With stories about OpenAI’s ChatGPT and Microsoft’s revamped Bing Search, natural language processing (NLP) is dominating the news cycle. Adding to this coverage is new research released by the University of British Columbia: researchers have successfully used NLP to analyze oncologist notes for patients’ initial consultation visits and predict the six-month, 36-month, and 60-month odds of survival for each patient with more than 80 percent accuracy. The findings were recently published in JAMA Network Open.
NLP is a form of artificial intelligence (AI) in which computers analyze, understand, and even generate text the same way human beings do. According to IBM, “NLP combines computational linguistics—rule-based modeling of human language—with statistical, machine learning, and deep learning models. Together, these technologies enable computers to process human language in the form of text or voice data and to ‘understand’ its full meaning, complete with the speaker or writer’s intent or sentiment.”
AI is particularly good at pattern recognition. With NLP, this translates to picking up on patterns across texts that may be easily missed by human readers. Researchers from the University of British Columbia leveraged this capability by training a NLP model to read oncologists’ notes. These notes include a variety of patient characteristics, such as age, type of cancer, health conditions, family history, etc. Weighing all of these factors to develop the best treatment plan is complex, so the AI “brings all this together to paint a more complete picture of patient outcomes,” said lead author John-Jose Nunez, PhD.
After training the model, the researchers tested it on 47,625 patients across six British Columbia Cancer sties. All the data was stored securely at BC Cancer and presented anonymously, which leads to an additional benefit of using AI to analyze notes: AI can maintain patient confidentiality more easily than chart reviews by humans.
“The great thing about neural NLP models is that they are highly scalable, portable, and don’t require structured data sets,” Nunez said. He explained that such models can be quickly trained on local data to optimize the model for that particular region.
Nunez and his team hope that their technology finds its place as a virtual assistant for oncologists, helping them make data-driven decisions and improve patient treatment plans.