How Agentic AI Aims to Streamline Clinical Trial Enrollment

Artificial intelligence (AI) is becoming increasingly integrated into oncology care, with applications ranging from administrative tasks to clinical decision support.¹ A newer evolution of this technology, agentic AI, is designed to go beyond simply generating responses or retrieving information; these systems can use AI to work through complex, task-specific problems, drawing on multiple sources of information and helping users move through a series of decisions or actions. In oncology, where treatment decisions often depend on a combination of clinical history, molecular findings, guidelines, and other patient-specific factors, these capabilities could have meaningful applications across both clinical care and research.

In an interview with Targeted Oncology^TM, Shaalan Beg, MD, MBA, FASCO, chief medical officer at ConcertAI and gastrointestinal medical oncologist, highlighted how agentic AI can reduce cognitive and administrative burden for oncology teams, improve the efficiency of clinical trial screening, and potentially expand access to trials for patients receiving care in community settings.

Targeted Oncology: Could you explain the premise of agentic AI and how it is being applied in oncology practice to inform treatment decisions?

Shaalan Beg, MD, MBA, FASCO: I think about agentic AI as tools that are developed to solve specific problems, using AI [along with] the totality of evidence that we have based on data which are available to the folks that have developed it, additional context and information related to the question that we're looking to solve from the patient, and other general data which are available outside of both of these contexts.

What we've seen over the evolution of the last few years as AI tools continue to develop is that general-purpose AI doesn't always understand the nuances of a task. If we look at the oncology use case specifically, there are elements of applications of AI—like interpreting molecular profiles, understanding what line of therapy a patient is on, interpreting oncology guidelines in the context of a specific patient—that require training of tools using more data than is otherwise available. I think about these agents as specialized tools that are there to fix specific problems that are necessary to solve along the continuum of different challenges that we face as we're addressing the care of a patient, either in clinical care or research.

How is agentic AI facilitating the matching of patients to clinical trials? How can it integrate into existing oncology workflows without adding another layer of administrative burden?

When I think about administrative burden, I think about cognitive load; I think about the number of decision points that you introduce into, let's say, a clinician's workflow to get a task to [completion]. If we specifically look at [our] CancerLinQ tool [at ConcertAI], these tools are powered by our proprietary data sources and AI tools to make sense of the available data. We synthesize data from electronic medical records [EMRs], both structured and unstructured elements like our notes, pathology reports and radiology notes. Those data are supplemented by social determinants of health data, claims data, and those are the tools that we use to develop elements and to refine our agents…in order to determine molecular characterization of a patient, its association with national oncology guidelines and FDA-approved medications, and clinical trials that are currently developing that that patient may be eligible for.

If we take the clinical trial matching use case that we have available for our CancerLinQ member sites, in order to find the right patient, it's not enough to say that this person has cancer X with this biomarker status. A patient may not be eligible for a clinical trial until their cancer reaches a specific state. A lot of clinical trials are designed as a first line of therapy; others are designed as a third line of therapy, and the tools we have developed based on the data that we have access to are designed to identify when a patient's cancer journey meets the point that a clinical trial is interested in as well. In that instance, it would be to the line of therapy and how can you extract prior treatments that a patient has had from the EMR and the data available to say, “Hey, this is a [patient] who is going to be eligible for the clinical trial today,” as opposed to, “Keep an eye on it, they may become eligible in 6 months or 2 years; we can’t tell the difference,” but the tools that we have can help narrow down that element.

Coming back to the cognitive load question, if you're thinking about a coordinator who is screening patients for clinical trials, we provide them with the information they need in the sequence that they need it. We help prioritize the list of patients who are more likely to be eligible for clinical trials. We're not looking to remove all of the prescreening activities and the screening activities that a coordinator does, but we are focused heavily on making all that information available to them in the sequence that they need it.

On the clinician side, the tools [can flag and say], “Hey, there's a care gap in this patient. You haven't tested for this biomarker. We haven't identified this biomarker in this record.” So instead of the physician having to, first of all, remember the biomarkers which are relevant for that disease, seeing whether that's been done [and] whether that's positive or not, we do that work in the background and surface that this [information] is missing, and then the clinician has the option of acting on or dismissing that alert for whatever reason. So, by making the data available to them at the right time, we're reducing the cognitive load that’s required to go look for the data. Providing it at the right time for the right patient…improves the efficiency of the clinic and allows them to surface the care gaps that may exist in a patient's chart. Or if they're looking to compare their patient's journey with that of thousands of other patients who have had similar journeys in the past, we provide those views to them as well, so we can help guide the care that they have.

What does human oversight look like when a potential trial match is identified by AI, and how should oncologists think about validating or acting on such a result?

