The CTO Show with Mehmet Gonullu #231 Data Analytics and AI in Healthcare: A Conversation with Flatiron Health's CTO Catherine Miller
Oct 5, 2023
Catherine Miller, CTO of Flatiron Health and health-tech leader focused on real-world oncology data and AI. She explores how real-world data speeds drug development and improves trial design. She explains ML techniques to extract unstructured EHR information and multimodal approaches with images and genomics. She discusses AI risks, data privacy, and the need for equity-minded models in healthcare.
AI Snips
Chapters
Transcript
Episode notes
Unstructured Clinical Text Is The Bottleneck
- Cancer EHRs contain structured fields and a large volume of unstructured clinical text that must be extracted to enable research.
- Miller explains some extractions are simple like smoking status, while others like progression or surgical depth are complex and require advanced models.
Multimodal Data Unlocks Stronger Insights
- Imaging and genomic data are separate but complementary modalities that add power when linked to clinical records.
- Miller notes scans are often stored apart from notes, creating a multimodal integration problem that Flatiron is addressing.
Combine Models With Human Checks
- Don't trust LLM outputs blindly; design layered checks and human review to manage hallucination and model errors.
- Miller recommends hybrid workflows where models summarize or pre-extract and humans validate a percentage to monitor accuracy.