Paradigm Health

Language models do the eligibility reasoning, and the company published a direct comparison making that concrete: it evaluated its AI native matching against a traditional rule based platform relying on structured data and pre parsed fields across four cancer trials, reporting that in three of them the AI approach reduced screening volumes by 31 to 87 percent by surfacing candidates closer to full eligibility. The architecture is described specifically as structured data prefilters enhanced with LLMs reasoning over full text clinical notes. Trial eligibility criteria live in prose, which is precisely why rules engines underperform here.

The design keeps a human at every consequential step and pairs software with staff rather than substituting for them. AI surfaces candidates with clear rationale and supporting clinical context for site review, continuously monitors eligibility and prioritizes candidates, while recruitment coordinators work alongside site teams on a shared platform to confirm eligibility and progress patients. Surfacing rationale rather than a bare match score is what makes the site review meaningful, and final eligibility determination and consent remain with the research team.

Unusually specific about architecture and its evolution. The company describes moving away from an earlier approach of fine tuned specialized models built use case by use case, each requiring separate training and validation, toward general purpose LLM evaluation, and reports needing roughly one tenth of the data to validate the newer approach. Its current design is documented as structured prefilters plus LLM reasoning over clinical notes. What is not published is model versioning, ongoing evaluation methodology, or performance by trial complexity.

Strong on both scale and comparative rigor. Network reach is specific: 45 states, 166 provider organizations, 2,100 care locations including 123 community oncology practices and 43 health systems and academic centers of which 23 are NCI designated, which the company frames as putting 70 percent of the US cancer population within reach of a trial site. Outcome claims include health systems accruing four times faster and a named practice reporting a 45 percent enrollment increase over two years. Most importantly, the head to head comparison against rule based matching was presented as an abstract at a 2025 ASCO symposium rather than asserted in marketing. Figures remain vendor reported.

No published PHI framework or data governance disclosure was located, which is a notable gap given the platform reads full text clinical notes plus molecular and genomic testing results across 166 provider organizations, and routes candidate information toward sponsors. A buyer should map exactly what patient level data crosses from the provider to the sponsor and at what point in the funnel.

No HIPAA or BAA commitment was located in public materials. Business associate arrangements are structurally required given deep EHR integration at health systems, and the research context adds IRB and informed consent obligations layered on top, but nothing is published.

No SOC 2, HITRUST, or ISO 27001 attestation was located, and no trust center was found.

No FDA pathway applies to trial matching software, which identifies candidates for research rather than informing diagnosis or treatment. The governing regime is human subjects research regulation, meaning IRB oversight, informed consent, and Good Clinical Practice, plus sponsor obligations under trial protocols. Worth noting for buyers that an algorithm influencing who gets offered a trial has equity implications that sit inside research ethics rather than device regulation.

Equitable access is the company's stated founding mission, and the network strategy is a substantive response: extending trial availability into 123 community oncology practices rather than concentrating in academic centers directly addresses the geographic and socioeconomic skew in trial participation. That is more than positioning. What is missing is measurement: no published analysis of whether AI surfaced candidates differ demographically from those found by manual screening, which is the specific risk when a model trained on documentation patterns decides who gets seen.

Deep EHR integration is the precondition for the product and is described as such: the platform deploys inside participating provider organizations, integrating with site EHR systems to evaluate structured and unstructured clinical data including molecular and genomic testing results against protocol criteria, with embedded screening workflows in clinical workflow rather than a separate portal. Sustaining that across 166 provider organizations and 2,100 locations is a substantial integration footprint rather than a pilot.

No hosting, tenancy, or data residency terms were located. International expansion, including announced collaborations in Israel and Japan, raises cross border data questions the published materials do not address.

No published pricing, but the commercial structure is disclosed in outline and is unusual enough to matter: sponsors and CROs pay for access to a research ready provider network, while provider organizations gain trial access and recruitment support. The company reports relationships with 75 percent of the top 20 largest biopharma sponsors. A provider should establish what it pays, what it receives, and what obligations attach, since the economics flow primarily from the sponsor side.

Oncology is the clear center of gravity, evidenced by the NCI designated center count, community oncology network, and cancer population reach framing, though the platform supports Phase I through IV trials generally and the company has launched a trial design service for interventional and pragmatic studies embedded in routine care. Settings span community practice through academic medical center, which is the meaningful breadth here, with announced international expansion beyond the US network.