Can AI Help Find the Right Patients for Clinical Trials?

Identifying patients who may benefit from a clinical trial can require physicians to piece together information scattered across electronic medical records, including biomarkers, prior treatments, disease characteristics, and clinical notes. Shaalan Beg and Eron Kelly of ConcertAI sat down with BioPharm International for a four-part interview to discuss how artificial intelligence could help bring those data points together while also giving physicians a way to examine treatment patterns among real-world patient cohorts.

How can AI help physicians identify eligible patients and inform treatment decisions?

For oncology practices, finding the right patients for a clinical trial can involve more than searching structured fields in an electronic medical record. Physicians may need to determine a patient's biomarker status, previous lines of therapy, and other clinical characteristics, much of which can be contained in unstructured clinical notes.

According to Shaalan Beg, AI tools can help extract those details and synthesize them into a more comprehensive view of the patient.

“It's only the development in technology because of these LLMs and SLMs that we've developed internally that are allowing us to extract that information and tell the physician at the point of care that this is a person who meets these criteria,” Beg said.

Using a HER2 example, Beg described how the technology can identify patients who have a particular biomarker status, prior treatment history, and other characteristics that could make them eligible for a specific therapy or clinical trial. The approach is designed to look beyond structured medical record fields and extract relevant information from clinical notes.

The technology can also help surface potential care gaps. By synthesizing information on biomarkers, treatment history, and other patient characteristics, the system can identify areas where additional testing or treatment options may need to be considered.

A second application involves patients for whom clinical trial evidence may not provide a clear treatment path. Beg noted that real-world patients do not always resemble the populations enrolled in clinical trials, leaving physicians with situations where there may be limited level-one evidence to guide treatment decisions.

For these patients, Beg described a cohort matching tool that allows physicians within the CancerLinQ network to identify other patients with similar disease and treatment characteristics. Physicians can then examine which treatments those patients received and how they fared.

The approach can provide another layer of decision support, including information on whether treatments used in comparable patients were categorized as NCCN category 1 recommendations.

Kelly described the technology underpinning these applications as a digital representation of a patient's characteristics. That same representation can be used both to identify comparable patient cohorts and to evaluate whether a patient meets the inclusion and exclusion criteria for a clinical trial.

“The underpinning there is we're effectively creating a digital object that represents that patient's characteristics,” Kelly said.

The same technology can also connect clinical trial sponsors and sites. According to Kelly, using a common technology platform across both sides could make communication and collaboration more efficient, while helping sponsors identify patients who are appropriate for their trials.

For clinical trials, that alignment could serve two purposes, helping patients access potentially appropriate care while also helping sponsors enroll the patients needed to generate meaningful trial evidence.

Watch part one of this four-part interview here.