Figure 2.

ML models are trained using data sourced from Stanford’s clinical data warehouse (STARR). In production, real-time data are sourced from the EMR’s transactional database (Epic Chronicles) through Epic and FHIR REST APIs. STARR data are several ETLs (extract, transform, loads) removed from the transactional database. Data mapping is necessary at inference time to ensure features seen during training match features seen in production. Mappings and inferences are invoked in DEPLOYR-serve.