Tackling AI Bias, Expanding Its Role in Precision Oncology: Davey Daniel, MD

In the second and final part of an interview, Davey Daniel, MD, chief medical officer of OneOncology, emphasizes that bias in artificial intelligence (AI) must be addressed to reduce rather than amplify disparities. He also shares his vision for how AI in precision oncology can evolve to promote equitable access for all patients.

These topics were explored further during the panel discussion, "Access for Everyone: Using AI in Precision Oncology," at the Patient-Centered Oncology Care conference in Nashville, Tennessee, on September 26. Watch part 1 to learn about current applications of AI in precision oncology and barriers preventing its broader adoption.

This transcript has been lightly edited; captions were auto-generated.

Transcript

Bias in AI models is a well-known challenge. What strategies are needed to ensure these tools reduce disparities rather than reinforce them?

Bias in AI models is a real concern. Even before AI, we saw that early precision medicine datasets really didn't reflect the community. They often lacked diversity, and they excluded patients who were often too sick to travel. I think that a lot of that has resolved as next-generation sequencing and other biotesting markers have really become standard of care. It's essential that training models for AI are built on datasets that really reflect real-world populations that we serve, rather than being a subset.

Another risk comes from really incomplete or fragmented clinical data. If you think about it, the models that are built on only a part of the patient's story won't really generate reliable insights. It's really important that we get transparency on how models are built and validated. If we don't understand when or why they may fail, we're at real risk, and human oversight is really critical to all of this.

We need clinicians really embedded in how these are done. AI should support clinical judgment, not really replace it. Just as importantly, the design of these tools really needs to intentionally account for the broad range of patients, and we need insight into where these gaps may occur.

Looking ahead, how can AI build on its current applications to shape the future of precision oncology and ensure equitable access for all patients?

I'm really eager to see the current use cases of AI in precision medicine expand. At OneOncology, we're particularly excited about projects we have around identifying potential clinical trial options for patients and bringing those directly to their physicians at the point of care so that it's more likely that they'll be acted upon.

There are a lot of unanswered questions in oncology and novel therapeutics that need to be tested. Having the right patients find the right trial is really one of the most impactful ways, I think.

The other case that I'm really excited about is the real breakthroughs that could come from using AI to detect patterns in datasets that humans could not see. There's just too much data to see the relationship, so uncovering how different mutations, expression patterns, and other omics interact to shape disease biology will be really helpful. I think that deep understanding will be essential for developing new drugs and opening new therapeutic options for patients.