Patient Identification at Population Scale
The eligible patient population for most trials is not hidden. They're in the healthcare system — in EHR records, claims databases, disease registries, and patient advocacy networks. The problem is connection: eligible patients and open trials don't find each other reliably.
Why patients don't get referred
The primary care physician managing a patient with a rare disease is not systematically thinking about clinical trial eligibility. The oncologist managing a patient with a rare mutation is not automatically aware of trials specifically designed for that patient. Referral to clinical trials is reactive and relationship-dependent, not systematic.
This means that for most trials, a significant fraction of eligible patients are never screened — not because they don't exist, but because no one made the connection.
AI-enabled patient identification approaches
Claims-based identification. For many indications, claims data can identify patients who match key eligibility criteria: diagnosis codes, procedure codes, prescription histories. AI analysis of claims databases can produce lists of potentially eligible patients, segmented by geography and referring physician, that site coordinators can use for outreach.
EHR-based phenotyping. Within participating health system networks, AI can analyze structured and unstructured EHR data to identify patients who match trial eligibility criteria — including criteria that appear in clinical notes rather than structured fields. This is increasingly common in academic medical center partnerships.
Patient registry matching. Disease advocacy organizations often maintain patient registries. AI-assisted matching of trial eligibility criteria against registry data surfaces interested and eligible patients faster than any other channel.
Digital cohort identification. For conditions where patients self-identify online — rare diseases, chronic conditions with active patient communities — AI analysis of social media and patient forum data can identify potential eligible patients for outreach through advocacy partnerships.
Governance requirements
Patient identification using real-world data requires careful governance. HIPAA applies to any use of identified patient data. IRB approval is required for any prospective patient contact. Claims-based identification is typically done at population level without identifying individuals until consent allows further contact.
The governance layer is not optional — but it's also not a reason to avoid the approach. Patient identification using AI and real-world data is standard practice at leading clinical development organizations.