A landmark systematic review and meta-analysis confirms what the healthtech world has long anticipated: artificial intelligence (AI) is not just promising but already effective in optimizing patient recruitment for cancer clinical trials.

In an analysis of over 50,000 patient records across 19 datasets, AI algorithms consistently outperformed or matched manual screening methods in identifying eligible patients. With summary sensitivity at 90.5% and summary specificity at 99.3%, the evidence points toward an AI-augmented future in oncology trial design, recruitment, and execution.

Closing the Enrollment Gap

Despite being central to medical innovation, clinical trials suffer from low participation, only about 5% of adults with cancer enroll in trials, and nearly 20% of studies are terminated early due to insufficient accrual. This leads to underpowered results, delayed drug approvals, and wasted resources.

The study demonstrates that AI, when applied to structured and unstructured clinical data, including natural language from patient charts, can dramatically improve trial matching accuracy, speed, and efficiency. Several included studies reported time savings of up to 90%, with one AI system screening in hours what took humans weeks.

Technical Insights: Structured + Unstructured Data, NLP, and Hybrid Models

The reviewed studies utilized a range of AI techniques from rule-based systems and support vector machines to random forests and commercial natural language processing (NLP) pipelines. All models leveraged both structured electronic health record data (like labs, diagnoses, age, staging) and unstructured sources (e.g., physician notes, pathology reports).

A unifying trend across the tools was the use of NLP to interpret free-text clinical notes, a significant advancement that allows AI to parse nuanced data usually locked in narrative form. In one study, incorporating NLP reduced the pool of non-eligible patients by over 85%.

Beyond Speed: Accuracy and Generalizability

Among the 19 datasets analyzed:

• Accuracy exceeded 80% in all but one dataset
• Sensitivity exceeded 80% in 16 out of 17 datasets
• Specificity exceeded 80% in 15 out of 16 datasets
• Negative Predictive Value (NPV) exceeded 80% in all datasets

This means AI can not only find eligible candidates quickly but safely rule out ineligible patients, potentially minimizing human error and reducing false exclusions.

Interestingly, commercial systems (e.g., IBM’s Watson for Clinical Trial Matching) generally reported higher and more consistent predictive performance than in-house models, suggesting value in deployment-ready solutions while still advocating for open-source innovation.

Supporting Diversity and Global Equity

Clinical trials have historically failed to reflect the full spectrum of patients affected by cancer. Women, older adults, racial minorities, and individuals from underrepresented geographies continue to be excluded from enrollment pipelines. This lack of diversity limits the generalizability of trial results and downstream AI tools, reinforcing a cycle of inequity.

AI offers a path forward. By assessing clinical trial eligibility based on data-driven criteria rather than subjective, manual screening, AI systems may mitigate human bias and promote fairer inclusion. The reviewed studies show promise in identifying patients who were missed by traditional screening methods, with models accurately parsing structured and unstructured health data across large populations.

This connects directly to the Remaining84 challenge: if only a small fraction of real-world patients make it into trials and even fewer into AI development, then any tool that expands equitable access upstream is critical. While these AI systems must be validated further in diverse populations and low-resource settings, they represent a scalable step toward inclusive research and personalized oncology.

Conclusion

This meta-analysis provides robust validation for the integration of AI into the clinical trial workflow, especially in oncology. AI tools can reduce screening burdens, lower trial costs, accelerate timelines, and support diversity and inclusion. However, broader deployment will depend on transparent development, regulatory alignment, and access equity.

As the clinical research ecosystem evolves, AI is not just a tool but it’s a catalyst for smarter, faster, and fairer clinical trials.

References

Chow R, Midroni J, Kaur J, et al. Use of artificial intelligence for cancer clinical trial enrollment: a systematic review and meta-analysis. J Natl Cancer Inst. 2023;115(4):365–374. https://doi.org/10.1093/jnci/djad013