The short answer is that [agentic AI] makes the humans’ actions a lot more efficient. It's collecting, synthesizing, and making the data available to the human abstractor in a much more uniform way. So, it's not taking away the need that a human will have to look at those records and talk to a patient, explain the study, and have to get a signature on the trial, but it reduces the screening activity from 1 hour or 90 minutes down to a few minutes by making all that information available to them.

Imagine a new patient who comes into the clinic; they may have been to a few different hospitals in the past. They have medical records that are written in the doctor's progress note. There could be information on an outside doctor's progress note, which is scanned under a media tab… Molecular reports aren't available; you're waiting for that fax to come in. Collecting that information and making that entire packet for a coordinator takes a lot of time, and things get missed or misinterpreted. The tools that we've developed are there to review that information and rank the patients based on their likelihood of matching, so that the actual human activities that take place are much more efficient and they can move forward.

Right now, for oncology clinical trials, I'm not seeing these solutions as being completely automated, where all of the eligibility determinations could take place using AI, and AI also executes a consent without human oversight. I do imagine that we may get there for some clinical trials that are looking for outcomes for certain studies; more pragmatic clinical trial designs are employing some of these tools. But for most oncology use cases, the human oversight is very appropriately going to continue being there, but we can improve the efficiency and the number of patients who are able to go on a study by making that experience of the clinicians and the coordinators much more practical.

Do you think such tools could help address existing disparities in clinical trial enrollment, particularly for patients treated outside of academic centers?

One hundred percent. Even right now, if we look at the academic medical centers, a lot of them have outlined community facilities. A lot of those programs are figuring out how to staff those centers with the appropriate clinical research staff and how to have clinical trial investigators participate in those centers so they could be screening and evaluating patients as well. There are many factors that go into whether a patient who's seeing a physician goes on a clinical trial or not. Certainly, there are factors like patient [interest] and [eligibility], but there are also clinician factors: Is it easy for them to refer to the clinical trial? Do they have the information available for the trial? And then institutional factors: Does that center have the ability to execute the clinical trial or not? By delivering these tools, we are able to extend the reach of the main research sites to oversee activity even at satellite centers so we know if there's an eligible patient who’s showing up at the satellite sites, and those patients can be screened at the same time as a patient who's coming to the main center.

When we talk about disparities in care, we're talking about many different aspects; we're talking about race and ethnicity, but we're also talking about insurance status, their ZIP code. We know one of the biggest barriers for enrolling in a clinical trial is their distance from a clinical trial site. And using the tools that we have, we're able to screen and evaluate patients who are being seen closer to home with their community clinicians.

Looking ahead, what is needed for these AI tools to scale up and become a routine part of clinical practice, rather than at select sites?

Success begets success. I think as a lot of the stories that we're talking about come out and they're known to the public, more people will embrace these tools. The trust in these tools will increase as well, and the tools will continue to get better as we get feedback. One of the challenges that we have in these tools is simply implementation at a site. You can have clinicians and research staff who really want a tool, but the process of implementing into a clinical site can take time, and a lot of health centers are still trying to define what their governance looks like for informatics, AI, and agentic AI tools and what their risk looks like. We're seeing, in some instances, [changing] state regulations that can impact how these tools are deployed. And then there are financial considerations [of] how a site or a tool defines its value in terms of the work and energy that goes into implementing and running that. As those value propositions become more clear, we're going to see more deployment of these tools, and it's not just developing the tool, but it's also maintaining the tool and making sure that it continues to improve.

I'm strongly of the opinion that tools that can give multiple solutions to a site with a single implementation process are the ones that are going to win. Many years ago, when sites were deploying elements like Microsoft Office, the fact that Office would come with Word, Excel, and PowerPoint [made] the experience easier because there's one deployment, one subscription, one learning curve with those tools, and you can move forward. And I'm seeing large-, medium-, and small-sized health systems look for solutions that are able to provide a battery of solutions to them, as opposed to having to pick individual solutions as they come. On the ConcertAI and CancerLinQ side, we have solutions that are supporting not just clinical care but also clinical research care. In oncology, that's very important, because clinical care and research happen in parallel. Unlike other diseases, where you may be seeing a separate research and clinical physicians, in oncology, most of the times your clinical investigator is also managing your cancer. The solutions that we've talked about are supporting both of those because we know where our patients and clinicians live: [at] the intersection of clinical care and clinical research.

REFERENCE

1. Li J, Zhang L, Yu Z, Bao Z, Li D, Wang L. The impact of AI on modern oncology from early detection to personalized cancer treatment. npj Precis Oncol. 2026;10(1). doi:10.1038/s41698-026-01276-6