Over the past decade, oncology drug development has undergone a profound scientific transformation. Biomarker-driven therapies, adaptive designs, and tissue-agnostic indications have expanded what is clinically possible. Yet the operational engine of clinical trials, patient identification and enrollment, has not evolved at the same pace.
A recent framework published in npj Precision Oncology makes a clear and important argument: recruitment inefficiency is no longer a peripheral operational challenge but a structural constraint on oncology innovation. The authors note that roughly one quarter of oncology trials fail to meet enrollment targets and that only a small proportion of eligible patients ultimately participate in studies, typically just 3–5% of adult cancer patients in the United States. 1 For sponsors operating in increasingly competitive therapeutic landscapes, these statistics are not merely operational, they are strategic.
The central question facing clinical development leaders today is therefore not whether recruitment needs improvement, but how fundamentally the screening paradigm must change to keep pace with precision oncology.
Precision Medicine Has Rewritten the Screening Playbook
The modern oncology trial bears little resemblance to its predecessor from the cytotoxic era. Where once eligibility might have been defined primarily by tumor type and line of therapy, today’s protocols frequently require intricate combinations of genomic alterations, prior treatment exposures, organ function thresholds, and performance status criteria.
Horgan and colleagues identify three structural forces driving the current enrollment bottleneck: escalating eligibility complexity, fragmentation of clinical data, and persistent operational constraints at the site level. Each of these forces is independently challenging; together they create a combinatorial burden that overwhelms traditional workflows. 1
Consider the implications of increasingly common molecular enrichment strategies. Trials targeting alterations such as NTRK fusions, RET rearrangements, or rare KRAS variants often require screening large numbers of patients to identify a handful of candidates. In these contexts, manual chart review becomes economically and operationally untenable.
This shift is particularly visible in basket and umbrella trials, where eligibility logic must operate across tumor types and molecular strata simultaneously. The screening task becomes less about finding patients within a known pool and more about continuously surveilling a dynamic oncology population.
The Limits of Traditional and Rule-Based Digital Screening
Health systems and sponsors have not been passive in the face of these pressures. Over the past decade, a range of digital prescreening approaches has emerged, including structured EHR queries, molecular tumor boards, and genomic matching portals. These tools have delivered measurable gains in certain settings, particularly in well-resourced academic centers.
However, the Nature Portfolio framework is explicit that rule-based digital screening remains constrained by underlying data realities. Much of the clinically decisive information in oncology, functional status, progression context, toxicity nuance, continues to reside in unstructured clinical narrative rather than discrete fields. As a result, purely rules-driven approaches often operate on an incomplete representation of patient eligibility. 1
This limitation becomes evident in everyday scenarios. Performance status, for example, may be captured variably across institutions: structured in some systems, narrative in others, and occasionally absent altogether. A rigid query looking only for coded ECOG values will inevitably miss patients whose functional status is documented descriptively. Similar challenges arise with prior therapy timelines, radiographic progression language, and laboratory trends.
Prospectively captured effort data in oncology trials suggests eligibility screening can create substantial nonreimbursed burden, with estimated screening costs ranging from $129 to $336 per enrolled patient and annual screening costs exceeding $90,000 in one cancer trial program. 2
Independent analyses have shown that many clinically relevant eligibility variables in oncology remain embedded in unstructured records, limiting the performance of purely rule-based screening approaches and reinforcing the need for more advanced methods. 1 Emerging AI-based systems aim to address these gaps by extracting eligibility signals directly from clinical narrative and multimodal data. 3
AI as an Inflection Point in Trial Matching
Against this backdrop, artificial intelligence, particularly natural language processing and large language models, has emerged as a potential inflection point. Recent evaluations of large language model–based systems have demonstrated near-expert performance in criterion-level eligibility assessment tasks. 3, 4
AI-enabled approaches can substantially reduce the manual burden of prescreening by rapidly narrowing large patient populations to smaller sets of high-probability candidates.1 Complementary work in multimodal oncology artificial intelligence demonstrates that integrating imaging with rich clinical-text context (and, in broader multimodal pipelines, molecular data) can improve predictive and stratification performance. 5
What distinguishes the current generation of clinical trial matching tools from earlier waves of clinical informatics is the shift from static querying toward more continuous, automated eligibility assessment within the EHR environment. In advanced implementations, the arrival of new genomic, pathology, or treatment data can trigger reevaluation against active protocols, enabling more longitudinal patient surveillance rather than episodic screening. 1
Early operational deployments reported in the clinical research community suggest meaningful reductions in manual screening workload and faster identification of eligible patients when AI is embedded directly into site workflows. Yet the literature remains appropriately cautious about overinterpreting these gains. 6
Human Oversight Remains Essential
Despite rapid progress, the most credible analyses, including the Nature Portfolio framework, stop short of endorsing fully autonomous trial matching. The reasons are instructive and highly relevant for sponsors considering aggressive automation strategies.
