Title: “A unified framework for pre-screening and screening tools in oncology clinical trials”
Authors: Denis Horgan, Joseph N. Paulson, Arturo Loaiza-Bonilla, Christer Svedman, Umberto Malapelle, Frédérique Penault-Llorca, Hadi Mohamad Abu Rahsheed, Paul Hofman, Stan Kachnowski, Daniel Schneider, and Vivek Subbiah
Background
Clinical trials remain central to progress in oncology, offering patients access to investigational therapies while generating the evidence needed to improve cancer care. Yet, trial enrollment continues to be one of the most persistent challenges in cancer research. As oncology becomes more biomarker-driven, eligibility criteria have become increasingly complex, and the process of identifying suitable patients has become slower, more fragmented, and more resource-intensive.
The review, “A unified framework for pre-screening and screening tools in oncology clinical trials,” published in npj Precision Oncology, examines how clinical trial matching is changing in the era of precision medicine, digital health, and artificial intelligence. The authors emphasize that oncology trial enrollment is limited not only by patient awareness but also by fragmented clinical data, complex inclusion and exclusion criteria, limited research staff, and the underrepresentation of rural and underserved populations.
Several numeric findings highlight the scale of the problem. Approximately one-quarter of cancer clinical trials fail to enroll the planned sample size, and 18% close with less than half of their intended participants. From 2019 to 2023, the average enrollment duration for industry-sponsored interventional oncology trials increased by six months, representing a 19% relative rise.
Despite the potential value of trial participation, only 3–5% of eligible cancer patients enroll in oncology clinical trials. At the same time, more than 15,000 interventional oncology trials are recruiting or preparing to recruit patients globally, creating an urgent need for more efficient and equitable matching systems.
Methods and Study Design
This article is a narrative review rather than a clinical trial. The authors evaluated current approaches to oncology clinical trial pre-screening and screening, including manual workflows, electronic health record–based systems, health system tools, patient-facing recruitment models, and emerging artificial intelligence-enabled platforms.
The review also explored the role of large language models in trial matching. These included zero-shot approaches, retrieval-augmented generation, few-shot prompting, fine-tuned models, and hybrid AI-rule-based frameworks. The authors compared these tools in terms of scalability, accuracy, explainability, data requirements, workflow integration, and ethical challenges.
A major focus of the review was the need to move beyond a single-model solution. The authors proposed a unified framework that combines patient engagement, health-system integration, structured data, automated screening, clinician oversight, and equity-focused implementation.
Results
The review found that no single screening approach is sufficient for modern oncology clinical trials. Manual screening remains valuable because oncologists and clinical research teams can apply clinical judgment, interpret nuanced patient histories, and evaluate factors that may not be fully captured in structured databases. However, manual screening is labor-intensive, difficult to scale, and vulnerable to missed opportunities when data are incomplete or research staff are limited.
Automated screening tools offer speed and scalability. In one cited community oncology center study, AI-supported screening reduced the time needed to screen patients by 80%. These systems can rapidly process large volumes of data and may help identify eligible patients earlier in the care pathway.
However, their performance depends heavily on the quality, structure, and availability of clinical data. In oncology, key information is often scattered across electronic health records, pathology reports, molecular testing results, clinic notes, and external documents.
Large language models offer a promising but still developing pathway for oncology trial matching. Zero-shot models may help with broad pre-screening, but they can struggle when eligibility criteria require detailed interpretation of biomarkers, prior therapies, laboratory thresholds, comorbidities, or washout periods.
Retrieval-augmented generation and few-shot methods can improve performance by providing the model with relevant context and examples. Fine-tuned models trained on oncology-specific data may offer greater accuracy, especially when combined with structured inputs and domain-specific rules.
The authors emphasize that hybrid models appear most practical. In this approach, automated systems perform the first layer of pre-screening, while clinicians and research teams review and validate eligibility. This balances the efficiency of automation with the safety and judgment of human oversight.
