News

Creating Smarter Healthcare IT: Putting the Doctor back in the driver’s seat

Published on February 1, 2013


Electronic health records (EHRs) have transformed healthcare by providing doctors with rapid access to clinical data. But too often, healthcare leaders have let the EHR take the wheel, allowing the computer to set priorities rather than allowing doctors to lead. This is obviously a mistake: the EHR has great potential, but it’s not ready to be in the driver’s seat.

EHRs try to give helpful advice, but often they’re just annoying, as in “Golly doc, did you know that narcotic you ordered could make the patient sleepy?” In this study from 2000, the authors note, “In one particular case a physician had to override the same allergy alert… for the same patient 106 separate times.”  There hasn’t been much improvement since, just look at this study, or these.

Most EHRs force doctors to dumb-down documentation, moving them away from information-rich natural language (like speech and free text typing) and toward time-intensive forms filled with dozens of checkboxes.  This may help computers, but it hurts patients by making the documentation less useful to human caregivers, stealing caregiver time away from patients, and  creating harmful delays for patients still waiting to be seen.

EHR medical advice systems are built by human experts who craft rules based on what they know. However, the rules are often too simplistic to be useful, such as “if the potassium level is low, then suggest ordering potassium tablets.” The reason for building such basic rules is due to the fact that checkboxes simply can’t capture enough detail to support sophisticated reasoning.  Experts are also terrible at reproducing the complex inner workings of their own brains. Ask an expert “How did you know that?”  Their answer is likely “experience,” or “I just know.”

Fortunately, machine learning systems have matured dramatically in the past decade, and are much more effective than human experts at creating sophisticated, accurate algorithms. Amazon’s recommendations, Google’s AdSense and email spam filters demonstrate the power of these systems that depend on “Big Data” – huge datasets that machine learning toolsets can use to create excellent prediction algorithms. Cloud-based medical software can provide the “Big Data” needed to power machine learning systems. Machine-learned algorithms actually improve when fed information-rich raw data like natural language documentation.

The original Yahoo! approach to searching the Internet used teams of librarian experts to manually catalog the information on the Web. Yahoo! was quickly usurped by Google’s machine-learning search techniques that scaled massively and generated much better results with far less effort. Medicine is on the cusp of a similar transition.

Cloud-based medical speech recognition enables machine-learning so that every time a doctor dictates a patient chart, the speech-to-text translation engine improves. Cloud-based natural language understanding engines leverage machine-learned algorithms to automatically translate text into codes a computer can understand. Cloud-based machine-learning can be applied to create medical advice systems that can help doctors instead of annoying them.

Doctors have been sitting in the back seat for way too long, and patients have paid the price. But the Cloud may finally provide the EHR with the information it needs to finally earn its learner’s permit.