George Kumar, Senior Director of Medical Diagnostics, Pan Tumor and GI Cancers at AstraZeneca, shared a post on LinkedIn about a recent article by Ravi B. Parikh et. al, published in Journal of Clinical Oncology:
“Stop building medical AI that just predicts risk. Start building AI that guides decisions.
A critical takeaway from the recent Journal of Clinical Oncology (JCO) guide on testing AI models highlights a major gap in the current landscape.
Too many current AI tools excel at prognostication—telling us a patient is at high risk—but fail to offer actionable guidance on which therapeutic strategy would be most effective.
If an AI tool doesn’t help an oncologist make a better decision at the point of care, its clinical utility is limited.
The new JCO guidelines set a high bar for bringing trustworthy AI to the clinical forefront, emphasizing:
- True Clinical Utility: Moving beyond risk scores to decision support.
- Validation over ‘Explainability’: Prioritizing rigorous external validation as the primary driver of trust.
- Real-World Representativeness: Ensuring training data reflects diverse patient populations to prevent baked-in bias.
We need to shift focus from ‘Will it work in the lab?’ to ‘Will it improve patient outcomes in the clinic?’
The full guide is an essential read for ensuring your AI tools are ready for the real world.”
Title: Bringing Trustworthy Artificial Intelligence to the Clinical Forefront at JCO: A Guide for Studies Testing Artificial Intelligence Models
Authors: Ravi B. Parikh, Alexia Iasonos, Andrew Ko, Jeremy Warner, Kathy Miller, Jonathan W. Friedberg
You can read the full article in Journal of Clinical Oncology.
More posts featuring George Kumar.