The right patients revealed: A fresh approach to rare disease trials

Introduction

Rare disease programs rarely fail because a therapy lacks promise. More often, they falter under the weight of bad data, underpowered designs, and recruitment tactics from another era. The impact is profound: millions in development spend evaporate, and patients wait, sometimes indefinitely, for therapies that could change or save their lives.

What makes rare disease programs uniquely fragile is the gap between clinical documentation and research needs: administrative code sets miss critical nuance, while high-value details (genotype, severity, phenotype, temporality) are buried in unstructured notes. Without a way to surface and standardize that information across sites, feasibility, cohorting, and analytics rest on a weak foundation.

This eBook explains why failures begin long before the first patient in. It then lays out a next-generation model rooted in data normalization, intelligent cohorting, and workflow-native activation to shorten timelines, reduce costs, and raise enrollment quality so no eligible patient is left behind.

Chapter One – Why rare disease trials fail before they start

About 50% of rare disease randomized controlled trials are discontinued or remain unpublished.¹ That figure represents more than sunk costs; it reflects missed patients who qualify but are never found, and promising therapies that stall or are abandoned. 

Many otherwise eligible patients, often highly knowledgeable about their condition, still never learn about open studies. This under-enrollment fuels what some call orphan, lost, or unseen diseases, and the burden extends beyond sponsors to families and society.

Two intertwined drivers underpin the problem:

• Operational drag: labor-intensive pre-screening, narrow site networks, and slow activation that suppress throughput and limit access to community settings.
• Data fragmentation: insufficient representation of rare disease in electronic health record (EHR) nomenclature and inconsistent, non-research-grade detail flowing into feasibility and accrual.

When the data foundation is thin, the processes built on top of it buckle. Sites overestimate eligible pools, screen failures spike, protocol amendments are required to address criteria that proved unworkable in practice, and timelines slip. 

The common root is clear – rare conditions are poorly represented in administrative code sets, and clinical data lacks the standardization research requires.

Chapter Two – The operational challenge

Sites often struggle to identify and enroll eligible patients fast enough to meet timelines.

Many teams still rely on manual code curation and chart review by coordinators and investigators who already carry clinical loads. That burden discourages community sites, where many patients actually receive care, and concentrates research in large academic centers. At many community sites, protocol complexity, limited IT capacity, and scarce research time lead to non-participation altogether.

Even when sites do participate, many underperform, not because patients are absent, but because International Classification of Diseases (ICD)-based filters may pull in clinically similar conditions, inflating lists with false positives and burying true eligibility under hours of manual verification. Administrative lists can make pools look large on paper, while coordinators lack the capacity to locate truly eligible candidates. The result is time diverted from patient outreach and consent to adjudicating false positives.

The common “fix” is to add more sites, which increases oversight complexity and cost without restoring throughput or diversity. 

Operational friction accumulates from feasibility to first-patient-in unless eligibility logic is delivered directly inside site workflows.

Chapter Three – The data challenge

Healthcare data is heterogeneous and siloed. The same clinical reality is encoded differently across systems, and more than 80% of the meaningful detail lives in unstructured text,² including Subjective, Objective, Assessment, Plan (SOAP) narratives where disease course, severity, and genetics actually reside. Downstream analytics and AI inherit these inconsistencies. Ultimately, without normalization and context-aware extraction, feasibility skews, cohorts drift from clinical intent, and statistical power erodes.

Because the data isn’t aligned to common standards, teams spend significant time and budget on clean-up, and AI models trained on messy inputs lose accuracy and propagate bias. Making the data usable at scale requires consistent mapping of labs, medications, procedures, and diagnoses to industry standards (e.g., LOINC®, RxNorm®, CPT®, SNOMED CT®) so outputs remain comparable across sites and studies.

Compounding the challenge is schema drift across institutions, such as different field names, units, reference ranges, and local vocabularies. Even high-quality single-site datasets can fail to generalize when pooled, unless harmonization and codification are enforced upfront.

