Automated interviews on clinical case reports to elicit directed acyclic graphs

Area of Science:

• Medical Informatics
• Clinical Decision Support
• Artificial Intelligence in Medicine

Background:

• Hospital information systems facilitate clinical data recording.
• Directed acyclic graphs (DAGs) are crucial for probabilistic models like Bayesian networks in medicine.
• Medical practitioners often lack training in eliciting DAG features for causal reasoning.

Purpose of the Study:

• To design an automated interview to tutor medical doctors in developing DAGs.
• To represent clinicians' understanding of clinical reports using causal models.
• To address the difficulty in applying causality concepts before identifying all relevant patient variables.

Main Methods:

• Analysis of medical notions to identify patterns in clinical reasoning.
• Development of algorithms to support causal DAG elicitation.
• Definition of clinical relevance to focus questioning on causally related variables.

Main Results:

• An automated interview with questions phrased in medical terms was developed.
• The interview elicits information on chief complaints, diagnostic hypotheses, manifestations, and risk factors.
• Subsequent sessions refine causal paths by incorporating syndromes, dysfunctions, and modifiers, illustrated by case studies.

Conclusions:

• The elicitation framework demonstrates consistency with medical knowledge.
• It progressively introduces relevant medical topics for DAG development.
• Elicited DAGs are validated against established biomedical knowledge sources.