Arturo Loaiza-Bonilla, MD, MSEd, and Shaalan H. Beg, MD
Dr. Arturo Loaiza-Bonilla
Key Points:
-
- Multiple studies presented at the 2025 ASCO Annual Meeting examined how artificial intelligence (AI) can power the patient–clinical trial matching process.
- In one study, multiagent, oncology-tuned AI reduced manual matching from months to minutes while sustaining an F1 score > 0.80; in another randomized study, joint human and AI prescreening showed improvement over coordinators or algorithms alone, raising chart-level accuracy by 5%.
- Future developments in this space demand curated data pipelines, continual drift monitoring, and deliberate upskilling so that every clinic visit automatically queries the trial universe.
Why Trial Matching Still Breaks Down
Only 2% to 8% of U.S. adults with cancer enroll in therapeutic trials, partly because eligibility review is a manual detective exercise buried in unstructured records.1 Coordinators must juggle changing protocols, incomplete staging data, and fast-evolving biomarkers—an error-prone workflow that disadvantages community practices and under-represented populations. Multiple studies presented at the 2025 ASCO Annual Meeting addressed these challenges and offered potential AI-powered solutions (Table 1).
Table 1. Key Results From ASCO25 Abstracts
Abbreviations: AI, artificial intelligence; LLM, large language model; pts, patients.View larger.
What Was New at ASCO25
Multiagent AI plus oncology knowledge graph
Our prospective evaluation of a 3-agent large language model stack linked to an oncology-specific knowledge graph processed 157,367 clinical pages for 3,804 patients with metastatic cancer in 2024.2 The system surfaced matches that oncologists could sign off on in approximately 1 hour instead of the thousands of hours a purely manual process would have required. The key metrics were a sensitivity of 0.84, a specificity of 0.84, and F1 0.82—well above zero-shot GPT-4o baselines.
Randomly assigned human plus AI teaming trial
Parikh and colleagues randomly assigned 2 research coordinators to review 356 lung and colorectal charts, either solo or with an AI that highlighted eligibility cues.3 Human plus AI review was noninferior and then superior (accuracy 76.1% vs 71.5%; P < .001). The hybrid arm outperformed in the hardest tasks—T, N, and M staging—without slowing review time (median 32 minutes per chart).
Dr. Shaalan H. Beg
Educating the workforce for an AI-enabled future
In May 2025, we outlined a blueprint for multidisciplinary AI adoption in the ASCO Educational Book.4 That framework—emphasizing transparent model auditing, equity safeguards, and workforce reskilling—underpins both of the previously mentioned abstracts. The article, “Driving Knowledge to Action: Building a Better Future With Artificial Intelligence–Enabled Multidisciplinary Oncology,” outlines how integrating AI across cancer care, from early detection and imaging to treatment decision support and follow-up, can enhance diagnostic precision and personalized therapy, leveraging rich, multimodal data sources (imaging, genomics, clinical, and real-world). Importantly, we emphasize transparent model governance, ensuring developers provide clear documentation on data provenance, performance metrics, bias detection, and continuous validation so clinicians and regulators can scrutinize and trust AI tools. To prevent deepening disparities, the blueprint recommends diverse training datasets, subgroup performance audits, and bias-mitigation strategies tailored especially to low- and middle-income settings. Recognizing that AI is not “plug-and-play,” the work calls for coordinated workforce reskilling, engaging clinicians, nurses, pathologists, informaticians, and engineers through targeted education in AI interpretation, validation, and oversight. Finally, the article urges establishment of governance and ethics frameworks—including data privacy, informed consent, accountability pathways, and regulatory alignment—to ensure AI deployment is transparent, equitable, and sustainable worldwide.4
Beyond Accuracy: 4 Practical Imperatives
In this rapidly evolving space, 4 practical necessities guide human-centered AI trial matching:
-
-
- 1. External Generalizability: Models trained at quaternary centers must perform on noisy community electronic health records. Federated learning shares weights, not data, protecting privacy while stress-testing robustness.
- 2. Drift and Bias Audits: Eligibility language and staging rules evolve. Automated dashboards should trigger reannotation when performance slips below preset thresholds.
