Using electrocardiogram (ECG) results, demographic data, and six standard lab tests, scientists trained machine learning models to estimate a patient’s mortality risk. This new AI-based learning system could assist medical professionals in making quicker, more precise medical decisions.
Predicting mortality risk could assist healthcare providers in streamlining and prioritizing patient care and treatment. However, current methods for diagnosing patients and determining treatment plans are restricted to interpreting test results such as ECGs, X-rays, and bloodwork.
Now, Canadian researchers have trained machine learning programs to read and analyze data from these tests in order to estimate a patient’s risk of mortality. Their findings, published in npj Digital Medicine, demonstrate that these artificial intelligence-based models can quickly identify patients who are at high risk for short- or long-term mortality — at the bedside of the patient.
To examine the performance of machine learning in predicting mortality, the research team developed and trained machine learning programs using 1.6 million ECGs from 244,077 Canadian patients between 2007 and 2020.
The machine’s algorithm accurately predicted mortality risk for each patient at one month, one year, and five years with an accuracy of 85 percent. In addition, the models classified patients into very low, low, medium, high, and very high risk categories.
Incorporating demographic information such as age and sex and six standard blood test results — hemoglobin, glomerular filtration rate (GFR), troponin I, creatinine, sodium, and potassium — improved the accuracy of mortality predictions.
In a news release, lead researcher Padma Kaul states, “These findings demonstrate how machine learning models can be used to transform routinely collected clinical data into knowledge that can be used to enhance decision-making at the point of care within a learning healthcare system.”
According to the study, the team’s future investigations will focus on incorporating additional labs, such as AST, ALT, and Hba1c, into the machine learning models. Additionally, they would like to fine-tune the models for specific patient subgroups and concentrate on predicting cardiac mortality. In addition, the team intends to investigate the clinical applicability of the mortality risk assessment models.
