How Google is using deep learning to understand EHR data and reduce readmissions

At its Google I/O 2018 conference, CEO Sundar Pichai shared how Google is betting on artificial intelligence to sort through the vast amounts of data in EMRs and predict better health outcomes, according to ZDNet.

Much of healthcare data is unstructured, and Google wants to apply its models and algorithms to fix that. In a paper introduced at the conference May 8, the company described its work with Stanford (Calif.) University, UC San Francisco and the University of Chicago exploring deep learning models as they can be applied to health data. It plans to scan thousands of patient data points, such as medication and lab results, to "quantitatively predict" patients' chances of being readmitted.

The No. 1 challenge for Google is data prep, which has hindered many other analysis efforts. However, Google expects deep learning may hold the key to success.

"We hypothesized that these techniques would translate well to healthcare; specifically, deep learning approaches could incorporate the entire EHR, including free-text notes, to produce predictions for a wide range of clinical problems and outcomes that outperform state-of-the-art traditional predictive models," the paper reads. "Our central insight was that rather than explicitly harmonizing EHR data, mapping it into a highly curated set of structured predictors variables and then feeding those variables into a statistical model, we could instead learn to simultaneously harmonize inputs and predict medical events through direct feature learning."

The models have shown their ability to scale vast amounts of data and accurately predict unexpected readmissions, discharges and inpatient mortality, ZDNet reports, but Google said it's just getting started. The predictions and deep learning models were using retrospective data, but in the future, Google could explore data gleaned from its partnership with Fitbit or that stored on its cloud platform.

More articles on EHRs:
Meditech launches Meditech UK to serve UK, Ireland
Interoperability nonprofit The Sequoia Project takes aim at health IT disaster responseEpic's App Orchard to include Excel Medical's patient surveillance, predictive analytics tools

© Copyright ASC COMMUNICATIONS 2019. Interested in LINKING to or REPRINTING this content? View our policies by clicking here.

 

Top 40 Articles from the Past 6 Months