Apollo: AI Model Creates Virtual Patient Representations from 25B Records
Apollo serves as a multimodal temporal foundation model, consolidating more than 30 years of longitudinal hospital data from a prominent US hospital network. It has been trained on 25 billion records, encompassing 7.2 million patients, and spans 28 medical modalities along with 12 specialties. The model develops a cohesive representation space for upwards of 100,000 distinct medical events, images, and clinical narratives, facilitating the generation of virtual patient profiles that encapsulate complete care pathways. This methodology tackles the issue of fragmented medical information spread across isolated systems.
Key facts
- Apollo is a multimodal temporal foundation model
- Trained on 25 billion records from 7.2 million patients
- Covers 28 distinct medical modalities and 12 major medical specialties
- Integrates over 100,000 unique medical events
- Learns a unified representation space for images, text, and structured data
- Creates virtual patient representations from entire care journeys
- Based on over three decades of longitudinal hospital records
- Developed using data from a major US hospital system
Entities
Locations
- United States