Back to Our Work
|
Healthcare

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
TypeCancer Research Center
IndustryHealthcare
SizeEnterprise
RegionOhio, United States
Users50+
1c. Motivation
Clinical Researchers
Weeks spent manually matching patients to trial eligibility criteria
Oncologists
Delayed identification of trial-eligible patients during treatment planning
Patients
Missed opportunities for precision therapies due to slow matching
Trial Coordinators
Administrative burden tracking eligibility across dozens of active trials
Data Science Team
No unified pipeline for genomic data processing and analysis
Compliance/IRB
Required explainability for patient selection decisions

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.

2b. Constraints
Compliance:FDA regulations and Common Rule for human subjects research
Integration:Epic EMR via FHIR APIs and institutional biobank data systems
Privacy:HIPAA required for all patient data; de-identification for research datasets
Processing:Sub-10-second query response for interactive cohort exploration
Scale:100,000+ patient records with genomic profiles across 40+ active trials
Explainability:Auditable reasoning for every patient-trial match decision

3. Solution

3a. Architecture
3b. Implementation
Discovery
12 weeks
Development
32 weeks
Integration
32 weeks
Deployment
20 weeks

4. Result

4a. DUBEScore™
4.3/5
D - Delivery4.4
U - Utility4.3
B - Business4.5
E - Endurance4.2
4b. Outcomes
Patient-trial matching time2 days (-85%)
Genomic records processed100,000+
Active trials supported40+
Match accuracy87%
Query response time<8 seconds
Researcher adoption50+ users
4c. Learnings
1

Epic EMR integration took 40% longer than estimated. Early data audits would have identified gaps sooner.

2

Explainability shaped the ML architecture. Building audit trails as a core feature enabled smoother IRB reviews.

3

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.