AI system shows promise in matching patients with clinical trials

A new artificial intelligence system called TrialGPT could help solve one of healthcare’s persistent challenges: matching patients with appropriate clinical trials. Researchers from the National Institutes of Health (NIH) and University of Illinois have developed an end-to-end framework that uses large language models to streamline and improve the trial matching process.

The study, published in Nature, demonstrates that TrialGPT can significantly reduce screening time while maintaining high accuracy in patient-trial matching. The system achieved an accuracy rate of 87.3% in evaluating patient eligibility criteria, approaching the performance level of human experts.

“Our study shows that TrialGPT could help clinicians connect their patients to clinical trial opportunities more efficiently and save precious time that can be better spent on harder tasks that require human expertise,” said Zhiyong Lu, PhD, NLM Senior Investigator and corresponding author of the study, in a press release.

TrialGPT works through three key components. The system begins by efficiently filtering through large databases of clinical trials, identifying the most promising matches while examining only a fraction of the total collection – achieving over 90% accuracy while reviewing just 6% of available trials. Once it identifies potential matches, TrialGPT conducts a detailed analysis of each trial’s eligibility criteria, providing clear explanations for why a patient may or may not qualify. In the final stage, the system generates comprehensive ranking scores that help clinicians quickly identify the most suitable trials for their patients.

Related articles
First fully AI-generated drug enters clinical trials
Tackling dismal diversity rates in oncology clinical trials

In a pilot user study conducted at the National Cancer Institute, TrialGPT demonstrated significant efficiency gains. Medical professionals using the system completed trial screening 42.6% faster on average compared to traditional manual methods. The time savings were even more pronounced for complex cases, with screening time reduced by up to 43.5% for longer patient cases.

The researchers evaluated TrialGPT using three public datasets containing 183 synthetic patients and over 75,000 trial annotations. The system outperformed existing methods by 43.8% in ranking and excluding trials. Importantly, TrialGPT provides clear explanations for its recommendations, allowing medical professionals to understand and verify its decisions.

While the results are promising, the researchers acknowledge several limitations. The current system focuses primarily on analyzing written patient summaries and trial criteria. Future work will need to incorporate additional data types like lab values and imaging results. The researchers also note that real-world implementation would need to consider factors like trial locations, recruitment status and  bias within the AI model itself.

“Machine learning and AI technology have held promise in matching patients with clinical trials, but their practical application across diverse populations still needed exploration,” said NLM Acting Director, Stephen Sherry, PhD in the release. “This study shows we can responsibly leverage AI technology so physicians can connect their patients to a relevant clinical trial that may be of interest to them with even more speed and efficiency.”