Journal: NPJ precision oncology
This publication is a narrative review of current and emerging approaches to identifying and enrolling oncology patients into clinical trials, with a focus on pre‑screening and screening innovations.
Key points:
- Problem framing: Trial accrual in oncology is hampered by complex eligibility criteria, biomarker-based stratification, and fragmented clinical data, which prolong recruitment and keep participation rates low.
- Current approaches:
- • Manual workflows (chart review, ad hoc referrals, tumor board discussions) remain common but are labor-intensive, slow, and highly dependent on local infrastructure and personnel.
- • Health system–embedded digital tools (EHR-based trial finders, structured alerts, and rule-based matching systems) improve scale and standardization but struggle with unstructured notes, evolving criteria, and cross-institution variability.
- AI and large language model (LLM)–based methods:
- • The review describes emerging LLM strategies, including retrieval-augmented systems (linking trial criteria with patient data sources) and domain-adapted models trained on oncology-specific corpora.
- • These approaches aim to improve scalability and accuracy in matching complex eligibility criteria—including nuanced biomarker and comorbidity profiles—to real-world patient records.
- • The authors highlight equity considerations, noting that algorithmic bias, incomplete data, and uneven digital infrastructure can exacerbate disparities if not explicitly addressed.
- Comparative assessment:
- • Manual review: highest contextual nuance but poor scalability.
- • Rule-based/digital tools: better throughput but rigid and brittle to complex or ambiguous criteria.
- • LLM-based tools: promising for handling free text and complex logic, but require careful validation, governance, and monitoring to ensure safety, reliability, and fairness.
- Proposed direction:
- • Hybrid frameworks—automated pre-screening using AI or advanced informatics followed by clinician review—are presented as the most practical near-term model.
- • Such systems can streamline workflows, shorten time to identification of eligible patients, potentially increase trial representativeness, and improve timely access to investigational therapies, while keeping oncologists and trial teams as final decision-makers.
Overall, the article positions AI-enabled and especially LLM-driven trial matching as an important evolution in oncology research operations, with the greatest potential realized when integrated thoughtfully into clinician-centered, equity-focused workflows.