[Rido/Adobe Stock]
Focusing on relapsed or refractory multiple myeloma, the study pitted an AI model designed to identify potentially eligible patients for a phase 3 therapeutic trial against a traditional manual approach. The AI-based method ultimately tripled the number of patients screened with similar outcomes as compared to traditional screening approaches. The company expects further AI-based breakthroughs in identifying and predicting patient progression.
The need for such innovations is only growing. Clinical research programs are grappling with staffing shortages that were exacerbated in the wake of the COVID-19 pandemic. Meanwhile, the number of Americans diagnosed with cancer continues to grow. “It’s not like we’re replacing the humans,” ConcertAI CEO Jeff Elton said. “They can’t find the humans to hire,” he said, referring to the staffing shortage often plaguing clinical research programs.
Jeff Elton [Image courtesy of ConcertAI]
From chaos comes order
The challenge lies in unlocking the insights buried within a mountain of unstructured data. “Most of the features that you’re looking for in clinical trials are sitting in unstructured data,” he explains. “That unstructured data is in the notes, it’s in documents, and all sorts of other aspects that you need to fill inclusion/exclusion criteria.”
While structured data is sometimes available in the form of genetic targets or specific biomarkers, they aren’t necessarily easier to work with. While generally faster to process than unstructured data, automated methods to interpret them can lead to “so many false positives over to the team that it’s like doing manual review,” Elton said.
When inefficiency is a barrier to quality of care
In cancer care, time lags stemming from such inefficiencies can have significant consequences. Patients who might have qualified for a life-extending trial could be left behind. The goal then is to not only drive clinical trial efficiency but to improve patient outcomes and accessibility to experimental therapies. “We’re processing the unstructured data, we’re then using models to…find patients based on 60–70 different criteria,” Elton said. The aim is to generate insights that are as reliable as those of a human expert reviewing the same data.
Connecting the dots in cancer data
One core component in the push to accelerate oncology research is ConcertAI’s CARA AI platform, a multi-modal data management, predictive AI and generative AI platform. It supports a variety of data types, including radiological imaging studies, digital pathology and real-world data (RWD) in both structured and unstructured formats.
By combining diverse data types, the platform can identify patterns and connections that are often missed, such as genetic details of a tumor or nuances in a physician’s notes. The AI surfaces insights that might otherwise go unnoticed.
To handle this data at scale, ConcertAI has partnered with NVIDIA, a collaboration highlighted ASCO 2024. ConcertAI’s collaboration with NVIDIA is integrated into their CARA AI platform as well as in TriaLinQ, which is designed to improve clinical trial screening and the TeraRecon Oncology Suite, which equips oncologists with advanced diagnostic capabilities, including AI tools for lung cancer detection, prostate cancer identification and brain tumor characterization.
Part of the aim is to develop a deeper understanding of biological mechanisms. “We’re really looking at cancer as a disease that has a whole bunch of underlying mechanisms,” Elton explained. “We’re trying to predict a system, because your body is a system.”
Collaborative in spirit
The relationship with NVIDIA is collaborative in nature. The GPU giant will reach out, saying “‘We just opened up a sandbox where you could pull [pre-release tech] down to try integrating that into your workflows, up into your CARA AI stack,’” Elton said. “We get it integrated in 45 minutes on our Slack channel. Three hours later, we’re providing feedback on things,” he added. “There is a speed and dynamism of going back and forth that we’ve rarely seen between two organizations walking through things.”
Toward smarter cancer trials
While cancer outcomes have notably improved in past decades, future gains could come from developing a holistic view of how it operates in all of its guises. “We’re really looking at cancer as a disease that has a whole bunch of underlying mechanisms,” explains Jeff Elton, CEO of ConcertAI. “We’re trying to predict a system, because your body is a system.”
The ability of AI to speed the matching of patients in cancer trials could point the way to further gains in cancer outcomes. Elton paints a picture of AI that goes beyond surface-level data: “If you’re trying to predict certain aspects, like your prediction whether a particular sub-cohort is going to respond positively. Will they maintain their response?” Or imagine if a patient develops resistance to a particular therapeutic approach. “Can we simulate what we think the outcomes might be of that study, change the design, prioritize the one to take into first-in-human that has a higher likelihood of success?” Elton asks.
Filed Under: clinical trials, Drug Discovery, machine learning and AI, Oncology
