New ML model predicts cancer survival from care team collaboration patterns

Cancer care is a longitudinal, team-based process where multiple healthcare professionals (HCPs) coordinate treatments over time. While prior research has focused on clinical and demographic factors, little attention has been paid to how HCP collaboration evolves during the delivery phase. Huang, Lu, and Ma from UC Davis fill this gap by modeling EHR-mediated HCP interactions as dynamic networks and applying machine learning to predict patient survival. Their approach extracts predictive signals from the structure of these collaborative networks, identifying which teamwork patterns are associated with better outcomes. The model is validated through robustness analyses to ensure stability, and its insights align with established medical literature, providing empirical evidence for previously hypothesized links between team coordination and patient prognosis.

This work contributes a practical workflow for leveraging digital traces of collaboration to evaluate and strengthen team-based healthcare. By pinpointing specific network characteristics and dynamic patterns, the model can guide data-informed interventions to optimize care delivery. For example, identifying that certain patterns of handoffs or communication frequency correlate with improved survival could lead to redesigned team workflows. The study underscores the untapped potential of EHR metadata as a source for understanding human factors in medicine, moving beyond clinical data alone. As healthcare increasingly relies on coordinated teams, this ML-driven approach offers a scalable way to measure and improve teamwork, with direct implications for patient outcomes in oncology and beyond.