“Tell Me Where IT Hurts” Podcast: Guest Tim O’Connell, M.D., Founder and CEO, emtelligent, on AI

As a practicing radiologist and vice-chair of Clinical Informatics in the Department of Radiology at the University of British Columbia, Tim O’Connell brings a unique, hands-on perspective to his role as CEO of emtelligent, which offers clinical-grade Natural Language Processing (NLP) technology to help healthcare providers strategically structure their unstructured medical data. He also holds a master’s degree in engineering and a Bachelor of Science in neuroscience. His career includes 25 years of IT experience, with companies such as Bell Canada and Nortel Networks.

In this episode, Dr. Anders and Dr. O’Connell take a frank, hopeful, and sometimes skeptical look at OpenAI and generative AI and how it all relates to healthcare––the good, the bad, the ugly, and the crazy (to quote Dr. Anders). Where does it stand? Where is it heading? How is it best used now, and how should it be used in the future?

As veterans of both clinical medicine and healthcare IT, Dr. Anders and Dr. O’Connell have more than passing interests in how AI can or will be deployed to make clinician-facing technologies “delightful” to use. In this wide-ranging discussion, the two touch on many topics, including:

  • Anders tees up the conversation with his observations on OpenAI and generative AI in overall
  • Has OpenAI’s opened a can worms when it comes to public knowledge of generative AI?
  • emtelligent’s approach to unlocking good clinical data from free text notes and NLP––a “medical information extraction system”
  • Training models specifically on clinical data, precision-annotated by physicians
  • The non-determinism of generative AI
  • Hallucination is a major issue where models make up convincing but incorrect responses.
  • Lack of explainability around generative model outputs.
  • Where generative AI fits in now –– improving human-machine interaction
  • “Trust but verify” will always be key!
  • High quality dataset annotation and “fine-tuning” is essential for accurate NLP performance. Leveraging medical ontologies improves consistency. Will AI help streamline the auditing function?
  • Huge potential to use NLP to efficiently find patient cohorts for research studies vs manual chart review––and analyze large datasets to reveal new insights.
  • Future opportunities for AI to enhance genomic analysis and make EHR use more efficient and satisfying. But need to thoughtfully validate applications.
  • The pivotal role of terminology
  • Healthcare organizations should pressure IT vendors to leverage this technology
  • “If you had a magic wand and could change anything in healthcare IT today, what would it be?”

Show Links

 

###