ICDA: a platform for Intelligent Care Delivery Analytics

Abstract

The identification of high-risk patients is a critical component in improving patient outcomes and managing costs. This paper describes the Intelligent Care Delivery Analytics platform (ICDA), a system which enables risk assessment analytics that process large collections of dynamic electronic medical data to identify at-risk patients. ICDA works by ingesting large volumes of data into a common data model, then orchestrating a collection of analytics that identify at-risk patients. It also provides an interactive environment through which users can access and review the analytics results. In addition, ICDA provides APIs via which analytics results can be retrieved to surface in external applications. A detailed review of ICDA's architecture is provided. Descriptions of four use cases are included to illustrate ICDA's application within two different data environments. These use cases showcase the system's flexibility and exemplify the types of analytics it enables.

Figure 1:

The ICDA architecture consists of an analytics subsystem to coordinate analytic processing, and an access subsystem to expose the resulting risk assessments. Analytic plugins (each consisting of an analytic engine, a serializer, and a display package) can be deployed in arbitrary combinations to support specific use cases.

Figure 2:

Data flow within the ICDA system. (a) New patient data arrives via an ETL process and is stored in the staging database. (b) The analytic controller then runs the analytic engines and stores results in the case database. (c) When the results are ready, the deployed and staging databases are swapped. (d) Finally, the new staging database is synchronized to prepare for the next iteration of analytics.

Figure 3:

The web interface for ICDA provides users with (a) an inbox showing patients flagged as high risk, (b) a patient summary view which provides a temporal summary of the medical record for an individual patient, and (c–f) targeted reports that convey detailed information supporting each of a patient’s identified risks. Zooming in on the images in electronic versions of this paper will reveal the full resolution of the screenshots.