Children’s Hospital of Orange County Begins Machine Learning Journey | Healthcare Informatics Magazine | Health IT | Information Technology Skip to content Skip to navigation

Children’s Hospital of Orange County Begins Machine Learning Journey

June 26, 2018
by David Raths
| Reprints
William Feaster, M.D., M.B.A., CHOC’s chief health information officer, describes progress on turning data into intelligence

Leveraging its enterprise data warehouse and population health platform, Children’s Hospital of Orange County (CHOC) has developed smart registries and is now working on using automation for alerts to patients with asthma. It also is working with Cerner as a development partner on HealtheDataLab, a new big-data platform using Amazon Web Services.

In a June 25 webinar sponsored by the eHealth Initiative, William Feaster, M.D., M.B.A., CHOC’s chief health information officer, described how CHOC is leveraging its investment in EHR technology, data warehouses and machine learning.

Feaster started his presentation by quoting W. Edwards Deming: “In God we trust. All others must bring data.” He said he would amend that slightly to “all others musts bring intelligence,” and his talk was a demonstration of how a provider organization can turn data into intelligence.

He spoke about how CHOC is leveraging data to advance the health of its population. It is using Cerner’s HealtheIntent population health platform, which gives it the ability to bring data from multiple sources into a big data arena based on Hadoop. The data is cleaned up, normalized and matched to patients.

That has allowed the organization to work with Cerner to develop seven specific disease-specific “smart registries,” including one on asthma. “Registry data is available right when someone is seeing a patient at the point of care,” instead of a month or three months later as is the case with claims data, he said. In addition, clinical workflow has been augmented to integrate registries into daily activities. CHOC was able to focus on local variations in care to reduce frequent ED visits and hospital admissions. It was able to track improvements in compliance with registry metrics over time. “With other specialty registries, we can choose metrics that we can measure for improvement,” he said. After the data is mapped, the registry measures are validated against source data and credibility is established with providers, Feaster added.

During the same webinar, Marc Overhage, M.D., Ph.D., chief medical informatics officer in population health for Cerner, said the company is focused on the concept of an evolving longitudinal care plan as a way to inject intelligence into the care continuum. Such a plan requires notifications and actions to be shared among a team of providers, caregivers and a patient. Cerner is analyzing which have to be human tasks and which a computer or automation can do well. A machine might be able to notify the care team of a new patient in a care management program or identify negative trends in the data in order to free up human time to do higher-value tasks, Overhage said. “We have begun to create virtual assistants,” he said. Cerner is focusing on how to translate those activities using FHIR resources into clinical and patient engagement processes that already exist. There is a lot of work to be done to standardize how intelligence is incorporated into the work flow, he added.

Here is an example from CHOC: It is building an alerting program for asthma that takes advantage of automation and machine learning. Data sources such as remote monitoring device data and environmental data and social determinants will be imported into the system. Alerts are constructed and fed into the patient portal and care management platform to preemptively alert patients and caretakers.

An example would be a poor air quality warning triggering automatic alerts to the parent of a child with asthma. The alert through the portal and texts might say that Maria has a history of problems on poor air quality days and ask her to use a peak-flow meter to measure it. The reading might put her in the yellow zone of her asthma action plan, and a care manager is alerted to help her with care, and an appointment is scheduled automatically.

After talking about some ways CHOC is applying data science to improve quality and patient satisfaction, Feaster said there are some limitations to doing data science work using discrete data in the enterprise data warehouse. “We are limited to the data we think might be appropriate, rather than letting the machine learn from much larger data sets,” he said.

CHOC is working as a development partner with Cerner on  HealtheDataLab, a big-data environment spun up in Amazon Web Services (AWS). “There is no limit to the computing resources. Our data scientist will begin in July working full time in that environment,” he said. Two early projects involve patient deterioration in the hospital and patients with no-show tendencies in primary care. The data elements are modelled on HL7 FHIR standards as applicable, and other data sets can be loaded into the same AWS “bucket” for analysis without performance and size limits, he said.

Feaster wrapped up with another quote, this one from George E.P. Box: “All models are wrong, but some are useful.”

“That is true in general, he said. “Nothing is perfectly right, but we are using data and data science to improve care.”




Get the latest information on Health IT and attend other valuable sessions at this two-day Summit providing healthcare leaders with educational content, insightful debate and dialogue on the future of healthcare and technology.

Learn More