AI learns to identify childhood cancer survivors who need extra support through patient conversations

AI Analyzes Patient Conversations to Transform Survivorship Care

Artificial intelligence is learning to identify childhood cancer survivors who need extra support by analyzing patient conversations, according to research published in Communications Medicine by scientists from St. Jude Children's Research Hospital1

1

. The study demonstrates how large language models can process interviews with young survivors and their caregivers to detect multiple symptoms causing severe disruptions in daily life, offering physicians a new tool to improve survivorship care.

The research addresses a critical challenge in pediatric oncology: identifying which childhood cancer survivors have symptoms severe enough to warrant targeted intervention. "About 40%-60% of a clinical encounter is a patient talking to their physician about symptoms and related health experiences," said corresponding author I-Chan Huang, PhD, from St. Jude Department of Epidemiology & Cancer Control2

2

. "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."

Complex AI Prompting Strategies Outperform Simple Methods

The researchers interviewed 30 survivors between 8 and 17 old along with their caregivers, generating more than 800 analyzable pieces of conversational data1

1

. Two human experts first analyzed the conversation transcripts for signs of excessive pain and fatigue, categorizing symptoms by severity and their physical, cognitive or social impact. The scientists then tested two large language models—ChatGPT and Llama—using four different AI prompting strategies to see how well they matched the human experts' assessments.

The study compared two simple approaches—zero-shot and few-shot prompting, which provide minimal information beyond basic instructions—against two complex prompting methods: chain-of-thought prompting and generated knowledge prompting. The simple approaches produced unstable and inaccurate results. "We found that simple prompts were not effective," Huang explained. "However, our more sophisticated prompting strategies performed significantly better and had a higher concurrence with our human reviewers"2

2

.

Chain-of-thought prompting uses step-by-step logical instructions, while generated knowledge prompting asks the model to generate background information before receiving instructions. Both complex prompting methods successfully distinguished the physical and cognitive impact of symptoms on survivors, though they showed only moderate ability to detect social impacts1

1

.

Unlocking Hidden Information in Patient-Physician Conversations

Children treated for cancer face unique challenges because treatment occurs during critical developmental periods, creating a ripple effect that can manifest long after the initial disease is cured. Cancer- and treatment-related effects can emerge years later, but identifying which survivors need additional support proves difficult for physicians. Much of the relevant data exists in transcripts of conversations and answers to open-ended survey questions that cannot be reviewed quickly in clinical settings2

2

.

"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"1

1

.

Future Integration Requires More Testing

While clinical use will require much more testing, these early results suggest that chain-of-thought, generated knowledge or similar prompting methods should guide future integration of AI into clinical workflows. The findings provide one of the first concrete examples of how artificial intelligence may improve long-term survivorship outcomes for childhood cancer survivors. The study, led by first author Jin-ah Sim and supported by grants from the National Cancer Institute, offers a data analysis framework that could transform how physicians use health experiences shared during clinical encounters to deliver better care2

2

.

For the growing population of childhood cancer survivors, this research represents a potential shift in how physicians process conversational data to identify symptom severity and provide physician decision-making support. By leveraging artificial intelligence to analyze patient conversations that currently go underutilized, healthcare providers may soon have a more efficient way to ensure survivors receive the targeted support they need.