Machine learning algorithms that can aid doctors in more quickly finding data in a patient’s health record have advanced thanks to research. According to research, even when a doctor has been taught to use an electronic health record (EHR) system, it can still take more than eight minutes to locate the answer to just one query.
Machine learning algorithms are being developed by many researchers to speed up the process by automatically locating the data that doctors require in an EHR. But effective models must be trained on large datasets containing pertinent medical questions, but these might be difficult to find because of privacy laws. To address this data shortfall, MIT researchers collaborated with medical professionals to explore the inquiries doctors make when looking over electronic health records (EHRs). Then, they created a dataset that is open to the public and contains more than 2,000 clinically pertinent questions created by these medical professionals.
“Two thousand questions may sound like a lot, but when you look at machine-learning models being trained nowadays, they have so much data, maybe billions of data points. When you train machine-learning models to work in health care settings, you have to be really creative because there is such a lack of data,” says lead author Eric Lehman, a graduate student in the Computer Science and Artificial Intelligence Laboratory (CSAIL).
“Realistic data is critical for training models that are relevant to the task yet difficult to find or create,” says Peter Szolovits, a Professor in the Department of Electrical Engineering and Computer Science (EECS), and the lead author. “The value of this work is in carefully collecting questions asked by clinicians about patient cases, from which we are able to develop methods that use these data and general language models to ask further plausible questions.”
The MIT researchers collaborated with practising physicians and final-year medical students to develop their dataset. More than 100 EHR discharge summaries were provided to these medical professionals, and they were instructed to look through each one and ask any questions they might have. Additionally, they asked the medical professionals to point out the specific “trigger text” from the EHR that prompted each query. For instance, the trigger language “prostate cancer” would prompt the expert to inquire about the date of diagnosis or previous treatments.
They used their dataset of test questions and related trigger text to train machine learning models that would generate new questions based on the trigger text. Then the medical experts determined whether those questions were “good” or not. The researchers discovered that it was able to come up with a good question 63 percent of the time. The team is motivated by the impressive early results they showed with this dataset, even if there is still much work to be done before that model may become a reality.