Kaiser Permanente is exploring the use of artificial intelligence to cull through doctors’ medical reports and help identify patients with aortic stenosis, a common heart valve disease, and other chronic health conditions.
As part of ongoing research into the predictors of valvular heart disease, researchers looked at whether they could train AI software to spot which patients have valvular heart disease by studying echocardiogram reports, according to Matthew Solomon, M.D., Ph.D., a physician researcher with the Kaiser Permanente Division of Research and a cardiologist.
According to the team’s research findings, published in Cardiovascular Digital Health Journal, a computer taught to intelligently recognize certain abbreviations, words and phrases was able to read through nearly a million electronic health records and echocardiograms from within Kaiser Permanente and identify 54,000 patients with aortic stenosis.
“In our case, the large data set was a giant trove of echocardiogram reports that were collated over the past decades,” said Solomon, lead author of the study.
The goal of the project was to improve care for patients with valvular heart disease, Solomon said.
Grants from the Permanente Medical Group Delivery Science and Applied Research and Physician Researcher Programs supported the work of the researchers.
Solomon noted how complex written text can appear when it is computerized, so natural language processing (NLP) algorithms helped medical professionals make sense of the data. Kaiser uses IQvia’s Linguamatics platform to construct the NLP algorithms.
“It’s a complicated problem because doctors write their reports quite differently,” Solomon explained. “They don’t always write them in the same way, and as we all know, doctors like to use a lot of abbreviations. So there was a lot to teach the computer.”
In addition, by using AI, the researchers were able to reduce the amount of time to cull a million EHRs down to mere minutes, Kaiser Permanente said in a blog post. It would take doctors years to read through this volume of medical records, according to the health system. NLP algorithms can scan through medical records just as Google can scan web pages, Solomon notes.
For the study, Solomon and his team studied the echocardiograms of Kaiser Permanente Northern California patients from January 2008 through December 2018. The team used about 1,000 of the echocardiogram reports to teach the NLP algorithms to understand various ways that doctors describe aortic stenosis and other heart findings.
NLP helped researchers overcome the limitations of procedure codes or diagnosis codes, according to the study’s senior author Alan Go, M.D., a senior research scientist at the Division of Research and the regional director of the Kaiser Permanente Northern California Clinical Trials Program.
“There are a lot of limitations to only using procedure codes or diagnosis codes to identify populations of patients with a condition of interest,” Go said in the blog post. “We were able to train the computer to do what a physician or trained abstractor would do, but on a large scale and without ever getting tired or making mistakes. Importantly, we were also able to train the system to look at the information and measurements from the exams to tell us not only whether a patient had aortic stenosis, but the severity of their condition.”
Healthcare organizations are ramping up their use of AI and NPL to analyze clinical data and aid in decision-making at the point of care. The American College of Cardiology (ACC) is planning a trio of studies that will measure whether personalized clinical guideline support delivered by an AI tool at the point of care can improve heart patients’ outcomes.
Mayo Clinic recently launched a new initiative to collect and analyze patient data from remote monitoring devices and diagnostic tools and to use artificial intelligence to accelerate diagnoses and disease prediction.
Healthcare executives at hospitals, life sciences companies, health plans, and employer organizations say they are accelerating or expanding their AI deployment timelines in response to the pandemic, according to an Optum survey published in November. For hospitals, AI is being used primarily to improve reimbursement coding, monitor the Internet of Things (IoT), and accelerate research, the survey found.
The Kaiser Permanente research team intends to train the computer to identify other types of heart conditions. If this work is successful, the next step would be to teach the computer how to analyze patterns in the medical records that could identify patients at risk for aortic stenosis, which would boost the use of AI for actual disease prevention, according to Kaiser Permanente researchers.
In the future, AI tools such as NLP will help medical professionals identify who is sick or at risk for a certain disease so they can improve treatment. They will learn from the work the team has done with people with hypertension and diabetes, according to Solomon.
“In the future AI techniques might be able to identify patients who are at risk for chronic diseases, so we can intervene earlier and prevent them, rather than simply identifying those who are already sick, so we can make sure we take better care of them,” he said.