Using social networks to predict cognitive resilience among adults aged ≥ 80 years: insights from the Framingham Heart Study

Aim: Social engagement is associated with preserved cognition in older adults. We evaluated the predictive value of structural and functional social network characteristics for cognitive resilience among adults aged ≥ 80 years using machine learning models. Methods: We included Framingham Heart Study Offspring participants aged ≥ 80 years and free of dementia at Exam 8 (2005–2008). Eleven social network measures were assessed using the Berkman-Syme Social Network Index. Cognitive resilience was defined as remaining dementia-free through December 2024. Minimum Redundancy Maximum Relevance (mRMR) identified informative features, and principal component analysis (PCA) characterized their multidimensional structure. Logistic regression, classification and regression trees, random forest, XGBoost, and LightGBM were evaluated using five-fold cross-validation. Performance was assessed using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, positive predictive value (PPV), negative predictive value, F1 score, and balanced accuracy. SHapley Additive exPlanations (SHAP) was used to interpret the best-performing model. Results: The sample of 141 participants consisted of 56.7% females with a median age of 82 years. In total, 107 participants were cognitively resilient. The most frequently selected feature was “Feeling love and affection”. PCA identified one component reflecting overall social connectedness and support (PC1) and another reflecting interaction with friends and social activities (PC2). LightGBM including PC1 and PC2 had the highest discrimination (AUC = 0.694; 95% CI: 0.456–0.897), with PPV = 0.850, F1 = 0.654, and balanced accuracy = 0.616. Within LightGBM, AUC increased from 0.613 to 0.694 after adding PC1 and PC2. SHAP ranked PC1 and PC2 among the four most influential predictors. Conclusions: Social network characteristics reflecting perceived love, emotional support, family connections, and social interaction may provide incremental predictive information for cognitive resilience. However, overall predictive performance was modest, and larger, more diverse studies incorporating broader biological, behavioral, and clinical factors are needed.