Not Another EHR: Reimagining Physician Information Needs with Generative AI Technology

A team of 18 researchers from Microsoft, including Eric Horvitz and David Rhew, published a position paper titled 'Not Another EHR: Reimagining Physician Information Needs with Generative AI Technology' on arXiv. The paper argues that while EHRs have improved data accessibility, they have also introduced significant cognitive burden for physicians due to the sheer volume and complexity of patient data. The authors conducted semi-structured interviews with internal physicians at Microsoft to identify key pain points in data navigation and synthesis during diagnostic workflows. They found that clinicians struggle with fragmented data, excessive clicks, and time-consuming manual synthesis of information.

The researchers propose leveraging large language models (LLMs) to create dynamic, adaptive user interfaces that can anticipate physician information needs and surface relevant data proactively. For example, an AI-powered system could summarize a patient's history, highlight critical lab results, or suggest differential diagnoses based on the current context. The paper also examines how physicians conceptualize AI assistance, noting that trust and interaction expectations depend on transparent explanations and human-in-the-loop control. The authors outline design considerations for generative user interfaces that prioritize clinician-centered workflows, aiming to reduce cognitive load and improve diagnostic accuracy. This work aligns with broader efforts to integrate AI into healthcare IT, potentially reshaping how doctors interact with patient data in clinical settings.