A new study “A machine learning approach to multifactorial modeling of episodic memory performance” led by Dr. Marianne Chanti-Ketterl, Associate Core Leader at the Duke-UNC Alzheimer’s Disease Research Center (ADRC), Dr. Evan Fletcher from UC-Davis, and Dr. Zvinka Zlatar used machine learning to uncover which life experiences and habits are most closely linked to memory performance in older adults. The research, conducted between Duke University, UC Davis, and University of California San Diego, analyzed data from over 2,200 people aged 54 to 90, using a powerful algorithm called XGBoost to find patterns in how 15 different lifestyle and demographic factors relate to memory.

Key Takeaways

• Top Influencers: Education level, income, and sleep quality were the strongest predictors of better memory.
• Other Helpful Factors: Volunteering and being female were also linked to better memory performance.
• Less Impactful: Race and ethnicity had little effect overall, though there were some differences for White and Latinx participants.
• Negative Influences: Smoking, heavy drinking, and past head injuries had moderate negative effects.
• Interesting Patters:
  • Memory benefits from education seemed to level off after about 12 years of schooling.
  • Higher income was linked to better memory with no clear limit.
  • Sleep showed a “just right” pattern – too little or too much wasn’t ideal.

Why It Matters

This study shows how machine learning can help untangle the complex mix of factors that affect brain health as we age. Understanding these relationships could help guide strategies to prevent Alzheimer’s and other forms of dementia—especially in diverse populations.

---

Fletcher, Evan & Chanti-Ketterl, Marianne & Hokett, Emily & Lor, Yi & Venkatesan, Umesh & Whitmer, Rachel & Chen, Ruijia & George, Kristen & Zlatar, Zvinka. (2025). A machine learning approach to multifactorial modeling of episodic memory performance. Alzheimer’s & Dementia. 20. 10.1002/alz.089749.