Skip to main content

AI Powers Progress of UPMC Clinical Trial

Analysis  |  By Scott Mace  
   July 13, 2022

The Pittsburgh health system is using technology to extract clinical notes and more quickly qualify participants.

Technology in use at UPMC is accelerating the speed with which the Pittsburgh-based health system can identify and qualify participants in a clinical trial targeting colorectal cancer prevention.

A technology platform that combines natural language processing and artificial intelligence is more efficient than traditional patient identification methods, which are often manual, time-consuming, and costly for clinical trial sites.

"In this case, the trial is about people who have a particular kind of polyp," says Robert Schoen, MD, MPH, chief of the Division of Gastroenterology, Hepatology, and Nutrition at UPMC and a professor of medicine and epidemiology at the University of Pittsburgh. "Then you would go through all those colonoscopy exams and pathology manually, and select them one by one. It's an effective way to do it, but it's very inefficient."

Schoen is principal investigator for FORTE, a study led by NRG Oncology, part of the National Clinical Trials Network, sponsored by the National Cancer Institute.

Robert Schoen, MD, MPH, chief of the Division of Gastroenterology, Hepatology, and Nutrition at UPMC.

"This time saved allows FORTE sites to increase recruitment efforts and continue to build their FORTE referral networks, and also make the overall study more likely to hit its participant recruitment goals," Schoen says.

The primary goal of FORTE is to compare the colorectal cancer rates between the two study groups (repeat colonoscopy at five and 10 years vs. repeat colonoscopy at 10 years) to see if the rates are equivalent. Each study group had one or two small benign polyps removed during a previous colonoscopy.

If those rates are equivalent, the study could prove that a five-year exam isn't necessary, and that clinicians can focus on recommending an exam every 10 years. Participants will be asked to donate blood and stool samples and will be followed annually.

The study organizers expect to enroll 9,500 participants, which requires finding and engaging a sufficient number of eligible participants from sites across the country.

"We want to query the medical record for the six to eight criteria that we know are important, and Pieces helps us do that," Schoen says.

According to a 2015 study, as many as 86% of clinical trials do not meet recruitment goals within their proposed time periods. This often makes clinical trials more expensive and time-consuming, which in turn can affect results and clinical outcomes.

Pieces Predict is the enabling technology in the screening process. Developed by Pieces Inc., a Texas-based AI company, Pieces Predict expedites this trial recruitment process by scanning hundreds of thousands of colonoscopy notes in a secure manner at selected study sites to identify participants who have a high probability of meeting the study's eligibility criteria.

The technology won't eliminate all the steps to validate potential participants.

"We may have to ask the patient some questions that aren't contained in the record, and of course, the patient has to decide if they want to participate in the study," Schoen says. "It's purely voluntary. But just going from 10,000 [prospects] to 1,500 is an enormous time saver."

This kind of technology assistance is not common in federal national studies, so researchers are hoping that FORTE can serve as a potential model for future studies, Schoen says. NRG Oncology alone has hundreds of studies ongoing, enrolling patients in cancer treatment trials, he says.

At this point, Schoen is unable to predict precisely how much faster the technology will be able to find the desired 9,500 patients.

"We know that we have an instrument that does what we want it to do well, and we will let it do its thing," he says.

The FORTE study is also being publicized through social media and other previously used methods of attracting participants, but by zeroing in on patient records at UPMC, study leaders can identify a pool of patients more quickly, Schoen says.

"It would be a very different thing to say I'm going to look through all the EMRs at Aetna Insurance, but we don't have permission for that," he says. "We only have permission if the patient has granted us permission, or if we have an accepted license to review their medical information."

On the plus side, communities of patients gathered by organizations such as the National Clinical Trials Network are also effective ways of finding participants.

"The question is, can we marry the technology to the administrative and legal structure of how you conduct these trials at these different sites?" says Ruben Amarsingham, chief executive officer of Pieces, Inc. "That was the first question Dr. Schoen asked. The more that organizations like ours can do that, I think we can make more progress."

"Individual sites can be large academic centers, or GI practices, that have multiple providers and take care of thousands of patients," Schoen says. "It can manifest in a lot of different ways."

Although the trial will take place over a longer time span, within a year or two, FORTE leaders will complete analyses looking at how well the technology has worked at some sites, in terms of records analyzed and patients identified, Schoen says.

The NCI Division of Cancer Prevention leads the NCI Community Oncology Research Program (NCORP). While NRG Oncology is leading the FORTE study, other network organizations participating include the Alliance for Clinical Trials in Oncology, ECOG-ACRIN Cancer Research Group, and SWOG Cancer Research Network.

“This time saved allows FORTE sites to increase recruitment efforts and continue to build their FORTE referral networks, and also make the overall study more likely to hit its participant recruitment goals.”

Scott Mace is a contributing writer for HealthLeaders.


Natural language processing and AI are culling through identified and consented patient records to speed up the identification process for clinical trials.

The UPMC project aims to compare rates of colorectal cancer in patients who repeat colonoscopies at 5-year and 10-year intervals.

As many as 86% of clinical trials do not meet their recruitment goals using traditional, more manual methods of patient identification.

Get the latest on healthcare leadership in your inbox.