Ask ChatGPT if you are eligible for a clinical trial: It may help speed up the screening process, study finds

ChatGPT Could Speed Up Clinical Trial Screening, Study Finds (Image: Canva)

Imagine finding eligible patients for clinical trials in minutes, not hours. A new study from UT Southwestern Medical Centre suggests that ChatGPT may make this a reality.

How ChatGPT May Improve Trial Screening

The research, published in Machine Learning: Health, tested ChatGPT’s ability to assess if patients could join clinical trials. The AI was used to analyse data from 74 patients being considered for a head and neck cancer trial. It screened each case far quicker than humans normally can.

Today, screening patients is a slow and manual task. Doctors must go through every medical record, which can take up to 40 minutes per patient. This often leads to delays, added costs, and sometimes even failed trials.

Many electronic health records contain unstructured text like doctor’s notes. These details are hard for traditional software to read, meaning some eligible patients are missed. ChatGPT, however, was able to search through this information in just 1.4 to 12.4 minutes.

What Methods Did the Study Test?

Researchers used GPT-3.5 and GPT-4, trying three ways to guide the model. The most effective version, GPT-4, gave more accurate results but took slightly more time and cost. Each test ran between $0.02 and $0.27 per patient, offering a low-cost alternative to manual reviews.

ChatGPT was asked to give structured answers, explain its reasoning, or decide how to respond freely. The last two methods performed better in flexible trial scenarios. The researchers noted that while AI could not replace human reviewers, it can support them by saving time.

What Does This Mean for Future Care?

Low enrollment has long been a problem in trials. One in five cancer trials linked to the National Cancer Institute fails for this reason. Missed enrollment targets delay drug approvals and reduce chances of treatment success.

Lead author Dr. Mike Dohopolski said that large language models like GPT-4 are especially helpful when trial rules are more flexible. In strict cases, AI still struggles, but it can improve how fast and effectively trials operate.

What Else Are the Researchers Working On?

The same team is exploring AI in radiation therapy. Their system, GeoDL, can adjust treatment doses in real time. It uses CT scans and treatment data to estimate 3D radiation doses in just 35 milliseconds. This could allow doctors to personalise therapy during surgery.

This study is among the first published in the Machine Learning series by IOP Publishing. The journal is the first open-access series focused solely on AI and machine learning in science.

With tools like ChatGPT and GeoDL, researchers hope to ease pressure on clinical staff and bring new therapies to patients faster.