Reading time: 5 minutes
Charles Okila, MPH
Clinical trials provide an avenue for many patients suffering from cancer to obtain access to new therapies and more intensive care. However, only a small percentage of cancer patients participate in these clinical trials. The lack of participation is seldom a matter of unwillingness. Instead, it is the result of challenges such as complex inclusion and exclusion criteria, lack of awareness, geographical distance, and the difficulty of finding the right patient for the right trial at the right time. Recently, artificial intelligence (AI) has been suggested as a solution to these setbacks. It is essential to note that these technologies do not replace doctors; instead, they are designed to search medical histories, understand criteria for trials, and identify potential matches at a faster rate compared to humans. However, it is important to inquire how effective the AI innovation is in cancer clinical trials, and whether it benefits those who have been traditionally excluded.
Why trial matching is so difficult
Cancer trials have a long list of inclusion and exclusion criteria, including diagnosis, cancer stage, previous treatment, lab results, biomarkers, and performance status. This review process is labor-intensive and can lead to patients being overlooked, particularly in busy clinics.1 As a consequence, patients may not be informed about a trial, and many trials fail due to insufficient enrollment.2 These challenges are particularly common in rural and community hospitals, where research infrastructure and staff support are limited.⁷ Cancer trials also struggle to enroll enough patients from racial and ethnic minority groups, impacting equity and the ability to apply findings across diverse populations.6,8
What AI-based trial matching actually does
The majority of AI trial matching systems apply natural language processing and machine learning algorithms to read unstructured notes and structured data in electronic health records. The system then matches patient data to the eligibility criteria of a trial and produces a list of potential matches ranked as fit.5 Contemporary AI systems are designed to cope with uncertainty, synonyms, and missing information, rather than relying on simple rule-based matching. Some newer systems also strive to be “transparent”, in that they show clinicians how and why a patient was matched or excluded, instead of providing just a black or white answer.1
What the evidence shows so far
Several studies indicate that AI-based matching performs at least as well as manual matching. A systematic review and meta-analysis of AI tools for matching patients to clinical trials demonstrated high sensitivity and specificity across a number of datasets in oncology, with AI-based systems typically faster and less reliant on human resources to identify eligible patients.2 These results are supported by real-world validation studies. A retrospective analysis of patients with hepatocellular carcinoma in China found that an AI-based matching system was highly accurate and took a fraction of the time of manual screening.10 More importantly, the system was particularly adept at excluding ineligible patients, thus allowing clinicians to focus on the more complex patients. Early-phase oncology trials may benefit as well. A proof-of-concept study with explainable AI showed that automated matching for phase I trials could be reasonably accurate, while also providing explanations for its decisions.1 While such systems are still experimental, they illustrate how to speed up one of the costliest phases of cancer research.
Beyond matching: eligibility and access
AI systems are also being tried for other tasks, such as eligibility for cancer screening, or to improve risk-based strategies for trial recruitment.3 These are not specific to trials, but demonstrate how algorithm-based decision support could personalize access to prevention and research. However, technology alone cannot fix structural barriers. Research consistently finds lower trial participation among people in rural communities, where access to transportation and special care is limited.7 There are also significant disparities in trial participation by race and ethnicity, despite decades of discussion and advocacy.6,8 AI might help to identify eligible patients, but the patient cannot enroll without the trial being offered, properly explained, and designed to facilitate participation.
Equality: challenges and opportunities
A common promise of AI-based trial matching is that it will decrease inequity by equalizing the assessment of eligibility for trials. In principle, an algorithm will not “overlook” patients due to time constraints or unconscious bias. But in reality, AI-based trial matching may perpetuate disparities. If previous trials were under-enrolled, data used to train matching tools will reflect this.4,9 Without careful control, AI may inadvertently favor those who already have greater access to health care. Professional bodies make it clear that enhancing diversity in cancer trials needs careful design, engagement and accountability – not just new technology.6 AI can help achieve these outcomes, but it cannot do them alone.
What patients should know
For patients, trial matching with AI will typically occur behind the scenes. It may be embedded in a hospital’s electronic health record (EHR) system or used as a tool by researchers. An algorithm will not be directly used by the patient.
But patients can inquire:
• How are clinical trials identified here?
• Is screening for eligibility done electronically?
