Explainable SHAP-XGBoost models for identifying important social factors associated with the atherosclerotic cardiovascular disease risk score using the LASSO feature selection technique.

Abstract

Objectives: Extensive evidence indicates that social factors play an essential role in explaining atherosclerotic cardiovascular disease (ASCVD). This study aimed to examine which social factors are associated with the estimated 10-year ASCVD risk score among male and female adults, incorporating both multifaceted social network components and conventional risk factors.

Methods: Using data from 4368 middle-aged Korean adults, we explored factors most likely to explain ASCVD risk with interpretable machine learning algorithms. The ASCVD risk was determined using the 10-year ASCVD risk score, as calculated using pooled cohort equations. Social network components were assessed through the name generator module. A total of 52 variables were included in the model.

Results: For male participants (area under the receiver operating characteristic curve [AUC] = 0.65), the average years known for network members contributed most to ASCVD risk prediction (mean Shapley additive explanations [SHAP] value = 0.31), followed by spouse's education level (0.22), medical history with diagnosis (0.18), and snoring frequency (0.14). By contrast, for female participants (AUC = 0.60), medical history with diagnosis was the strongest predictor (0.47), followed by logged income (0.21), education level (0.19), and the average number of years known in network members (0.17).

Conclusion: Several important social factors were associated with the ASCVD risk score in both male and female adults. However, longitudinal research is needed to determine whether these factors predict future ASCVD events.