Can Artificial Intelligence Change the Practice of Oncology?

Artificial intelligence (AI) is all the rage. Although AI has been a part of our lives for a long time, the release of the online chatbot ChatGPT by the company OpenAI on November 30, 2022, has ushered in a new era of enthusiasm about the potential for AI to improve our lives. If you haven’t tried ChatGPT yet, you’re probably one of the few. Based on current estimates, the ChatGPT website has approximately 25 million visitors daily. Microsoft has invested $10 billion in OpenAI. Google has announced that it will be incorporating conversational AI into its search engine. And in medicine, the New England Journal of Medicine has announced that it is planning a new online journal called NEJM AI “to identify and evaluate state-of-the-art applications of artificial intelligence to clinical medicine.”

What could AI do for physicians in general and hematologists/oncologists in particular? To answer that question, I turned to ChatGPT itself.

First, AI could help radiologists read scans better and reduce the rates of false-positive and false-negative results, thereby detecting malignant tumors on imaging studies more accurately. Second, in the era of targeted therapies, AI could identify biomarkers and help physicians choose correct treatments.

Third, AI could improve remote monitoring of patients by interpreting data from wearable devices. This feature would be particularly useful as we start to use bispecific antibodies and monitor patients’ vital signs carefully for signs of cytokine release syndrome, but it would also help patients who are receiving conventional chemotherapy.

Fourth, AI-powered robots could assist surgeons performing operations, enabling a greater degree of precision and accuracy. Fifth, chatbots could listen to a visit between physician and patient and write a note summarizing the content.

Sixth, chatbots can already help us think through diagnostic challenges presented by patients. For example, I asked ChatGPT what diagnoses I should consider for a patient reporting blisters on the back of her hands. It listed a differential diagnosis of 6 conditions (although it failed to mention porphyria cutea tarda, which is what I was thinking).

I’m sure there are countless more examples of tasks that AI could help physicians with. In fact, this entire article about AI in oncology could have been written for me in approximately 10 seconds by ChatGPT. I know this because I tried it, and it worked. But I didn’t publish that version. That would be cheating.