Data fatigue: Machine learning predicts post-echocardiography survival with EHR assistance
Machine learning can accurately predict survival after echocardiography by analyzing unique data produced from heart images and electronic health record (EHR) information, according to a June 13 study in the Journal of the American College of Cardiology: Cardiovascular Imaging.
As the amount of acquired image data and measurements from images multiplies, Manar D. Samad, PhD, with the department of imaging science and innovation at Geisinger, and colleagues argue “humans may no longer be able to fully and accurately interpret these data.”
“To date, there has been no effort to use machine learning to take advantage of standard clinical and echocardiographic data to help physicians predict outcomes in the broad population of patients who undergo echocardiography during routine clinical care,” they wrote.
Samad and colleagues studied the mortality of 171,510 patients who underwent 331,317 echocardiograms in a large regional health system. They compared the mortality prediction accuracy of a nonlinear machine learning model against a linear logistic regression method.
They used three different inputs as criteria:
- Clinical variables, which included 90 cardiovascular-relevant International Classification of Diseases, Tenth Revision, codes, age, sex, heart rate, blood pressures among others.
- Clinical variables with physician-reported ejection fraction (EF).
- Clinical variables and EF, plus 57 additional echocardiographic measurements.
Results showed the machine learning models were “significantly higher” in their prediction accuracy compared to common clinical risk scores.
Authors discovered it took only 10 variables to achieve 96 percent of the maximum prediction accuracy, and six of those variables were interpreted from echocardiography. The random forest model, which included all echocardiographic measurements was the most accurate.
“Machine learning models can be used to predict survival after echocardiography with superior accuracy by using a large combination of clinical and echocardiography-derived input variables,” authors concluded.
The research also produced what Samad et al. described as a “major” finding. Tricuspid regurgitation jet maximum velocity was determined to be more important than physician-reported left ventricular ejection fraction for predicting patient survival. Additionally, pulmonary artery acceleration time and slope were also among the 10 most valuable variables for mortality prediction.
“These high rankings suggest that measures of pulmonary systolic pressure derived from echocardiography may in fact be more important than previously recognized,” Samad et al. wrote.
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.