Chapter Four – When codes erase clinical intent

Administrative code sets, such as ICD-10-CM, were built for reimbursement, not for research. In rare disease, that mismatch is decisive:

• Only about 7% of rare diseases have a unique ICD-10-CM code; therefore, of the approximately 7,000 rare diseases, only 500 or so have a dedicated ICD-10-CM entry.³
• Codes under-capture disease. For example, sepsis is only correctly identified in approximately 35% of cases by coding.⁴
• Approximately 20% error rates have been reported for key conditions.⁵
• Biomedical evidence doubles roughly every 12 months, yet major code set transitions, including ICD-10-CM to ICD-11-CM, unfold on decade timelines.⁶

Even with ongoing additions driven by rare disease communities, administrative coding still fails to carry the nuance research needs. The practical outcome is that distinct diseases documented at the point of care collapse into shared administrative categories, stripping out stage, histology, genotype, and severity, which are precisely the details trials rely on for eligibility, stratification, and endpoints.

Chapter Five – Preserving intent with granular clinical terminology

A better foundation begins with the language clinicians actually use. 

IMO Health’s interface terminology captures real clinical phrasing and maps it into administrative code sets, ensuring that billing and operations continue to function without sacrificing clinical intent. The terminology is continuously curated by medical terminologists and practicing clinicians, building on decades of terms that are released up to five times per year to meet both regulatory and non-regulatory needs – including client term requests.

Consider muscular dystrophy disorders: Duchenne and Becker are clinically distinct yet often share an ICD category. In IMO Health’s model, they are separately represented, linked to distinct SNOMED concepts, and mapped appropriately to ICD-10-CM. At this level of detail, concepts can also carry genotype and etiologic signals such as biomarkers, specific mutations, and relevant family history, so rare-disease cohorts preserve the biology that eligibility and endpoints depend on.

Scale and interoperability matter:

• ICD-10-CM: Approximately 75,000 codes
• IMO Health: More than 1 million clinical concepts across the ecosystem (including synonyms, acronyms, misspellings, variants, mutations, severity, familial patterns) and over 5 million terms that account for complex clinical nuances through greater specificity and language that accurately represents patient care
• Each concept maps cleanly across ICD-9-CM, ICD-10-CM, and SNOMED CT, enabling consistent cohorting, surveillance, and analytics across datasets and time.

Independent validation indicates a near-identical match to manual review. In a CDC study comparing manual chart review (the gold standard) with ICD-only and ICD with structured-data algorithms for a cardiac condition, the IMO Health terminology-based method matched manual review almost one-to-one, differing on only one chart^.7 In practice, that allows machine-driven identification to approach the fidelity of human chart review, without human cycle times.

Chapter Six – Clinical trial enablement solutions

With IMO Health terminology and EHR-native workflows as the foundation, trial operations can change end-to-end and stay inside the tools sites already use:

• Site selection: Combine claims-based market intelligence with clinician activity signals to identify where providers are actively diagnosing and treating patients with the target condition. Validate those signals against real-world EHR data to confirm actual protocol-aligned patient counts before site activation.
• Site activation: Provide electronic health record (EHR)-native assets that sites can import and use directly in their existing workflows, with minimal IT build. Sponsors typically save up to two months of IT enablement. Coordinators can start pre-screening immediately, and providers receive point-of-care alerts inside the EHR.
• Patient identification: Use high-precision filters to remove noise at the source. In documented programs, this has cut false positives by up to 99%, which reduces screen failures and lets clinical research coordinators (CRCs) spend time on patient outreach instead of adjudication. Once filters and work queues are live, sites report eliminating most manual chart reviews.
• Trial result collection: Pull structured variables directly from clinical notes and auto-populate the electronic data capture (EDC) from the EHR, eliminating duplicate data entry. This improves completeness (fewer missing fields), reduces entry errors, and speeds interim analyses and data cleaning.

The net effect is faster trial start-up, more accurate site lists, fewer non-performers, cleaner datasets, and a higher proportion of qualified referrals.