Dementia Models Exhibit Socioeconomic Status Performance Disparities
Researchers at Mayo Clinic (JMIR Med Inform, 2026) analyzed two cohorts—the Mayo Clinic Study on Aging (n=3,041) and the Rochester Epidemiology Project (n=19,572)—to evaluate SES and sex bias in four ML models predicting 1-year dementia onset using 5-year EHR data. They found higher balanced error rates for lower SES groups, that the individual-level HOUSES index detects different biases than area-level ADI, and that SMOTE-NC oversampling reduced some disparities but sometimes worsened others.
Key Points
- 1Show higher balanced error rates for lower SES groups across two cohorts (n=3,041; n=19,572).
- 2Indicate HOUSES individual-level SES detects bias differently than area-level ADI, altering fairness assessment.
- 3Recommend targeted oversampling reduces SES disparity but can worsen fairness under alternative SES definitions.
Scoring Rationale
Peer-reviewed, multicohort evidence of SES bias with practical mitigation, limited to dementia models and definition-specific tradeoffs.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
