LEWES, Del., March 24, 2026 (GLOBE NEWSWIRE) -- John Snow Labs, a healthcare AI company, is proud to announce that it has been named the winner of the Real World Evidence (RWE) Catalyst Challenge at PHUSE US Connect 2026. The award recognizes the company’s groundbreaking framework for automating oncology data abstraction – a process traditionally so complex it has remained largely manual until now.
The winning submission, titled "Scaling Regulatory-Grade RWE: A Hybrid NLP, SLM, and Deterministic Reasoning Framework for Automated Cancer Registry Abstraction," was presented by David Talby, CEO at John Snow Labs and Veysel Kocaman, CTO at John Snow Labs. The project addresses the critical delay in oncology data — where datasets are typically 12–24 months old by the time they are used — by introducing an automated, highly accurate alternative to manual curation.
The Challenge of Complexity and Volume
The primary reason oncology registries have resisted automation is the sheer complexity of the data. Modern staging guidelines, such as SEER and AJCC (versions 8 and 9), span over 2,500 pages of intricate, version-dependent rules. This high level of specialization makes it exceptionally difficult for general-purpose AI to apply rules consistently without hallucinating or losing the necessary clinical context.
This complexity is compounded by unstructured data noise at an unprecedented scale. A typical cancer patient in the US generates more than 1,000 pages of text per year, leaving a registrar with several thousand pages of clinical notes, pathology reports, and imaging results to parse through for a single patient's history. Compounding this is the "needle-in-a-haystack" problem: empirical analyses show that 96% of electronic pathology reports processed by health systems are non-reportable, meaning human experts currently spend most of their time searching through irrelevant data rather than performing expert abstraction.
Achieving Regulatory-Grade Accuracy for Complex Cancer Registries
John Snow Labs’ framework is the first to achieve the precision required for regulatory-grade real world evidence (RWE) in oncology by utilizing a "governance-by-design" architecture. The solution is now a core component of the company’s Patient Journey Intelligence Platform, which is fully aligned with the latest FDA guidance on using RWE as primary evidence for regulatory decision-making. The framework utilizes a multi-layered approach to ensure success:
- Healthcare-Specific NLP: Specialized models perform massive-scale "relevance filtering," successfully triaging thousands of pages of noise to identify the few critical documents containing reportable cancer cases.
- Medical Small Language Models (SLMs): Unlike frontier LLMs that require data to be sent to external APIs, SLMs are deployed locally within a secure environment. They’re fine-tuned specifically for clinical reasoning, ensuring high precision in extracting variables like Primary Site, Histology, and TNM Stage.
- Deterministic Reasoning: A specialized logic layer applies the 2,500+ pages of SEER and AJCC rules to the extracted data, ensuring that the final output follows strict medical guidelines rather than probabilistic guesses.
- Audit-Readiness and Full Traceability: For the first time, every single extracted variable is mapped directly to source evidence in the medical record. This provides the full data provenance and transparency required for auditors to verify the accuracy of the automated findings instantly.
The performance results demonstrate a radical shift in what’s possible for RWE, including:
- A 98% Reduction in Effort: Total abstraction time per case was slashed from approximately 120 minutes of manual searching to less than two minutes of verification.
- Regulatory-Grade Precision: The system demonstrated high concordance with Certified Tumor Registrar (CTR) validated reference standards, meeting the strict requirements for regulatory submissions.
- Real-Time Active Surveillance: The solution enables a shift from 12-month retrospectivity to near-real-time updates, allowing researchers to monitor treatment outcomes as they happen.
"For the first time ever, we have demonstrated that AI can reach the level of accuracy required for the world’s most complex cancer registries," said Talby. "By reaching regulatory-grade accuracy, we are moving beyond simple data extraction to true evidence generation, enabling researchers to access high-quality, audit-ready oncology data in days rather than months, and meeting fundamental requirements for the FDA-ready patient journey intelligence we provide to our partners."
About John Snow Labs
John Snow Labs, the AI for healthcare company, provides state-of-the-art software, models, and data to help healthcare and life science organizations put AI to good use. Developer of Medical LLMs, Healthcare NLP, Spark NLP, the Generative AI Lab, and the Patient Journeys Platform, John Snow Labs’ award-winning medical AI software powers the world’s leading academic medical centers, pharmaceuticals, and health technology companies. Creator and host of the Applied AI Summit (formerly the NLP Summit), the company is committed to further educating and advancing the global AI community.
Contact
Gina Devine
Head of Communications
John Snow Labs
gina@johnsnowlabs.com
