ChatGPT Shows Promise in Emergency Diagnoses at WVU Health

Researchers at West Virginia University evaluated the effectiveness of various ChatGPT models in assisting emergency room physicians with disease diagnoses based on exam notes. The study found that while these AI models performed well with commonly presented conditions, they struggled significantly with atypical cases, such as pneumonia without fever, highlighting their limitations. Despite a modest improvement in accuracy for top diagnoses in newer models, the overall performance remained similar across versions, underscoring the critical need for human oversight in medical diagnostics.

Opens on WVU Today