Using raw data from the entirety of a patient’s electronic health record, Google researchers have developed an artificial intelligence network capable of predicting the course of their disease and risk of death during a hospital stay, with much more accuracy than previous methods.
The deep learning models were trained on over 216,000 deidentified EHRs from more than 114,000 adult patients, who had been hospitalized for at least one day at either the University of California, San Francisco or the University of Chicago. For those two academic medical centers, the AI predicted the risks of mortality, readmission and prolonged stays, as well as discharge diagnoses, by ICD-9 code.
The network was 95% accurate in predicting a patient’s risk of dying while in the hospital—with a much lower rate of false alerts—than the traditional regressive model—the augmented Early Warning Score—which measures 28 factors and was about 85% accurate at the two centers. The researchers’ findings were published last month in the Nature journal npj Digital Medicine.
The algorithm incorporated the entire EHR, including free-text clinical notes from doctors and nurses and other less-structured data, which bought in more than 46 billion pieces of data for predictions made at discharge.
Rather than working to explicitly harmonize the EHR data and then mapping it on to variables in a statistical model, the researchers instead developed a generic data processing pipeline that takes raw data as input and produces outputs in a more-flexible EHR data structure called FHIR, for Fast Healthcare Interoperability Resources.
Typically, with other methods, up to 80% of the work in predictive models goes toward making the data presentable, the paper's Co-author Nigam Shah, an associate professor at Stanford University, told Bloomberg.
The Google system, on the other hand, would make it relatively easy to deploy to a new hospital, the researchers wrote, although models trained on a particular site’s data may not be immediately transferable to a new site.
To address the clinicians’ wariness of asking the “black box” of a neural network to simply spit out a diagnosis, Google’s AI illustrates the data points it pulled from to come to its conclusion, including parts of the patient’s medical history, radiology findings or a provider’s notes.
The researchers believe that more accurate predictions can lower healthcare costs, while fewer false alerts could reduce alarm fatigue for physicians and nurses. However, they called for more prospective trials and future research to demonstrate the AI’s clinical usefulness and scalability, noting their retrospective study’s limitations.
Google’s AI programs have previously partnered with U.K. hospitals and the National Health Service for mapping radiotherapy for head and neck cancer patients and analyzing eye scans for evidence of age-related macular degeneration and sight loss resulting from diabetes. In 2016, the NHS planned to apply Google’s algorithms to 1 million anonymous optical coherence tomography scans.