Predictive Models of Decision Making in Medical Triage

Artificial intelligence (AI) is increasingly leveraged for decision support in complex, high-stakes environments such as medical triage. Training AI models typically relies on aligning with expert consensus; however, in these environments, expert decisions often diverge due to individual differences and contextual factors. Key Decision-Making Attributes (KDMAs)individual characteristics that interact with context to shape decisions-play a critical role in this divergence. Understanding and modeling these attributes can enhance AI's ability to align with trusted human decision-making. This research explores rulebased predictive modeling of KDMAs to improve AI decisionsupport systems in medical triage. We collected self-reported KDMA rankings and scenario-based decision-making data to develop models that predict individual decision-maker responses. By evaluating models based on single and multiple KDMAs, we identified which attributes best predict decision outcomes. Comparing predictive accuracy across these models allowed us to determine the most influential KDMAs, establishing a framework for AI that adapts to individual decision-making styles.