Related Experiment Video
Updated: May 8, 2026
A Machine Learning Approach to Design an Efficient Selective Screening of Mild Cognitive Impairment
Published on: January 11, 2020
The Brain Vulnerability Index: Development and Validation of a Machine Learning-Derived, Community-Informed
Yusuf Tamer1, Autumn Noon1, Zhiyi Yang2
1Parkland Center for Clinical Innovation.
Background:
Social determinants of health (SDOH) are increasingly recognized as important contributors to cognitive impairment, including Alzheimer's disease and related dementias. Existing indices are heavily weighted toward financial indicators and are not validated against cognitive outcomes. We aimed to develop and validate a novel brain-specific SDOH index that identifies high-risk communities for cognitive impairment.
Methods:
The Brain Vulnerability Index (BVI) integrates patient-level electronic health record data with neighborhood-level SDOH from the Community Vulnerability Compass, a population health analytics framework. In the model development phase, electronic health record data were obtained from patients seen at Parkland Health (Dallas, TX). Cases were defined as individuals with at least one ICD-10 diagnosis suggestive of cognitive impairment from 2015 to 2023 and controls were age-matched with no ICD-10 diagnosis. External validation against serial cognitive performance measured by the Montreal Cognitive Assessment (MoCA) and clinically adjudicated consensus diagnosis of cognitive impairment of any etiology was performed in community-based (Dallas Heart Study) and clinically based (Alzheimer's Disease Research Center) cohorts.
Results:
The model development sample included 39,570 cases and 192,060 controls. Derived BVI at the block group level achieved detection of electronic health documentation suggestive of cognitive impairment at a balanced accuracy of 54.6%. In external validation (n = 1,395), higher BVI values were associated with lower MoCA scores (β = -0.35; p = 0.009) and with faster cognitive decline (- 1.04 vs - 0.42 MoCA points/year in high- vs low-risk groups; p < 0.0001). BVI was also associated with adjudicated cognitive impairment in the community-based cohort only (p = 0.04). Across analyses, BVI correlated with but outperformed existing neighborhood indices.
Conclusions:
BVI, a novel multidimensional brain-specific SDOH index, is clinically anchored and is externally validated. It may assist in earlier identification of at-risk communities with accelerated cognitive decline and aid in resource allocation for targeted prevention strategies.
