Machine Learning to Identify Physician Actions Associated with Patient Experience of Compassion

Area of Science:

• Medical Education
• Patient Experience
• Health Outcomes

Background:

• Physician compassion is linked to improved patient outcomes.
• Specific physician behaviors influencing patient compassion experience require identification.

Purpose of the Study:

• To identify physician behaviors most strongly associated with patients' experience of compassion.
• To explore potential differences in these associations based on patient race and gender.

Main Methods:

• Cross-sectional study in two urban academic emergency departments (EDs).
• 1025 adult patients completed validated compassion measures and behavior checklists.
• LASSO regression identified key predictive physician behaviors.

Main Results:

• "Listen carefully to what you had to say" was the strongest predictor of patient-reported compassion.
• Nine behaviors collectively explained approximately one-third of the variation in compassion scores.
• Differential associations between behaviors and compassion were observed for women and Black patients compared to men and White patients, respectively.

Conclusions:

• Specific, identifiable physician behaviors are significantly associated with patient-reported compassion.
• Attentive listening is a critical behavior for enhancing patient compassion.
• Further research should focus on interventions to cultivate these behaviors and prospectively validate their impact.