Key Findings
A central finding of the review is that oncology trial matching must be dynamic. Patient eligibility can change quickly because of disease progression, laboratory changes, new imaging, prior treatment exposure, toxicities, or biomarker results. The authors note that the decision to enroll a patient in a trial often needs to occur within approximately two weeks, making timely pre-screening as important as accuracy.
The review also highlights the importance of performance metrics. Trial matching tools should not be evaluated only by how many patients they identify. They should also be assessed using sensitivity, specificity, positive predictive value, negative predictive value, enrollment rate, time to match, cost-effectiveness, diversity index, and transparency. These metrics can help determine whether a tool is clinically useful, equitable, and trustworthy.
The review further stresses that trial matching varies by therapy type. Biomarker-driven studies require accurate interpretation of molecular data, including mutation type, fusion status, allelic frequency, tumor mutational burden, and prior targeted therapy exposure.
Trials involving rare biomarkers, such as NTRK or RET fusions, face special challenges because these alterations occur in less than 1% of cancers overall. High-prevalence biomarkers, defined in the review as those present in more than 25% of patients, require scalable workflows that can handle larger patient pools without sacrificing precision.
Immuno-oncology trials introduce additional complexity. Screening must consider PD-L1 testing, prior therapies, autoimmune disease, organ function, and the presence or absence of targetable genomic alterations. Cell and gene therapy trials add logistical issues, including referral to specialized centers, manufacturing timelines, and the need to coordinate treatment readiness with product availability.
Equity and Ethical Considerations
The review places strong emphasis on equity. Rural patients, older adults, patients from lower socioeconomic backgrounds, and individuals from underserved regions remain underrepresented in oncology clinical trials. Major academic centers continue to dominate trial recruitment, which can limit access for patients who live far from these institutions or lack transportation, financial resources, caregiver support, or trial awareness.
The authors discuss the importance of patient-focused, health system-focused, and hybrid recruitment strategies. Patient-focused approaches use social media, advocacy groups, and patient information platforms to raise awareness. Health system-focused approaches integrate trial matching into clinical workflows, molecular tumor boards, and electronic health records. Hybrid approaches combine both strategies and may offer the strongest model by empowering patients while also supporting clinicians.
Ethical trial matching also requires attention to autonomy, privacy, algorithmic bias, multilingual consent, and transparency. AI tools should not function as black boxes. Clinicians need to understand why a patient was included or excluded. The review points to explainable systems, rule-based layers, structured rationale outputs, and retrieval-grounded architectures as important ways to make AI-assisted matching safer and more auditable.
Key Takeaway Messages
Oncology clinical trial enrollment remains limited despite the growing number of available studies.
Only 3–5% of eligible cancer patients participate in clinical trials, while more than 15,000 interventional oncology trials are recruiting or preparing to recruit patients.
Manual screening is clinically nuanced but difficult to scale.
Automated and AI-enabled tools can reduce screening time and improve efficiency, but they require high-quality data, strong governance, and clinician oversight.
Large language models may support trial matching, especially when combined with retrieval systems, oncology-specific training, structured data, and rules-based validation.
Hybrid screening models appear to offer the best balance between speed, accuracy, safety, and clinical trust.
Equity must be built into trial matching systems from the beginning, especially for rural, older, socioeconomically disadvantaged, and underserved patients.
Conclusion
This review presents a practical and timely framework for improving oncology clinical trial pre-screening and screening. As cancer care becomes increasingly personalized, the ability to identify the right patient for the right trial at the right time is becoming a core component of precision oncology.
AI-enabled tools, including large language models, can help address the growing complexity of trial eligibility and fragmented clinical data. However, automation alone is not enough. The most effective future model is likely to be hybrid: digital tools for speed and scale, combined with clinician oversight for judgment, safety, and patient-centered decision-making.
The review makes clear that trial matching should not be measured only by efficiency. A successful system must also improve access, protect patient autonomy, reduce disparities, and ensure that clinical trials better reflect the populations they are intended to serve.
By integrating AI, structured data, ethical oversight, and inclusive recruitment strategies, oncology clinical trials can become more efficient, more representative, and more accessible to patients who may benefit from investigational therapies.
More articles like this are available on OncoDaily
Written by Nare Hovhannisyan, MD