for Optum Health

Clinical decision support @ Scale

530

Million FHIR resources

4

milliseconds avg latency per CQL definition

100

Faster than CQF ruler

“By far the best team I’ve ever worked with. They truly understood the look I was going for and completely nailed it!”

What is Clinical Decision Support?

We helped Optum build a decision support tool that computes thousands of decisions per second for millions of patients enabling clinicians to have access to published care guidelines in real-time.

Clinical decision support at the point of care is critical in improving patient outcomes and reducing costs. Clinical Quality Language (CQL) is an emerging standard for describing and authoring clinical decision support rules. Current solutions for executing CQL do not scale in terms of big data.

We leveraged our expertise in Apache Kafka, Health IT standards, and high computational computing to design and build a solution that works.

We published the design as a podium abstract for the American Medical Informatics Association (AMIA), and the upcoming Austin Kafka Summit.

Standards Driven

Organizations often rely on our deep experience in health IT to implement solutions in FHIR and CQL. Here, FHIR serves as a data format while CQL handles query syntax, allowing clinical content authors to design rules in the language they feel most comfortable.

Standards are not only crucial in describing health technology, but also in the design of the computational system. Proven technologies like Apache Avro, Apache Kafka, and RocksDB provide a robust and compact method for data governance, compatibility, and data integrity.

Hello standards, meet performance

Standards

We benchmarked existing open-source solutions for executing CQL, and found none of them could perform fast enough to meet our need for computation on 5 million patients in real-time.

Performance

At the heart of this system are performance and scalability. So, we wrote one in Scala that runs on RocksDB and Apache Kafka that performed 10x faster than OpenCDS and CQF Ruler at CQL execution while delivering horizontal scalability and fault tolerance.