Research & Methodology
How we validate and benchmark the UniteRare platform.
Diagnostic Benchmark — GA4GH Phenopacket Store
Benchmarked against the GA4GH Phenopacket Store, a curated dataset of confirmed rare disease diagnoses from published clinical case reports. This is the same benchmark used by academic rare disease diagnostic tools.
These are internal benchmarks and have not yet been peer-reviewed. Publications are in preparation. Results reflect platform performance on the validation dataset and should not be interpreted as clinical diagnostic accuracy.
Multi-Pathway Consensus Approach
The UniteRare diagnostic assist engine uses five independent reasoning pathways, each processing phenotype data through a different methodology:
- Semantic phenotype matching — HPO term overlap with Orphanet disease definitions
- Differential rank fusion — weighted combination of independent differential diagnosis rankings
- Literature pattern matching — case report similarity using published phenopackets
- Genetic pathway analysis — gene-disease association reasoning (where genetic information is provided)
- Clinical trajectory modeling — age of onset, disease progression, and inheritance pattern fitting
Results across pathways are fused using a consensus algorithm. Conditions appearing across multiple pathways receive higher confidence scores.
Appeal Benchmark
The AI insurance appeal letter generator was evaluated against a set of denial scenarios across rare disease categories. Letters are assessed across multiple dimensions including:
- Medical necessity argumentation accuracy
- Regulatory citation accuracy (CMS, state-level appeal rules)
- Clinical evidence citation quality
- Letter structure and completeness
All appeal letters require review by the patient's healthcare provider before submission.
Data Sources
- Orphanet — disease definitions, prevalence, HPO annotations, gene associations
- ClinicalTrials.gov — recruiting clinical trial data (updated regularly)
- FDA Drugs@FDA & DailyMed — approved drug labels, orphan drug designations
- PubMed Phenopackets — clinical case reports for benchmarking
- NORD — rare disease organization and patient advocacy data
Publications
Publications describing the UniteRare platform methodology are in preparation. We are committed to publishing our benchmarks and validation methods in peer-reviewed venues.
If you are a researcher interested in collaborating, contact us at team@uniterare.com.
Intellectual Property
The UniteRare platform includes proprietary methods covered by pending patent applications. See our Intellectual Property page for details.