NIH-developed AI algorithm matches potential volunteers to clinical trials
Researchers from the National Institutes of Health (NIH) have developed an artificial intelligence (AI) algorithm to help speed up the process of matching potential volunteers to relevant clinical research trials listed on ClinicalTrials.gov.
A study published in Nature Communications found that the AI algorithm, called TrialGPT, could successfully identify relevant clinical trials for which a person is eligible and provide a summary that clearly explains how that person meets the criteria for study enrollment. The researchers concluded that this tool could help clinicians navigate the vast and ever-changing range of clinical trials available to their patients, which may lead to improved clinical trial enrollment and faster progress in medical research.
A team of researchers from NIH’s National Library of Medicine (NLM) and National Cancer Institute harnessed the power of large language models (LLMs) to develop an innovative framework for TrialGPT to streamline 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 from ClinicalTrials.gov for which a patient is eligible and excludes trials for which they are ineligible. TrialGPT then explains how the person meets the study enrollment criteria. The final output is an annotated list of clinical trials—ranked by relevance and eligibility—that clinicians can use to discuss clinical trial opportunities with their patient.
To assess how well TrialGPT predicted if a patient met a specific requirement for a clinical trial, the researchers compared TrialGPT’s results to those of three human clinicians who assessed over 1,000 patient-criterion pairs. They found that TrialGPT achieved nearly the same level of accuracy as the clinicians.
Additionally, the researchers conducted a pilot user study, where they asked two human clinicians to review six anonymous patient summaries and match them to six clinical trials. For each patient and trial pair, one clinician was asked to manually review the patient summaries, check if the person was eligible, and decide if the patient might qualify for the trial. For the same patient-trial pair, another clinician used TrialGPT to assess the patient’s eligibility. The researchers found that when clinicians use TrialGPT, they spent 40% less time screening patients but maintained the same level of accuracy.
Given the promising benchmarking results, the research team was recently selected for The Director’s Challenge Innovation Award to further assess the model’s performance and fairness in real-world clinical settings. The researchers anticipate that this work could make clinical trial recruitment more effective and help reduce barriers to participation for populations underrepresented in clinical research.
