Integrating Social-environmental Determinants into Machine Learning Algorithms in Dementia Care: A Systematic Review

Authors: Basma M. Saleh and Mohamed Salama Institution: Institute of Global Health and Human Ecology (IGGHE), AUC, Cairo, Egypt

1. Introduction & The Global Burden

• Global Crisis: Dementia affected 57 million people in 2021 → projected 153 million by 2050.

• Economic Burden: $1.3 trillion annually (2019); dementia = 7th leading cause of death, major driver of disability/dependency.

• Preventable Risk: 2024 Lancet Commission → 45% of dementia cases preventable via 14 modifiable risk factors.

2. Methodology

• PRISMA-Compliant Systematic Review:

  • Databases: PubMed, ProQuest, Scopus.

  • Initial results: 2,000 → narrowed to 24 pivotal studies.

• Publication Window: 2024–2026, focusing on latest AI/ML applications to social-environmental data.

• Targeted Search Strategy: MeSH terms combining “Dementia/Alzheimer” + “Artificial Intelligence” + “Social/Environmental factors.”

3. Analytical Framework

Multi-modal Data Fusion

• Clinical & genetic data.

• Environmental exposure metrics (air pollution, geospatial indicators).

• Education levels, HOUSES index.

• LLMs for SEDH from EHR.

Equity-Driven Modeling

• Socioeconomic status (SES).

• Social isolation.

• “Fairness-aware” algorithms to detect, quantify, mitigate bias.

• Focus on racial & socioeconomic disparities.

Translational Care Pathways

• AI-SARs (AI-based social assistive robots).

• Algorithmic insights → real-world digital health tools.

• Mobile ecosystems (e.g., “CareBuddy”).

4. Results & Insights

• Social Disadvantage vs Biology: Baez S. et al. (2026) → social disadvantage predicts dementia risk more accurately than purely biological/genetic knowledge.

• Enhanced Predictive Accuracy: Integrating SEDH via multi-modal models improves AUC and clinical relevance vs single-modality approaches.

• Algorithmic Bias: Models underperform for lower SES groups; individual-level SES measures more effective for bias mitigation than area-level metrics.

5. Conclusion: The Future of Care

• Fairness-by-Design: AI architectures must embed ethics & equity from design phase to avoid amplifying disparities.

• Population-Centered Integration: Outputs should integrate into community-level interventions & clinical decision-making for equitable care.

6. Select Key References

• Global Forecasting & Trends: GBD 2019 Dementia Forecasting Collaborators, Lancet Public Health, 2022.

• Lancet Commission 2024: Livingston G. et al., Lancet, 2024.

• Social Disadvantage Study: Baez S. et al., Alzheimer’s & Dementia, 2024.

• Sociodemographic Impact: Liu X. et al., JMIR Medical Informatics, 2026.

Contact: Basma_Saleh@aucegypt.edu