate these will yield important insights into why women are at higher risk of developing AD.