Background

Social engagement has been linked with preserved cognition in older adults, but it remains unclear whether social network measures can predict late-life cognitively resilience.

Methods

We included 148 dementia-free participants from the Framingham Heart Study Offspring cohort aged over 80 years at baseline (2005-2008). Social network was assessed with the Berkman-Syme Social Network Index questionnaire, including 11 indices.  Individuals were cognitively resilient if they did not develop dementia by December 2024. Minimum Redundancy Maximum Relevance with five-fold cross-validation (CV), repeated 100 times, was used to rank feature relevance based on their selection frequencies. Eleven sequential classification and regression trees with CV were applied based on the base model, with each adding one social network feature in descending order of relevance. The base model included only sex, age, marital status, education, and APOE genotype. Model performance was evaluated by average Area Under the Curve (AUC), sensitivity, specificity, Positive Predictive Value (PPV), and Negative Predictive Value (NPV).

Results

The sample consisted of 83 females (57%) with a mean age of 82.6 years (SD=2.3). 115 participants were classified as cognitively resilient. The most frequently selected feature was “Feeling love and affection”, while the least selected feature was “Someone listens to you when you need to talk”. The AUC and sensitivity of the base CART model were 0.576 and 0.583, respectively (Figure 1). Adding social network features increased the sensitivity, peaking at 0.63 when the top 10 features were included. Other model performance measures showed minimal improvement.

Conclusion

Feeling loved emerged as the most important social factor associated with cognitive resilience in adults over 80. However, social network features did not substantially improve the predictive performance, underscoring the need for future studies using more refined measures and larger cohorts.