Loading Curie Briefs...
Getting the latest healthcare news for you
Getting the latest healthcare news for you
AI is transforming healthcare, but biased training data and inaccurate electronic health records are holding it back. Oncologist Chadi Nabhan, MD, MBA, argues that rigorous validation, diverse datasets, and human oversight are essential before AI can be safely deployed in clinical decision-making. He's cautiously optimistic the field will get there within five years.
AI is rapidly becoming a fixture in healthcare, but oncologist Chadi Nabhan, MD, MBA, says the technology has a serious data problem. Speaking at the COSMO Conference in Chicago, Nabhan explained that most AI models in oncology are trained on datasets skewed toward patients from large academic centers — often insured, predominantly white, and enrolled in clinical trials — making their outputs unreliable for the broader, more diverse patient population seen in community and rural settings.
Beyond bias, Nabhan flagged a subtler issue: data integrity. Electronic health records are riddled with copy-paste errors, outdated problem lists, and missing information, meaning AI can't distinguish between a patient who tested negative for a biomarker and one who was never tested at all. "Absence of evidence in the medical record is not evidence of absence," he noted.
Key Takeaways:
Why it matters: If AI is deployed without addressing bias and data integrity, it risks widening — not closing — healthcare disparities. Getting this right could democratize access to expert-level oncology care, especially in underserved and rural communities.
