Artificial intelligence may assist physicians in determining which childhood cancer survivors require additional support, according to research published on March 27 by scientists from St. Jude Children's Research Hospital in Communications Medicine.
The study is significant because many effects of childhood cancer and its treatment can last long after the disease is cured, making it difficult for doctors to identify survivors who need targeted care. Much of the information that could inform this process is found in transcripts of conversations and survey responses that are not easily reviewed by clinicians. The integration of AI could help analyze these large amounts of conversational data, potentially improving survivorship care.
Researchers interviewed 30 survivors aged between 8 and 17 years old along with their caregivers. Two human experts analyzed over 800 pieces of information from conversation transcripts to categorize symptoms by severity and impact. The same data was then analyzed using two large language models—ChatGPT and Llama—with four different prompting strategies. Results showed that more complex prompts led to better performance by the AI models compared to simpler approaches.
"About 40%-60% of a clinical encounter is a patient talking to their physician about symptoms and related health experiences," said I-Chan Huang, PhD, corresponding author from the St. Jude Department of Epidemiology & Cancer Control. "We have provided a proof of concept that large language models could help analyze that underutilized conversational data to detect symptom severity and its functional impact and assist physician decision-making to provide better care to survivors." Huang also said, "We found that simple prompts were not effective," but noted that more sophisticated prompting strategies had higher concurrence with human reviewers.
The study highlights chain-of-thought and generated knowledge as effective complex prompting methods for distinguishing physical and cognitive impacts on survivors, though they were less successful at detecting social impacts. While further testing will be needed before clinical use, researchers say these findings offer one of the first concrete examples showing how AI might improve survivorship care.
"These AI-driven approaches provide us with a new way to unlock the complex symptom information hidden in the wealth of patient-physician conversations that currently go unused," Huang said. "By making this information easier to capture and analyze, we can help physicians better identify survivors who need additional support in real time and improve care for this growing population."
