Brain predictors of episodic memory change across race/ethnic groups: A machine learning investigation.

Objective: It is known that brain measures predict cognitive trajectories and that these predictions are not uniform across ethnoracial groups. But it remains unclear whether group differences reflect differing brain-cognition associations or similar associations but different demographic and brain health distributions. Method: We used eXtreme Gradient Boosting regression tree modeling to predict episodic memory (EM) trajectories from 64 brain measures, including regional gray matter thickness and white matter tract diffusion-tensor free water, with age, sex, education, and ethnoracial indicators in four well-established, ethnically and racially diverse U.S.-based cohorts . eXtreme Gradient Boosting combines many simple regression trees to model nonlinear relationships among predictors and outcome; predictor importance was ranked using SHapley Additive exPlanation values. Results: In a sample of 1,463 participants (mean age, 77.0 ± 8.4 years; 503 African American , 168 Asian, 289 Hispanic, and 503 White), a computationally based gray matter signature and white matter tract hippocampal cingulum free-water brain measures ranked first and second in global importance to predict EM trajectory, followed by uncinate fasciculus free water. Generalized additive model analyses by ethnoracial group and age strata showed equivalent pairwise SHapley Additive exPlanation-brain dependency curves for the top two brain measures. Mean absolute deviations were less than 5% of between-participant standard deviation for EM slope , indicating that brain-outcome SHapley Additive exPlanation dependencies were similar across age and ethnoracial groups. Conclusion: EM trajectory prediction profiles reflect different sample compositions of brain health measures rather than different brain–EM associations. This suggests that observed variability in brain-EM predictors derives from group-specific distributions of demographic predictors and brain health. Key Points Question: Are observed differences across ethnoracial groups in brain-cognition associations related to differences in underlying brain correlates of cognition or to population differences in brain health markers ? Findings: Individual mappings of brain markers onto predictive importance for longitudinal episodic memory were largely similar in three age strata and four ethnoracial groups, but group levels of brain health differed. Importance: By pointing to group-level distributions of brain health metrics as the source of observed association variance, these results may help resolve ambiguity in the interpretation of that variance while suggesting targeted population interventions to support brain health. Next Steps: This study lays out a framework for analyzing the predictive importance of brain and life exposure predictors for cognition using machine learning, enabling detailed models tailored to individual populations.

Authors

Institutions

Publication Details

Journal
Neuropsychology
Published
2026-11-16
DOI
https://doi.org/10.1037/neu0001138
Primary Topic
Dementia and Cognitive Impairment Research
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Brain predictors of episodic memory change across race/ethnic groups: A machine learning investigation.

Philippe G. Sarrazin, Evan M. Fletcher, Brandon E. Gavett, Zvinka Z. Zlatar et al.
1 links
Neuropsychology
Dementia and Cognitive Impairment Research
article

Brain predictors of episodic memory change across race/ethnic groups: A machine learning investigation.

Philippe G. Sarrazin, Evan M. Fletcher, Brandon E. Gavett, Zvinka Z. Zlatar, Lisa Laverne Barnes, Keith F. Widaman, Dan Mungas, Pauline Maillard, Shraddha Sapkota, María M. Corrada, Marianne Chanti‐Ketterl, Paola Gilsanz, Sarah Farias, Rachel Whitmer
article en

Abstract

Objective: It is known that brain measures predict cognitive trajectories and that these predictions are not uniform across ethnoracial groups. But it remains unclear whether group differences reflect differing brain-cognition associations or similar associations but different demographic and brain health distributions. Method: We used eXtreme Gradient Boosting regression tree modeling to predict episodic memory (EM) trajectories from 64 brain measures, including regional gray matter thickness and white matter tract diffusion-tensor free water, with age, sex, education, and ethnoracial indicators in four well-established, ethnically and racially diverse U.S.-based cohorts . eXtreme Gradient Boosting combines many simple regression trees to model nonlinear relationships among predictors and outcome; predictor importance was ranked using SHapley Additive exPlanation values. Results: In a sample of 1,463 participants (mean age, 77.0 ± 8.4 years; 503 African American , 168 Asian, 289 Hispanic, and 503 White), a computationally based gray matter signature and white matter tract hippocampal cingulum free-water brain measures ranked first and second in global importance to predict EM trajectory, followed by uncinate fasciculus free water. Generalized additive model analyses by ethnoracial group and age strata showed equivalent pairwise SHapley Additive exPlanation-brain dependency curves for the top two brain measures. Mean absolute deviations were less than 5% of between-participant standard deviation for EM slope , indicating that brain-outcome SHapley Additive exPlanation dependencies were similar across age and ethnoracial groups. Conclusion: EM trajectory prediction profiles reflect different sample compositions of brain health measures rather than different brain–EM associations. This suggests that observed variability in brain-EM predictors derives from group-specific distributions of demographic predictors and brain health. Key Points Question: Are observed differences across ethnoracial groups in brain-cognition associations related to differences in underlying brain correlates of cognition or to population differences in brain health markers ? Findings: Individual mappings of brain markers onto predictive importance for longitudinal episodic memory were largely similar in three age strata and four ethnoracial groups, but group levels of brain health differed. Importance: By pointing to group-level distributions of brain health metrics as the source of observed association variance, these results may help resolve ambiguity in the interpretation of that variance while suggesting targeted population interventions to support brain health. Next Steps: This study lays out a framework for analyzing the predictive importance of brain and life exposure predictors for cognition using machine learning, enabling detailed models tailored to individual populations.

Neuropsychology
University of California, Riverside (US), Kaiser Permanente (US), Duke University (US), University of California, Irvine (US), University of California San Diego (US), Duke Medical Center (US), Rush University (US), University of California, Davis (US), Université Grenoble Alpes (FR)
Foundation for the National Institutes of Health, National Institutes of Health
Good health and well-being
Dementia and Cognitive Impairment Research
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon