The Challenge
Patients with rare diseases are systematically missed by traditional commercial identification methods. Diagnosis of rare diseases depends on findings buried in physician narratives, radiology reports, pathology notes, and specialist observations. This information sits entirely outside the ICD-10 codes that claims-based patient finding relies upon.
The problem compounds across multiple fronts:
- No Code, No Cohort: Of roughly 10,000 known rare diseases, only about 500 carry a specific ICD-10 code. When a code does exist, it's often grouped with the common form of the disease, making rare-disease patients indistinguishable in claims data.
- The Data Sits Untouched: The richest diagnostic signal lives in unstructured EHR notes, physician narratives, discharge summaries, and pathology reports, but traditional patient-finding approaches mine only the limited structured data that's easier to access.
- Trial Recruitment Bleeds Time and Money: Clinical research staff spend 30 to 60 minutes per patient manually reading charts to determine eligibility. With a screen failure rate of 81% in rare disease, most of that effort is spent on patients who never enroll.
- Commercial Teams Fly Blind: Field teams deploy against proxy cohorts built on adjacent conditions and procedure codes, missing the patients who don't fit the pattern and the HCPs who are quietly treating them.
The Solution
A reproducible, five-pillar methodology extracts clinical signals from unstructured EHR notes and combines them with lab and claims data to produce a probability-scored patient cohort and a prioritized HCP list. This solution is built with R&D-grade rigor and is scalable to commercial deployment:
- The Right Data: A rigorous, multi-vendor data assessment across claims, EHR, notes, and real-world data sources, evaluated across eight dimensions to identify exactly which sources will yield the highest return.
- Unstructured Data Extraction: Customized LLM extraction of physician diagnoses, symptoms, lab values, and treatment patterns from free-text notes, run on the client's own.
- Human-in-the-Loop Verification: Trained clinical reviewers validate a randomly sampled set of AI-extracted patients to establish a gold-standard cohort, creating a feedback loop that refines extraction accuracy until performance targets are met.
- Operational Guide: A documented protocol covering chart review methodology, reviewer decision rules, statistical sampling, and quality standards, making the approach repeatable across geographies and therapeutic areas.
- ML Look-Alike Models: Probability scoring and risk stratification that identifies patients in the broader population who share the clinical profile of the confirmed cohort, even without direct diagnostic evidence.
What This Means for You
Commercial & Brand Leadership
Replace proxy targeting with evidence-based prioritization
- Surface confirmed patients invisible to claims-only approaches
- Deploy field teams to highest-match HCPs within 60 days of data access
- Produce evidence-based prevalence models for defensible peak sales forecasting
- Convert identified patients into treated patients with patient-level clinical evidence
- Expand reach into community and rural practices missed by specialist-only targeting
Medical Affairs & Field Medical
Ground every KOL and HCP conversation in documented clinical evidence
- Identify misdiagnosed patients and deliver supporting clinical evidence to field teams
- Inform KOL engagement and medical education strategy with real diagnostic-delay patterns
- Support label expansion and lifecycle planning with real-world signal detection
- Build a validated cohort that serves as a foundation for ongoing RWE generation
- Characterize the patient journey and line-of-therapy progression without new data purchases
Market Access & HEOR
Turn patient-level evidence into payer and regulatory leverage
- Reveal step-therapy blocks and prior authorization denials at the patient level
- Build proprietary real-world evidence that strengthens payer negotiations
- Support regulatory submissions with documented, chart-validated clinical signals
- Accumulate compounding RWE assets over a multi-year horizon
- Avoid the cost of purchasing additional RWD sources for comparative effectiveness analyses
Clinical Development & Trial Operations
Cut the manual burden out of patient identification
- Apply the same extraction methodology to eligibility screening, not just commercial targeting
- Reduce the 30-60 minutes of manual chart review currently spent per patient screened
- Identify patients who have documented symptoms that are consistent with the diagnosis, but no confirmed diagnosis or confirmatory testing
- Confirm eligibility against complex, multi-source criteria with less inter-rater variability
- Lower the cost burden of an 81% average screen failure rate in rare disease trials
In This Paper, You'll Discover
- Why claims-based approaches to patient finding structurally fails for rare disease populations
- How three complementary data sources, unstructured notes, lab and biomarker data, and structured claims, work together to close the gap
- The full Five-Pillar methodology, from initial data assessment to deployable look-alike models
- A complete case study: how the approach found 789 confirmed patients for a rare endocrine disease with no ICD-10 code
- Quantified results, including a $400K estimated annual ROI per patient who is converted to treatment and a greater than 96% LLM recall rate
- How durable assets built for patient finding create a foundation for long-term real-world evidence and market intelligence
Download the Full Paper
Access the complete white paper to learn how LLM-powered patient finding turns unstructured EHR data into a validated, actionable rare disease patient cohort and HCP target list.
FAQs
