Healthcare outcomes are influenced by a combination of social influence and individual clinical factors. Therefore, integrating heterogeneous data sources for improving predictive performance in healthcare analytics can be an effective strategy. In this paper we propose a multi-modal machine learning framework that combines social network metrics and clinical health features to predict health intervention outcomes. We evaluate the proposed model on a healthcare dataset with both social connectivity and health status information. The multi-modal model is compared against single-modality baselines and our results show that the multi-modal approach achieves superior prediction accuracy, with lower mean error and higher explained variance. Through detailed analysis including SHAP feature attribution, training convergence behavior, and node influence dynamics, we find that while health features when considered separately exhibit strong predictive power, adding social network information captures complementary effects that further improve predictions. We discuss how our framework can inform targeted interventions and future directions for integrating diverse data sources in healthcare.