Over the last several years, the core features of EHRs have stabilized somewhat, with industry tools like HIMSS EMRAM Scale helping to lay out what constitutes an orderly accumulation of features and standard add-ons.
That being said, as newer data management approaches emerge in the IT world, we’re beginning to see the outlines of the next generation of EHR features. One example of what’s on the way comes from Google, which seems to be planning to roll out a new model relying on AI to present physicians with not only a patient’s existing health data but also predictions about their future condition.
According to an item in Politico’s Morning eHealth column, the U.S Patent Office has published a 40-page application for a new invention which “addresses a pressing question facing the physician in the hospital, namely, which patients have the highest need for my attention now.”
The new system would include a data storage element in which providers would aggregate health data from millions of patients, a deep learning engine analyzing those records and an interface which displays not only past health information but also predictions about future clinical events.
When queried by Politico about the new platform, Google had nothing to add other than directing the editor to a journal article published last year. The paper offers some juicy details about the direction of Google’s work, however.
In their research, a group including Google scientists and colleagues from three other universities used FHIR to format data and make retrospective outcome predictions, including in-hospital mortality, 30-day unplanned readmission, prolonged length of stay and all of a patient’s final discharge diagnoses.
Using deep learning, the team was able to predict multiple medical events from multiple centers without having to extract, clean, harmonize and transform relevant variables within those records. This obviously saved the team a ton of work. (It also suggests that hospitals and health systems might be able to duplicate this feat, which is intriguing.)
In a blog posting, Google admits that its project is at a very early stage. That’s clearly the case. But that doesn’t mean that its efforts aren’t noteworthy. What stood out for me, in particular, was the extent to which Google researchers were able to blaze ahead using FHIR as a data standardization tool.
In other settings, after all, it hasn’t been as easy for AI researchers to harmonize data gathered from multiple organizations. For example, a recently published study found that when AI tools are put to work crawling through imaging data across multiple health systems, it’s harder to do than when they stick to a single system’s data.
As interesting as the idea of presenting predictive information is, I haven’t heard of any health system which is doing this across all patients yet, though there are more than a few that have launched projects identifying patients that might acquire costly, dangerous conditions such as prolonged ventilation or sepsis. However, I have little doubt that such predictive capabilities will be standard at some point. Will Google be the one to deliver them? Anything is possible at this point.