Would you use AI to help with cancer treatment decisions? That was the question researchers explored in a recent study examining how humans interact with AI to make clinical decisions on therapy for non-small cell lung cancer and liver cancer. Read on for a link to that study and further details!

This study, partly funded by NCI’s Division of Cancer Treatment and Diagnosis, found mixed results. Some clinicians were skeptical of AI’s recommendations and unwilling to adjust their decisions. Others were willing to consider AI’s choices on a case-by-case basis, especially when the recommendations were clearly beneficial to patients. Those clinicians tossed out suggestions that could be potentially harmful (e.g., decisions that would put organs at risk or cause toxicity).

The researchers also found key themes underlying the complex interaction that’s at play between the clinician and the AI tool. If you’re considering using AI for making decisions, these themes might be familiar to you, including:

• Trust. Can you trust AI to make the right decision?
• Agreement. Do you agree with AI’s recommendation? If so, you’re more likely to take that approach. But what if you disagree? Will AI change your mind?
• Cancer type. Different cancers present different treatment challenges. In the study, clinicians were more likely to adjust their decisions based on AI’s recommendations for lung cancer, compared with liver cancer, especially when the recommendations deviated from a typical “standard of care.”

Senior author, Dr. Issam El Naqa, of the Moffitt Cancer Center, said “Today’s AI tools aren’t perfectly accurate, and can be biased and limited depending on the quality of their training data. However, when treating diseases like cancer, which can have a narrow window for treatment and a small margin for error, AI could help experts select more precise treatment and timing for that therapy.”

He added, “AI-assisted decision-making may be especially beneficial when choosing newly emerging, adaptive treatment strategies, which haven’t yet been widely incorporated into standard clinical practice. That’s an area where we’ll be focusing on next in studying collaborative decision-making.”