Reviewed by Lexie CornerNov 20 2024
Researchers led by Cedars-Sinai have developed and validated a new artificial intelligence (AI) method aimed at streamlining and accelerating the initiation of cancer clinical trials. This approach uses patients’ pathology reports to automate the classification of cancer severity, potentially reducing the time needed to identify suitable trial candidates. The findings, published in the peer-reviewed journal Nature Communications, represent a notable advancement in the application of AI in healthcare.
The new AI method, referred to as a model, offers a valuable alternative to tumor registries, which are databases managed by governments and hospitals.
Researchers commonly use tumor registries to screen cancer patients for clinical trials. However, this process depends on specialized personnel manually determining a patient’s cancer stage by reviewing laboratory reports, clinicians’ notes, and other documentation, making it both time-consuming and labor-intensive.
By the time a cancer patient’s data is entered into a tumor registry, months may have passed, along with the opportunity for the patient to participate in relevant clinical trials or other treatments. Our AI model can dramatically reduce that delay, accelerating the pace of research and expanding patients’ access to clinical trials.
Nicholas Tatonetti, Ph.D., Vice Chair, Computational Biomedicine, Cedars-Sinai
Nicholas Tatonetti is also the Associate Director for Computational Oncology at Cedars-Sinai Cancer and the corresponding author of the study.
The AI model developed by the team efficiently determines the cancer stage by analyzing and interpreting text from a single component of a patient’s electronic health record: the pathology report, which details pathologists' findings from tissue sample examinations. Tests involving thousands of patient records demonstrated the model's high effectiveness in accurately staging cancers.
The method utilizes a transformer AI model, designed to emulate the human brain’s complex decision-making processes. Initially, the team trained the model using publicly accessible pathology reports from The Cancer Genome Atlas, encompassing nearly 7,000 patients and 23 cancer types.
To ensure the model's adaptability across different environments, researchers applied it to approximately 8,000 pathology reports from a single medical center. The model's performance, assessed using a standard metric for AI evaluation, was rated as highly accurate.
This was an important finding because it means that our AI model is an ‘off- the-shelf’ tool that can be generalized to other institutions without requiring that it be trained for each location.
Nicholas Tatonetti, Ph.D., Vice Chair, Computational Biomedicine, Cedars-Sinai
According to Tatonetti., in addition to screening patients by cancer stage for clinical trials, the AI model can automate the classification of patients for observational studies, retrospective data analyses, and potential treatment options.
“Future research could build on our method to integrate the pathology text with other types of clinical data, potentially advancing personalized cancer treatment,” he added.
Previous research led by Tatonetti made the development of the AI model possible. This research addressed technical challenges in extracting and analyzing pathologists’ notes from electronic health records.
Notably, the researchers have made their AI model, BB-TEN (Big Bird—TNM staging Extracted from Notes), available to other institutions for academic and selected additional uses.
By speeding up the selection of candidates for cancer clinical trials, this innovative AI model shows promise for accelerating the development of relevant treatments and making them available to more patients.
Jason Moore, Ph.D., Chair, Department of Computational Biomedicine, Cedars-Sinai
The study's authors from Cedars-Sinai include Jacob Berkowitz, Jose M. Acitores Cortina, and Kevin K. Tsang, along with Jenna Kefeli as an additional contributor.
Kefeli and Tatonetti received support from award number R35GM131905, granted by the National Institute of General Medical Sciences of the National Institutes of Health.
Journal Reference:
Kefeli, J., et al. (2024). Generalizable and automated classification of TNM stage from pathology reports with external validation. Nature Communications. doi.org/10.1038/s41467-024-53190-9.