Machine Learning Programs Predict Mortality Risk by Analyzing Results from Routine Hospital Tests

Individuals having high blood pressure or symptoms of heart disease, such as chest pain, shortness of breath or an irregular heartbeat generally visit a hospital or an emergency department. In such cases, a clinician usually orders an electrocardiogram, or ECG - a standard test in which tiny electrodes are taped to the chest for checking the heart’s rhythm and electrical activity. Hospital ECGs are mostly read by a doctor or nurse at the patient’s bedside, but now researchers are applying artificial intelligence (AI) to gather additional information from those results to improve patients care.

A research team at University of Alberta (Edmonton, Alberta, Canada) has developed and trained machine learning programs using a massive dataset of 1.6 million ECGs performed on 244,077 patients spanning over a period from 2007 till 2020. The algorithm predicted the risk of death from all causes within one month, one year, and five years with an impressive 85% accuracy rate, ranking the patients into one of five categories, ranging from the lowest to the highest risk. The algorithm's precision was substantially enhanced when demographic information such as age and sex, along with the results of six standard laboratory blood tests (creatinine, kidney function, sodium, troponin, hemoglobin, and potassium) were incorporated into the analysis.

This study serves as a proof-of-concept for utilizing routinely collected data to enhance individual care, enabling the healthcare system to “learn” on the go. The initial phase of the study examined ECG results of all the patients. However, the research team aims to refine these predictive models to cater to specific subgroups of patients. In the subsequent phases, the study will also focus on forecasting heart-related causes of death. The researchers highlight the immense advantage of employing high-powered computing as it can simultaneously view the patterns in a multitude of data points.

“These findings illustrate how machine learning models can be employed to convert data collected routinely in clinical practice to knowledge that can be used to augment decision-making at the point of care as part of a learning health-care system,” the researchers concluded in the study.

Related Links:
University of Alberta