Artificial intelligence-generated feedback on social signals in patient-provider communication: technical...

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Clinical Communication Analysis

Background:

• Implicit bias in healthcare leads to inequities, often seen in nonverbal patient-provider cues.
• Automated communication assessment using artificial intelligence (AI) can help clinicians recognize bias in interactions.

Purpose of the Study:

• To investigate the technical performance and impact of AI for automated communication assessment and feedback in primary care.
• To raise clinician awareness of implicit bias through AI-driven analysis of patient-provider interactions.

Main Methods:

• Developed a machine-learning pipeline to detect social signals (dominance, warmth, engagement) in 145 primary care visits.
• Engaged 24 clinicians to co-design AI feedback mechanisms for workflow integration.
• Evaluated the AI approach's impact in a prospective cohort of 108 primary care visits.

Main Results:

• AI models successfully identified social signals in nonverbal communication, demonstrating feasibility.
• Clinicians preferred personalized dashboards but found social signals a useful alternative to nonverbal cues.
• AI models showed fairness and generalizability; stronger clinician implicit race bias correlated with less provider dominance and warmth.

Conclusions:

• AI-driven systems show promise for assessing communication and providing feedback on social signals to reduce healthcare disparities.
• Future research should focus on improving AI performance, personalization, and implementation in medical education and telehealth.