In a session at the 2024 Community Oncology Conference titled “Is AI Ready for Prime Time for Cancer Care?”, Jim Chen, MD, American Oncology Network, and Sanjay Doddamani, MD, MBA, Guidehealth, provided an overview of where artificial intelligence (AI) technology stands today and its potential for use in cancer care.

Dr Chen began the session with a brief history of the origins of AI. The development of AI began in the 1950s and went through different stages of models, such as expert systems in the 80s, statistical models that gave machines more flexibility during the 90s and early 2000s, and neural network models that had less human involvement in the output in the 2010s.

Dr Chen noted that within the past seven years, Google built the transformer architecture, which “kicked the door down” in processing language in different ways. Some examples of this transformer architecture technology include large language models such as ChatGPT, Facebook’s Meta, and Google’s Gemini. There are also tools available such as text to image, text to video, text to audio, and text to code.

In terms of cancer care, Dr Doddamani suggested that we are in the early stages of the use of AI in oncology. “We’re at the first round of what is going to be an explosion. There’s going to be AI-aided oncologists and non-AI-aided oncologists,” he said.

AI is already capable of simple tasks such as summarization. It can also save providers time on tedious tasks of data entry. AI has the potential to assist with aggregating the overwhelming amount of data, literature, and guidelines needed to treat patients. It can help integrate information from a patient’s chart and the information they give to their providers into actionable decisions.

One of the issues that Dr Doddamani said AI can assist with is the “tightening up” of the primary care–specialty relationship. He states that machine learning and algorithms can help with risk prediction, prognostication, and shared decision-making. Along similar lines, Dr Chen brought up that general physicians can use AI to help guide decision-making for the next step of care, since general physicians are not specialized in hematology.

Dr Doddamani also mentioned the importance of having partners in different fields. Dr Chen added that within the oncology team there should be investment in the language models as they are being developed, and at least one member of the team should become knowledgeable about the language. When working with AI companies to test or implement the technology, Dr Doddamani suggested considering four criteria: Are they saving you time? Are they saving you money? Are they increasing survival? Are they increasing human dignity?

Both speakers acknowledged there are certain challenges and limitations for using AI in cancer care. On the technical side, Dr Chen discussed the issue of “hallucinations,” which is when AI will make up words or ideas because the language model is trying to please the user. This can be addressed by changing how the numbers are added and multiplied through fine tuning and/or training. The AI must also be given text such that the data that feeds into it is trustworthy.

In terms of practical implementation of the technology, Dr Doddamani said there is an “adoption crisis” in health care and that it takes about 10 years to have universal adoption of new discoveries. “Questioning things is incredibly good, but having early adoption to test out and trial various solutions I think is really important,” he said.

Source: Chen J, Doddamani S, Rahman A. Is AI ready for prime time for cancer care? Presented at the 2024 Community Oncology Conference; April 4-5, 2024. Orlando, Florida.