Flatiron Health Launches Six AI Blood Cancer Datasets Covering 505,000 Patients
Flatiron Health has launched six new hematology datasets built using AI and large language models (LLMs), unlocking longitudinal clinical data for 505,000 patients with blood cancers. The release marks a reported six-fold increase in cohort sizes compared to Flatiron’s prior hematology datasets and expands access to rare and complex subpopulations.
The new datasets, launched under Flatiron’s Panoramic product line, cover five B-cell lymphoma subtypes and multiple myeloma. According to Flatiron, the datasets incorporate advanced molecular and treatment data, including coverage of measurable residual disease (MRD) testing and CAR-T therapy usage. These are designed to support both regulatory-grade real-world evidence (RWE) studies and exploratory research across traditionally underrepresented patient groups.
Flatiron applied its VALID data quality framework to maintain data fidelity while scaling extraction via LLMs. The company claims the datasets capture unstructured data from inpatient and outpatient settings and support endpoints such as clinical and molecular response, adverse events, and real-world progression.
Flatiron Health was founded in 2012 and in January 2013 raised an $8 million round led by Google Ventures. Roche acquired Flatiron in 2018 for $1.9 billion; the company continues to operate as an independent affiliate of the Roche Group.
Flatiron reports access to EHR-derived oncology data from more than five million patients across over 280 practices and 800 sites of care in the U.S., and it co-operates the Clinico-Genomic Database with Foundation Medicine, linking EHR outcomes with comprehensive genomic profiling from over 100,000 patients.
Regulatory collaborations include a five-year agreement launched in September 2023 with FDA’s Oncology Center of Excellence to study the use of real-world data and evidence. On methods, Flatiron has described its VALID framework for assessing accuracy of ML/LLM-extracted variables and has cited benchmarking against expert human abstractors.
Topic: AI in Bio
