Healthcare: Genomic Patient Matching for Clinical Trials
1. Problem
1a. Statement
Oncology clinical trials have a 3.4% success rate, the lowest of any therapeutic area, with poor patient-drug matching driving many failures. At an NCI-designated comprehensive cancer center, manually reviewing genomic profiles against trial eligibility criteria took weeks per patient cohort. With median trial timelines stretching to 13 years and costs exceeding $1.5 billion per approved drug, the center needed an ML-powered system to automate genomic patient matching while maintaining FDA-compliant explainability.
1b. Client Profile
1c. Motivation
2. Analysis
2a. Requirements
The solution required an ML platform capable of ingesting and processing next-generation sequencing data, including whole exome sequencing and targeted gene panels, matching patient molecular profiles against clinical trial eligibility criteria. The system needed to integrate with Epic EMR infrastructure to pull patient demographics, treatment history, diagnosis codes, and lab results alongside genomic data from the institutional biobank. Explainability was non-negotiable, with every patient-trial match requiring documented reasoning that clinical researchers and IRB reviewers could audit for FDA compliance. The platform needed human-readable explanations detailing which genomic markers, prior treatments, and clinical factors contributed to eligibility. Real-time processing was essential, enabling oncologists to identify trial-eligible patients during treatment planning with results returned in under 10 seconds across 100,000+ patient records and 40+ active trials.
The solution required an ML platform capable of ingesting and processing next-generation sequencing data, including whole exome sequencing and targeted gene panels, matching patient molecular profiles against clinical trial eligibility criteria. The system needed to integrate with Epic EMR infrastructure to pull patient demographics, treatment history, diagnosis codes, and lab results alongside genomic data from the institutional biobank. Explainability was non-negotiable, with every patient-trial match requiring documented reasoning that clinical researchers and IRB reviewers could audit for FDA compliance. The platform needed human-readable explanations detailing which genomic markers, prior treatments, and clinical factors contributed to eligibility. Real-time processing was essential, enabling oncologists to identify trial-eligible patients during treatment planning with results returned in under 10 seconds across 100,000+ patient records and 40+ active trials.
2b. Constraints
3. Solution
3a. Architecture
3b. Implementation
4. Result
4a. DUBEScore™
4b. Outcomes
4c. Learnings
Epic EMR integration took 40% longer than estimated. Early data audits would have identified gaps sooner.
Explainability shaped the ML architecture. Building audit trails as a core feature enabled smoother IRB reviews.
Interactive cohort exploration was the most valued feature. Prioritize dashboard usability over batch workflows.
Ready to Build Your AI Solution?
Let's discuss how we can deliver similar results for your organization.