M-PROVE: Model performance and output validation engine
Making Model Validation Fast and Easy
We offer model validation services using multi-modal health data for internal researchers and external researcher partners with a UMN collaborator.
Partner with our experts to validate your model
Validate your clinical predictive algorithm, or NLP routine in partnership with a UMN researcher. The Center for Learning Health System Sciences provides secure, compliant access to multimodal electronic health record (EHR) data to validate model performance on independent clinical cohorts.
Is your project a good fit for M-PROVE?
Promising performance
M-PROVE is meant for pre-trained models with promising performance that need to be tested on a new patient cohort.
Multi-modal data available
M-PROVE can provide secure compliant access to images, notes, structured EHR, genomic, and video data.
Collaborate with a UMN researcher
M-PROVE is available to internal UMN researchers and external researchers with a UMN collaborator (we can help with match-making if you don't have one).
Flexible Service Tiers & Pricing
Data & Computation Only: $5,000 avg.
Best for teams with internal analysts who need clean extracts and secure space.
- Custom cohort identification & data retrieval
- Cohort descriptive statistics
- Secure isolated workspace for your team to conduct annotation and model validation
- Technical onboarding for working in the secure environment
Full-Service Validation: $20,000 avg.
Best for investigators seeking hands-off, end-to-end evaluation reports.
- Custom cohort identification & data retrieval
- Cohort descriptive statistics
- Reference standard created by expert annotation
- End-to-end model execution
- Comprehensive custom performance report
Full-Service Validation Project Timeline
1. Services Contracting
Weeks 1-4
IRB, data use agreements, sales contracting.
2. Data Retrieval
Weeks 5-7
Cohort extraction & data procurement.
3. Model Run
Weeks 8-12
Annotation, debugging, execution, & analysis.
4. Reporting
Weeks 13-16
Statistical validation & final performance report.
Ready to connect? What we need from you:
To provide an accurate quote, our intake team requires:
- Model Architecture: Delivery format (Python, Docker, etc)
- Input Specs: Clinical elements & variables needed
- Target Outcomes: Defined endpoints & thresholds
- Regulatory Status: Current IRB approval status