As artificial intelligence (AI) technologies advance rapidly across healthcare domains, oncology researchers are discovering transformative applications for optimizing clinical trial design and execution. According to Sandip P. Patel, MD, a medical oncologist and professor in the Department of Medicine at the University of California San Diego Health, AI-powered solutions are already reshaping how cancer clinical trials operate.
“AI is here in cancer care, and one of the areas of interest [for AI] is in clinical trials,” Patel explained in an interview with OncLive®. “Clinical trials [involve] multiple processes related to the screening of patients, radiology measurements for tumor sizes, [and collecting] clinical data. [We are] thinking about AI tools that can help us in those venues.”
Revolutionizing Clinical Trial Design Through Machine Learning
The integration of AI and machine learning into oncology research represents a paradigm shift in how investigators approach trial architecture and patient selection. During the 2025 ASCO Annual Meeting, Patel presented groundbreaking research on AI innovations aimed at improving clinical trial design and enhancing patient representativeness in oncology studies.
His presentation outlined several promising applications of AI technology in clinical trial execution, including AI-assisted lung cancer screening, COVID-19-related pneumonia diagnosis, TROP2 normalized membrane ratio assessment, and sophisticated clinical trial matching algorithms. These innovations address longstanding challenges in oncology research, from initial patient identification through final data collection.
Large Language Models Transform Patient-Trial Matching
One of the most significant advances involves large language models (LLMs) that streamline the complex process of matching patients with appropriate clinical trials. A comprehensive review published in Current Opinion in Urology examined the current capabilities of LLMs in clinical trial matching, revealing impressive performance metrics.
The review highlighted TrialGPT, which leverages ChatGPT-3.5, ChatGPT-4.0, and MedCPT technologies. This sophisticated system successfully retrieved 90% of relevant trials from just a 6% initial collection, demonstrating remarkable efficiency. The matching accuracy achieved a rating of 0.873, while ChatGPT-4.0 alone displayed an impressive 88% agreement with initial physician assessments.
These automated clinical trial matching systems offer substantial benefits for both patients and healthcare providers. Patients gain improved accessibility to potentially life-saving trials without complete reliance on their providers’ knowledge of ongoing studies. Meanwhile, oncologists experience reduced administrative burden, as they no longer need to maintain detailed tracking of complex eligibility criteria across multiple trials.
“There are many friction points in clinical trials, starting at accrual,” Patel noted. “[We must] find patients who meet multiple eligibility criteria, sometimes including specific genomic or molecular aberrations. It’s important for us to have tools that can help screen patients for these clinical trials. Clinical trial matching is one of the AI tools that we see utilized.”
Acknowledging AI Limitations and Hallucination Risks
Despite promising advances, AI-based LLMs in oncology remain imperfect tools requiring careful human oversight. Patel emphasized that current systems still produce concerning errors that could impact patient care if left unchecked.
Research published in JAMA Oncology revealed significant limitations when an LLM provided treatment recommendations for stage I breast cancer patients. While the system offered at least one treatment option concordant with National Comprehensive Cancer Network (NCCN) guidelines for each case, 34.3% of outputs (35 of 102) also recommended treatments that contradicted NCCN guidelines. Even more concerning, the LLM produced hallucinated responses—recommendations not part of any established treatment protocol—in 12.5% of outputs (13 of 104).
Streamlining Data Collection and Management
Beyond patient matching, AI technologies offer substantial potential for improving clinical trial data management. Throughout study duration, researchers collect extensive clinical data including laboratory values, adverse effects documentation, and imaging assessments. AI-assisted transcription and electronic data capture systems can significantly reduce manual data entry burdens while improving accuracy.
“Having an AI assist with transcribing or manually editing electronic data capture systems, as opposed to a human, is another area [of opportunity],” Patel explained. “Increasingly we’re going to see [attempts to use] AI to address these friction points head-on, because they’re a barrier to us understanding how we can best help our patients in clinic, even though these are back-end systems.”
Enhancing Patient Diversity and Representation
Racial and ethnic diversity in clinical trials remains a persistent challenge in oncology research. Machine learning models present promising solutions for addressing recruitment biases by optimizing cohort selection and refining inclusion criteria. These advanced algorithms can analyze pathological data across large multiracial patient cohorts, facilitating improved patient stratification in biomarker-based cancer trials.
However, Patel cautioned that AI deployment must proceed carefully to avoid exacerbating existing healthcare disparities. “This is an area where we have to be very careful with how we utilize AI [and] not create a digital divide amongst the geographic and demographic divides that often exist in receiving optimal cancer care,” he warned.
Preventing Algorithmic Bias in Clinical Research
The data sources used to train AI algorithms—primarily clinical trials, electronic health records, and administrative claims—may contain inherent biases that could be amplified without proper oversight. Training models on specific populations may produce clinical predictors that lack validity for different demographic groups with distinct biological characteristics.
Addressing these challenges requires rigorous human supervision and multi-stakeholder collaboration to identify and eliminate known sources of machine learning bias, including evidence/research bias, expertise/provider bias, and exclusion/embedded data bias.
“As long as we’re cognizant about how we utilize these tools, AI can help to [alleviate] some of the problems we have with the digital divide, but we have to be careful and conscious about deploying AI in a way that aims to reduce disparities, not increase them,” Patel concluded.
