Paradigm

Solution

Paradigm replaced a slow patchwork of specialized healthcare ML and NLP models with GPT-4 delivered through the OpenAI API to analyze clinical-trial data and patient records. Instead of building, training, and validating a separate model for every data element, the company now uses GPT-4’s long context and multimodal capabilities to summarize unstructured medical text, identify relevant clinical details, and match patients against trial criteria. The system was evaluated on expert-curated golden datasets and outperformed state-of-the-art domain models by at least 10% on a blended precision/recall metric, with especially strong gains on complex information spread across multiple documents. Because the resulting output requires far less validation data and less manual checking from clinicians, Paradigm can expand faster to new providers and trials while moving toward interactive patient-eligibility conversations that further reduce burden on nurses and doctors.

Data flow

Medical records, notes, and trial documents are passed through the OpenAI API to GPT-4, which extracts structured facts and eligibility signals from unstructured content. Paradigm then validates outputs against expert-curated datasets and uses the resulting matches to support clinicians and provider partners in enrollment workflows.