Truveta Language Model

Truveta Language Model (TLM) is a large-language, multi-modal AI model purpose-built for healthcare, designed to clean and structure billions of EHR data points daily with industry-leading precision. It is built for life sciences researchers, clinical investigators, and data scientists who need research-ready, high-quality patient data at scale — addressing the core challenge that 95% of healthcare data goes unused due to fragmentation, inaccessibility, and lack of structure.

TLM is trained on the largest collection of complete medical records for more than 100 million patients, representing the full diversity of the United States. Unlike general-purpose large language models trained on public internet data, TLM's specialized training on healthcare data ensures clinical validity and the rigor required by regulators and scientific reviewers.

Core Data Cleaning and Normalization Capabilities

• Cleans all types of EHR data, including semi-structured data such as lab tests and diagnoses, as well as unstructured data such as clinical notes and imaging reports.
• Normalizes raw medical text to the most appropriate medical ontology with standard units of measurement, producing research-ready inputs.
• Maps clinical concepts — including lab test results, diagnoses, medications, and clinical observations — to standard medical ontologies within the Truveta Data Model (TDM).
• Processes and structures data continuously, with daily updates ensuring timely access to the latest patient information.

Unlocking Data Hidden in Clinical Notes

• Provides access to more than 7 billion clinical notes generated during patient care, including progress notes, nursing evaluations, procedure and operative reports, referral notes, and discharge summaries.
• Extracts and normalizes concepts from free-text clinical notes, eliminating the need for manual, human-based abstraction.
• Uncovers critical data points such as disease staging, adverse events, and medication rationale changes directly from unstructured note content.
• Supports research across a broad range of conditions, including colon cancer, migraines, seizures, NASH, heart failure, vessel disease, hypercholesterolemia, rare diseases, and more.
• Enables researchers to understand complete clinical context and pursue novel research questions that would otherwise be inaccessible.

Data Quality and Accuracy

• Aims to exceed the accuracy of clinical experts reviewing medical records, providing the transparency and rigor required for regulatory trust.
• Achieves high accuracy across diagnoses, medications, lab results, clinical observations, and additional data types.
• Outperforms general-purpose and specialized models including GPT-4, LogMap, and AML on medical normalization tasks.
• Specialized healthcare training is critical for ensuring clinical validity — general LLMs trained on public internet data are frequently inaccurate within the medical domain.

Broader Truveta Data Ecosystem

• TLM-cleaned data is part of Truveta Data, which covers more than 120 million patients representing the full diversity of the US.
• EHR data is linked with social determinants of health (SDOH), mortality data, and claims data for a comprehensive patient view.
• Data is complete, timely, and structured within the Truveta Data Model, enabling scientifically rigorous research workflows.

Truveta Language Model is supported by published whitepapers on both the model itself and data quality, providing scientific transparency for researchers and organizations requiring validated, regulator-trusted data infrastructure.