Researchers at the University of Pennsylvania have launched Observer, a multimodal medical dataset derived from anonymized recordings of real-world patient–clinician interactions in primary care. The dataset links video, audio, and transcripts from clinical visits and provides de-identified, structured data derived from these encounters—such as measures of speaking time, turn-taking, pauses, gaze, and explanation patterns—enabling researchers to study how communication and interaction shape patient understanding, experience, and health outcomes, and to support research in clinical communication, medical education, and medical AI.
Image credits: Brian Wangenheim
