Figure 3.

DEPLOYR triggering mechanisms. Models deployed with event-based triggering logic are exposed as REST APIs on the inference engine using a python Azure Function application (DEPLOYR-serve). An event (A) in the EMR (eg, clinician button-click initiating a laboratory order) transmits an HTTPS request (B) directed at the exposed DEPLOYR-serve function, which wraps an ML model. The function transmits HTTPS requests (C) to REST APIs documented in Epic’s App Orchard to collect a feature vector, performs model inference, and directs the inference and resulting clinical decision support via HTTPS request (D) back into the EMR to interface with end-users. Models deployed with time-based triggering logic perform inference at set intervals (E) through use of Azure Function timer triggers. Every time interval (eg, 15 min), a DEPLOYR-serve function transmits HTTPS requests (F) to REST APIs to retrieve a batch of patient identifiers for whom inference should be made. Feature vectors are collected for the batch of patients (G), and inferences are transmitted back into the EMR (H). SHC: Stanford Health Care.