Modeling and Interpreting Teamwork Dynamics in Cancer Care Outcome Prediction

2026-06-03

AI Summary

This study addresses a critical gap in cancer prognosis research by explicitly modeling the dynamic collaboration among healthcare teams as a temporal interaction network derived from electronic health records. Integrating dynamic network representation learning with interpretable machine learning, the work systematically uncovers prognostic signals embedded in team collaboration patterns. The proposed model robustly identifies specific team structures and interaction dynamics significantly associated with patient survival, with findings aligning well with established clinical hypotheses. Beyond confirming collaboration quality as a key prognostic factor, this research provides actionable, data-driven evidence to inform the optimization of care team composition and coordination in oncology settings.

Abstract

Cancer care requires a longitudinal approach in which treatments are planned and delivered over time according to the needs of each individual patient. While prior research has thoroughly explored how clinical and demographic factors, such as comorbidities and age, inform treatment planning, far less attention has been devoted to the delivery phase of care. Yet planning and delivery are both team-based processes that depend on coordinated efforts among multiple healthcare professionals (HCPs). As such, the human factors embedded in these collaborative practices are crucial to optimizing patient outcomes. Despite this importance, the existing literature on human factors in cancer care is limited, and very few studies have investigated how collaboration within care teams evolves over the course of treatment. To fill this gap, this work examine how HCPs' collaboration, captured through electronic health record (EHR) systems, affects cancer patient outcomes, with particular emphasis on teamwork dynamics. We represent EHR-mediated HCP interactions as networks and apply machine learning methods to identify predictive signals of patient survival embedded in these collaborative structures. We further interpret model predictions by pinpointing network characteristics and dynamic patterns associated with particular outcomes. We evaluate our model through robustness analyses to ensure that the findings are stable and not driven by stochastic variation in training. Additionally, our insights align with hypotheses proposed in the medical literature, and our results provide the empirical, data-driven evidence supporting these claims. Overall, our work contributes a practical workflow for leveraging digital traces of collaboration to evaluate and strengthen longitudinal team-based healthcare, offering actionable insights to guide data-informed interventions in healthcare delivery.

Problem

• teamwork dynamics
• cancer care
• healthcare collaboration
• patient outcomes
• longitudinal care

Innovation

• teamwork dynamics
• collaboration networks
• electronic health records (EHR)
• interpretable machine learning
• cancer care outcomes