Researchers from the University of Utah and Intermountain Primary Children’s Hospital recently developed artificial intelligence tools to help predict the onset and outcomes of cardiovascular disease.
In a study published this week in PLOS Digital Health, the scientists explained that they used machine learning software to mine the de-identified electronic health records of millions of patients.
They then identified the effects of comorbid conditions and demography on cardiovascular health.
“We can turn to AI to help refine the risk for virtually every medical diagnosis,” said Dr. Martin Tristani-Firouzi, the study’s corresponding author and a pediatric cardiologist at U of U Health and Intermountain Primary Children’s Hospital, in a statement accompanying the research.
“The risk of cancer, the risk of thyroid surgery, the risk of diabetes – any medical term you can imagine,” he added.
WHY IT MATTERS
The researchers note that using data science methods on EHRs has potentially broad applications – but that technological challenges still exist.
For instance, they note, room for improvement exists when it comes to teasing apart the impacts of comorbidities and demographic variables on patient health.
Using a comorbidity discovery method called Poisson Binomial based Comorbidity (PBC), the team searched 1.6 million patient records for comorbid diagnoses, procedures and medications.
“The result is a disease network, devoid of Protected Health Information, that is well-suited for powering downstream outcomes research,” they wrote in the study.
However, they observed, more computational heft is necessary to calculate the contributions of multiple, conditionally dependent variables on an outcome.
That’s where Probabilistic Graphical Models, or PGMs, come in.
“PGMs are capable of answering a prediction query for any variables conditioned on any set of inputs included in the model,” the researchers wrote.
The research team leveraged these technologies to focus on comorbidities around three areas: heart transplant, sinoatrial node dysfunction and congenital heart disease.
“Our results illuminate the comorbid and demographic landscapes surrounding these key cardiovascular outcomes in the U.S. intermountain west, and demonstrate how our approach can inform health care disparities with precise, quantitative results in the context of a specific healthcare system,” they wrote.
Among adults, the team zeroed in on several predictors, including an 86 times higher risk of heart transplantation among individuals diagnosed with cardiomyopathy and a 59 times higher risk for those diagnosed with viral myocarditis.
The strongest individual predictor of heart transplant was the use of the medication milrinone, which carried a 175-fold greater risk.
“Note that we are not suggesting milrinone causes heart transplant – rather that the prescription of milrinone in a patient’s medical record is a powerful predictor of future heart transplant,” they wrote.
Some combined morbidities had a greater risk: A cardiomyopathy patient requiring milrinone has a 407-fold increased risk for heart transplant, they found.
In addition, the scientific team also examined differences among populations.
For example, they found, in the University of Utah Hospital system, a Hispanic patient with atrial fibrillation has a 61-fold increased risk of sinoatrial node dysfunction, compared to 30-fold risk for white patients and 40-fold risk for Black patients.
“These results underscore the potential of our approach to inform ethnic [and] racial healthcare disparities with precise, quantitative results, and in the context of a specific healthcare system,” they wrote.
THE LARGER TREND
The sheer amount of information in EHRs – especially in large health systems – makes them ripe targets for analysis.
In May 2021, Derek Baird of Sensyne Health noted that researchers and companies can use AI to collect, store and analyze large data sets at a far quicker rate than by manual processes.
“This enables them to carry out research faster and more efficiently, based on data about genetic variation from many patients, and develop targeted therapies effectively,” he told Healthcare IT News.
ON THE RECORD
“No matter how aware you are, there’s no way to keep all of the knowledge that you need in your head as a medical professional in this day and age to treat patients in the best way possible,” said Mark Yandell, senior author of the study, in a statement.
“The computational machines we are developing will help physicians make the best possible patient care decisions, using all of the pertinent information available in our electronic age. These machines are vital to the future of medicine,” he added.