Sex differences in cognitive function trajectories and influencing factors in older adults: A machine learning study based on CHARLS and HRS.

XGBoost performed best among six machine learning models tested for predicting cognitive function trajectories.

• Six ML models were compared in the study
• XGBoost was identified as the optimal model across the analytical groups
• SHAP (SHapley Additive exPlanations) values were used to explain and rank feature importance in the XGBoost model
• The Boruta algorithm was used for variable screening prior to model application

Latent Class Growth Models identified distinct cognitive function trajectories among older adults in both China and the United States.

• Five waves of longitudinal data were utilized from both CHARLS and HRS
• CHARLS (China Health and Retirement Longitudinal Study) and HRS (Health and Retirement Study) provided the data sources
• LCGMs were employed to classify participants into trajectory groups
• The study covered older adult populations in two countries with different cultural and healthcare contexts

Education was the top influencing factor for cognitive function trajectories across all groups regardless of sex or country.

• Education ranked as the highest-importance feature based on SHAP values across all subgroups analyzed
• This finding was consistent across Chinese and American cohorts
• This finding was consistent across both male and female subgroups
• The result was identified using the Boruta algorithm for variable selection combined with SHAP importance ranking

Family factors had stronger influence on cognitive trajectories in the Chinese sample compared to the U.S. sample.

• Family-related variables ranked higher in feature importance for CHARLS participants than HRS participants
• The authors attributed this difference to traditional culture and differing social roles between the two countries
• Differences were also linked to public services available in each country

Medical and endowment insurance had stronger effects on cognitive function trajectories in the U.S. sample compared to the Chinese sample.

• Insurance-related variables showed higher SHAP importance values in the HRS (U.S.) group
• This contrast with China was interpreted in the context of differences in public service infrastructure and healthcare systems
• The finding highlights country-level structural differences in determinants of cognitive aging

Men's cognitive function trajectories were more influenced by external social networks, while women's were more influenced by family responsibilities and community environments.

• Sex-stratified analyses revealed differential feature importance rankings between male and female subgroups
• Social network variables ranked higher in importance for men
• Family responsibility and community environment variables ranked higher in importance for women
• The authors attributed these sex differences to differing social roles and traditional cultural expectations

Differences in cognitive trajectory influencing factors between countries and sexes were linked to public services, social roles, and traditional culture.

• Authors identified public services as a structural factor differentiating Chinese and American cognitive aging determinants
• Social roles were identified as a key explanatory mechanism for sex differences
• Traditional culture was cited as contributing to both sex-based and country-based differences
• These explanations were offered as interpretive context for the machine learning–derived feature importance findings