Artificial intelligence algorithm successfully selects potential volunteers to participate in clinical trials

A team of researchers from the NIH National Library of Medicine (NLM) and the National Cancer Institute leveraged the power of large language models (LLM) to develop an innovative TrialGPT framework to simplify the clinical trial matching process. TrialGPT first processes a patient summary, which contains relevant medical and demographic information. The algorithm then identifies relevant clinical trials on ClinicalTrials.gov in which the patient is eligible to participate and eliminates studies in which the patient is not eligible. TrialGPT then explains how the person qualifies for the study. The end result is an annotated list of clinical trials, ranked by relevance and acceptability, that physicians can use to discuss clinical trial opportunities with their patients.

“Machine learning and artificial intelligence technologies have shown promise in recruiting patients to clinical trials, but their practical application in diverse populations still needs to be explored,” said NLM Acting Director Stephen Sherry, Ph.D. “This study shows that we can responsibly use artificial intelligence technology so that doctors can connect their patients to relevant clinical trials that may interest them with even greater speed and efficiency.”

To evaluate how well TrialGPT predicted whether a patient would meet certain criteria for a clinical trial, the researchers compared TrialGPT’s results with those of three human physicians who rated more than 1,000 patient-criteria pairs. They found that TrialGPT achieved almost the same level of accuracy as doctors.

The researchers also conducted a pilot user study in which they asked two human physicians to review six anonymous patient records and match them with six clinical trials. For each patient and study pair, one clinician was asked to manually review the patient’s summary information, check whether the individual was eligible for the study, and decide whether the patient was eligible for the study. For the same patient-subject pair, another clinician used TrialGPT to assess the patient’s eligibility for participation. Researchers found that when doctors use TrialGPT, they spend 40% less time screening patients while maintaining the same level of accuracy.

Clinical trials provide important medical discoveries that improve health, and potential participants often learn about these opportunities from their doctors. However, finding a suitable clinical trial for interested participants is a time-consuming and resource-intensive process that can slow down important medical research.

“Our study shows that TrialGPT can help physicians connect their patients to clinical research opportunities more efficiently and save valuable time that can be better spent on more complex tasks that require human expertise,” said NLM senior investigator and study author Zhiyong Lu, Dr. philosophy.

Given the promising results of the comparative analysis, the research team was recently selected for The Director’s Challenge Innovation Award to further evaluate the effectiveness and validity of the model in real-world clinical settings. The researchers expect that this work could make clinical trial recruitment more efficient and help reduce barriers to participation for populations underrepresented in clinical trials.

The study was co-authored by researchers at the Albert Einstein College of Medicine, New York; University of Pittsburgh; University of Illinois Urbana-Champaign; and the University of Maryland, College Park.