February 06, 2019
The US Patent and Trademark Office on January 31 published Google's patent application for a "deep machine learning" system that uses longitudinal electronic health record (EHR) data to predict future health events.
The clinical decision support system can also help physicians identify patients who are most in need of help and can display the key clinical markers that underlie the predictions, according to Google.
Google announced the concept last May and filed the patent application in July. The Patent Office has not yet granted a patent for the predictive EHR system.
The Google system can aggregate and store EHR data for populations and individual patients. The company uses Health Level Seven's Fast Health Interoperability Resources framework to standardize the data extracted from disparate EHRs, a Google blog post said. Then, without users having to specify the variables of interest, the deep learning model for each prediction "reads all the data points from earliest to most recent and learns which data helps predict the outcome."
Google collaborated on a study of its system with the University of California, San Francisco, Stanford Medicine, and the University of Chicago Medicine. The Google system was able to predict in-hospital mortality, 30-day unplanned readmission, prolonged length of stay, and final discharge diagnoses with an accuracy superior to that of traditional predictive models, the company said.
Google's approach improved on the traditional approach to data aggregation for predictive modeling, the researchers noted. "Importantly, we were able to use the data as-is, without the laborious manual effort typically required to extract, clean, harmonize, and transform relevant variables in those records."
In the patent application, Google said that the predictive system could help physicians prioritize patients and could show which information to look for in a patient's chart. That information could help healthcare providers identify areas of concern or intervene to reduce the likelihood of an adverse event, the application noted.
Rise of the Machines
Deep machine learning and other types of artificial intelligence (AI) are increasingly being used to predict clinical events and refine clinical protocols.
For example, Penn Medicine has used machine learning to redesign care pathways and to successfully predict which patients are likely to develop sepsis, according to CIO.com. Flagler Hospital in St. Augustine, Florida, has used an AI solution to improve and standardize care paths for patients with pneumonia and sepsis, HealthcareITNews reports. And Grady Health System in Atlanta, Georgia, reportedly saved $4 million by preventing readmissions using a combination of AI-driven predictive analytics and patient-level community interventions, according to Mobile Health News.
Whether or not such tools will catch on among physicians depends largely on their accuracy. Commenting on the Penn Medicine program in the CIO.com article, Dean Sittig, PhD, a professor at the University of Texas Health Sciences Center in Houston, said that the predictions must be right more than half the time, or clinicians won't pay attention to them.
The trial of Google's predictive EHR system, according to an article in Nature Partner Journals: Digital Medicine, processed 46.8 billion data points collected from 216,221 adult patients hospitalized for at least 24 hours in two academic medical centers.
Using a scale in which 1.00 is perfect and 0.50 is no better than random chance, the models used in the study scored 0.86 in predicting whether patients would have a lengthy stay in the hospital, 0.95 in predicting inpatient mortality, and 0.77 in forecasting unexpected readmissions.
Health Solutions From Our Sponsors
SOURCE: Medscape, February 06, 2019.