- 3. Explainability: Exposing the rule path (“EGFR exon 20 insertion found; ECOG 0; prior osimertinib”) builds clinician trust and reveals why a patient was not matched.
- 4. Embedded Workflows: Fast health care interoperability resource–native microservices that deliver eligibility snippets directly into the electronic health record inbox beat standalone portals every time.
-
Furthermore, in Table 2, we outline current challenges, next steps, and success metrics for the near future of AI integration in oncology trial matching.
Table 2. Human Plus AI Trial Matching Playbook for 2025 to 2027
Abbreviations: AI, artificial intelligence; DUA, Data Use Agreement; FDA, U.S. Food and Drug Administration; IND, Investigational New Drug Application, pp, Percentage Points.
Controversies and Counterpoints
Critics worry about hallucinations or coordinator deskilling; yet, hybrid workflows catch outliers while freeing staff to counsel patients—a value-add rarely captured by time-and-motion studies. Early modeling suggests at least a 10-fold return on investment when reduced screen failure rates and faster accrual are included.6-7
Conclusion
Several ASCO25 abstracts showed that human-centered AI can transform trial matching from an artisanal chore into a real-time safety net. If we implement these solutions at the point of care as a federated engine that is contextual to the population in which they are deployed—and make the solutions patient-centric—we can transform the future of trials, activating them in real time at an economically feasible global scale. Our mandate is clear: integrate these tools responsibly, audit them continuously, and train the workforce so that every oncology visit automatically surfaces the best evidence-aligned options, no matter where a patient receives care.
References:
1. Unger JM, Hershman DL, Till C, et al. “When offered to participate”: a systematic review and meta-analysis of patient agreement to participate in cancer clinical trials. J Natl Cancer Inst. 2021;113(3):244-257.
2. Loaiza-Bonilla A, Kurnaz S, Tuysuz E, Huner O, Giritlioglu D, Noel Meza JP. Transforming oncology clinical trial matching through multi-agent AI and an oncology-specific knowledge graph. J Clin Oncol. 2025;43:163 (suppl; abstr 1554).
3. Parikh R, Kolla L, Beothy E, et al. Effect of human-AI teams on oncology prescreening: final analysis of a randomized trial. J Clin Oncol. 2025;43:16s (suppl; abstr 1508).
4. Loaiza-Bonilla A, Thaker N, Chung C, Parikh RB, Stapleton S, Borkowski P. Driving knowledge to action: building a better future with AI-enabled multidisciplinary oncology. ASCO Educ Book. 2025;45(3):e100048.
5. Templin T, Fort S, Padmanabham P, et al. Framework for bias evaluation in large language models in healthcare settings. NPJ Digit Med. 2025;8(1):414.
6. Bharadwaj P, Nicola L, Breau-Brunel M, et al. Unlocking the value: quantifying the return on investment of hospital artificial intelligence. J Am Coll of Radiol. 2024;21(10):1677-1685.
7. Kurnaz S, Loaiza-Bonilla A, Carvallo Castañeda D, Huner O, Giritlioglu D; Precision Cancer Consortium. Effect of a novel artificial intelligence (AI) –enabled multi-trial matching system on patient matching using real-world data. J Clin Oncol. 2024;42:16s (suppl; abstr e13501). https://doi.org/10.1200/JCO.2024.42.16_suppl.e13501
Author Bios
Dr. Arturo Loaiza-Bonilla is the co-founder and chief medical officer at Massive Bio, chief of hematology and oncology at St. Luke’s University Health Network, and associate professor at the Lewis-Katz School of Medicine at Temple University.
Dr. Shaalan H. Beg is adjunct associate professor of hematology/oncology at the University of Texas Southwestern Medical Center and vice president of oncology at Science 37. He is also an associate editor for ASCO Daily News.
Disclaimer: This article is taken from “Joint Human and AI Teams Bring Real-Time Trial Matching to the Point of Care”, originally published at ASCO Daily News: https://dailynews.ascopubs.org/do/joint-human-and-ai-teams-bring-real-time-trial-matching-point-care