• If there is no trial, may I ask if I would be eligible for one elsewhere?
Taking the initiative can help make sure technology will complement patient-focused care.
Looking ahead
AI will not solve the problem of clinical trial recruitment, but we know it can help reduce the time and effort involved in patient matching and identify previously unknown opportunities. The next steps will be to investigate the potential use of AI in a range of clinical settings, the transparency of algorithms and the effects on equity and outcomes. In the meantime, AI offers the prospect of a useful aid, one that might help more patients receive the message, “There is a trial for you,” rather than the question, “Is there a trial for me?”
Header Image Source: from Scienmag (image is under a Creative Commons CC0 1.0 license)
Edited by Jessica Desamero, PhD
References
1. Ghosh S, Abushukair HM, Ganesan A, Pan C, Naqash AR, Lu K. Harnessing explainable artificial intelligence for patient-to-clinical-trial matching: A proof-of-concept pilot study using phase I oncology trials. PLoS One. 2024 Oct 24;19(10):e0311510. https://doi.org/10.1371/journal.pone.0311510
2. Chow R, Midroni J, Kaur J, Boldt G, Liu G, Eng L, Liu FF, Haibe-Kains B, Lock M, Raman S. Use of artificial intelligence for cancer clinical trial enrollment: a systematic review and meta-analysis. JNCI: Journal of the National Cancer Institute. 2023 Apr 1;115(4):365-74. https://doi.org/10.1093/jnci/djad013
3. Callender T, Imrie F, Cebere B, Pashayan N, Navani N, Van der Schaar M, Janes SM. Assessing eligibility for lung cancer screening using parsimonious ensemble machine learning models: A development and validation study. PLoS Medicine. 2023 Oct 3;20(10):e1004287. https://doi.org/10.1371/journal.pmed.1004287
4. Askin S, Burkhalter D, Calado G, El Dakrouni S. Artificial intelligence applied to clinical trials: opportunities and challenges. Health and technology. 2023 Mar;13(2):203-13. https://doi.org/10.1007/s12553-023-00738-2
5. Lee K, Mai Y, Liu Z, Raja K, Jun T, Ma M, Wang T, Ai L, Calay E, Oh W, Schadt E. CriteriaMapper: establishing the automatic identification of clinical trial cohorts from electronic health records by matching normalized eligibility criteria and patient clinical characteristics. Scientific Reports. 2024 Oct 25;14(1):25387. https://doi.org/10.1038/s41598-024-77447-x
6. Oyer RA, Hurley P, Boehmer L, Bruinooge SS, Levit K, Barrett N, Benson A, Bernick LA, Byatt L, Charlot M, Crews J. Increasing racial and ethnic diversity in cancer clinical trials: an American Society of Clinical Oncology and Association of Community Cancer Centers joint research statement. Journal of Clinical Oncology. 2022 Jul 1;40(19):2163-71. http://ascopubs.org/doi/full/10.1200/JCO.22.00754
7. Bhatia S, Landier W, Paskett ED, Peters KB, Merrill JK, Phillips J, Osarogiagbon RU. Rural–urban disparities in cancer outcomes: opportunities for future research. JNCI: Journal of the National Cancer Institute. 2022 Jul 1;114(7):940-52. https://doi.org/10.1093/jnci/djac030
8. Monge C, Greten TF. Underrepresentation of Hispanics in clinical trials for liver cancer in the United States over the past 20 years. Cancer Medicine. 2024 Jan;13(1):e6814. https://doi.org/10.1002/cam4.6814
9. Saeed H, El Naqa I. Artificial intelligence in clinical trials. In Machine and deep learning in oncology, medical physics and radiology 2022 Feb 2 (pp. 453-501). Cham: Springer International Publishing. https://www.ijsrtjournal.com/article/Artificial+Intelligence+in+Clinical+Trials#
10. Wang K, Cui H, Zhu Y, Hu X, Hong C, Guo Y, An L, Zhang Q, Liu L. Evaluation of an artificial intelligence-based clinical trial matching system in Chinese patients with hepatocellular carcinoma: a retrospective study. BMC Cancer. 2024 Feb 22;24(1):246. https://doi.org/10.1186/s12885-024-11959-7
OncoBites 
Leave a comment