First, oncology eligibility criteria frequently contain conditional logic that remains difficult for even advanced models to interpret reliably. Determining whether radiographic change constitutes true progression, reconstructing complex washout windows from fragmented treatment histories, or interpreting investigator discretion clauses often requires contextual clinical judgment. 1
Second, real-world data heterogeneity continues to challenge model robustness. Documentation practices vary widely across institutions, community sites often have less structured data capture, and key variables may be inconsistently recorded. These factors introduce noise that can degrade algorithmic performance if not carefully managed. 1
Third, governance and risk considerations remain front of mind. Investigators and regulators are understandably cautious about delegating enrollment decisions entirely to algorithmic systems, particularly when misclassification could expose patients to inappropriate therapies or deny access to beneficial trials. 1
Taken together, these realities support a more nuanced conclusion: AI is transformative for scale and speed, but human oversight remains essential for clinical nuance and accountability.
The Hybrid Model Is Becoming Operational Reality
The most consequential insight from the recent literature is the growing convergence around hybrid screening architectures. In this model, AI operates as a continuous prescreening layer within the EHR ecosystem, while human investigators and coordinators perform targeted validation and patient engagement.
This division of labor reflects the natural strengths of each component. AI excels at breadth, persistence, and pattern recognition across large datasets. Clinicians excel at contextual interpretation, edge-case reasoning, and patient-centered decision-making.
Horgan and colleagues explicitly conclude that hybrid frameworks integrating automated screening with clinician oversight are likely to be the most effective approach for improving both efficiency and representativeness in oncology trials. Many leading cancer centers are beginning to operationalize this model. 1
In practice, hybrid workflows often begin with continuous AI surveillance of the oncology population. Potential matches are surfaced directly within clinical or research work queues, enabling coordinators to focus attention where signal is highest. Investigators then confirm clinical appropriateness and engage patients earlier in the care journey. Early adopters report reductions in manual chart review burden and improved time to first patient, although performance varies with data maturity and workflow integration.
Data Architecture
One of the most important, and frequently underestimated, messages in the Nature framework is that AI performance is fundamentally bounded by data infrastructure. Even highly sophisticated models cannot compensate for fragmented, poorly normalized, or incomplete clinical data.
High-performing programs tend to share several characteristics:
- Structured genomic pipelines rather than scanned reports
- Standardized eligibility ontologies
- Robust ingestion of clinical narratives
- Interoperability between sponsor and site systems.
The authors specifically emphasize the importance of integrating genomic data directly within EHR workflows to enable real-time precision matching. 1
For sponsors, this has practical implications. Deploying advanced matching algorithms into environments where molecular results arrive as PDFs or where treatment histories are inconsistently coded will almost certainly yield disappointing results. Conversely, investments in data harmonization often produce outsized gains in matching performance even before advanced AI layers are added.
Equity and Representativeness: Opportunity and Risk
A notable strength of the review conducted by Horgan and colleagues is its sustained attention to equity. Oncology trials have long struggled with underrepresentation of rural populations, racial and ethnic minorities, and patients treated outside major academic centers. The authors emphasize that improvements in screening technology must be evaluated not only on efficiency but also on representativeness. 1
AI-enabled prescreening introduces both promise and risk. Properly deployed, continuous population scanning can identify eligible patients who might otherwise be overlooked, particularly in community settings. However, models trained on biased datasets or deployed unevenly across health systems may inadvertently reinforce existing disparities. 1
Forward-looking sponsors are increasingly evaluating diversity metrics alongside traditional performance indicators for matching platforms, alongside sensitivity and positive predictive value. This shift aligns with increasing regulatory and payer attention to trial representativeness.
Strategic Implications for Sponsors and Development Leaders
What ultimately emerges from the cumulative evidence is a reframing of recruitment itself. Patient identification is evolving from a periodic, site-driven activity into a continuous, data-enabled capability that spans the development lifecycle.
Organizations at the forefront of this shift are already adapting. Protocol teams are beginning to consider computability during eligibility design. Clinical operations groups are investing earlier in prescreening infrastructure. Sponsor–CRO partnerships are increasingly evaluated on data interoperability and workflow integration rather than purely on site activation speed.
Perhaps most importantly, recruitment performance is becoming a durable competitive differentiator. In biomarker-rich oncology pipelines, the ability to identify eligible patients faster and more reliably can materially influence development timelines, amendment frequency, and overall program risk.
Key Takeaways
- Oncology trial enrollment is increasingly constrained by eligibility complexity, fragmented data, and site-level capacity limits
- Traditional manual and rule-based digital screening approaches cannot scale to modern precision oncology pipelines
- AI-driven prescreening can materially improve candidate identification, particularly when unstructured clinical data and genomics are incorporated
- Fully autonomous trial matching remains premature; hybrid AI–clinician workflows currently represent best practice
- Data architecture and interoperability are primary determinants of real-world performance
- Sponsors that operationalize hybrid screening early are best positioned to accelerate enrollment and improve trial representativeness
About Arc Nouvel
Arc Nouvel supports biopharma sponsors navigating the growing complexity of modern oncology development. We partner with clients to design fit-for-purpose clinical strategies, strengthen operational readiness, and optimize trial execution from early through late phases.
With deep oncology expertise and a focus on practical implementation, Arc Nouvel helps organizations translate emerging innovation into measurable clinical development performance, including the practical integration of AI to improve recruitment and trial execution.
