In the Journals

AI with ECG may predict sex, estimate physiologic age

The use of artificial intelligence with ECGs predicted a patient’s sex and estimated physiologic age, which may be able to measure overall health, according to a study published in Circulation: Arrhythmia and Electrophysiology.

“Being able to more accurately assess overall health status may help doctors determine which patients they should examine further to determine if there are asymptomatic or currently silent diseases that could benefit from early diagnosis and intervention,” Suraj Kapa, MD, assistant professor of medicine and director for augmented and virtual reality innovation at Mayo Clinic, said in a press release. “For people at large, an AI-enhanced electrocardiogram could better show there may be something going on such as a new health issue or comorbid condition that they were otherwise unaware of.”

Researchers utilized data from 774,783 patients (mean age, 59 years; 52% men) who had at least one 10-second 12-lead ECG to train a convolutional neural network. Of these patients, 399,750 were in the training set, 99,977 were in the internal validation set and 275,056 were in the holdout testing set. In addition, 100 patients with several ECGs were randomly selected to assess the ability of the convolutional neural network to estimate physiologic age.

The model was able to classify sex with 90.4% accuracy and an area under the curve of 0.97. The estimation of age was performed as a continuous variable with an average error of 6.9 years (R2 = 0.7).

Of the 100 patients with multiple ECGs, 51% had an average error of less than 7 years between their real age and what the convolutional neural network predicted. Patients with a predicted physiologic age greater than 7 years from their chronologic age often had hypertension, low ejection fraction and coronary disease (P < .01).

There were 27% of the patients in whom the correlation was greater than 0.8 between the physiologic age predicted with the convolutional neural network and their chronological age. These patients had no incident events during 33 years of follow-up.

“It could be hypothesized that the [convolutional neural network]-predicted age may not just reflect preexisting comorbidities or the impact of incident events, but that temporal progression of [convolutional neural network]-predicted age relative to chronologic age per individual ECGs obtained over time could serve as a predictive biomarker for future events,” Kapa and colleagues wrote. “Thus, future population-based research focusing on the potential for risk stratification based on the relative progression of [convolutional neural network]-predicted age compared with chronologic age is also necessary.” – by Darlene Dobkowski

Disclosures: The authors report no relevant financial disclosures.

 

The use of artificial intelligence with ECGs predicted a patient’s sex and estimated physiologic age, which may be able to measure overall health, according to a study published in Circulation: Arrhythmia and Electrophysiology.

“Being able to more accurately assess overall health status may help doctors determine which patients they should examine further to determine if there are asymptomatic or currently silent diseases that could benefit from early diagnosis and intervention,” Suraj Kapa, MD, assistant professor of medicine and director for augmented and virtual reality innovation at Mayo Clinic, said in a press release. “For people at large, an AI-enhanced electrocardiogram could better show there may be something going on such as a new health issue or comorbid condition that they were otherwise unaware of.”

Researchers utilized data from 774,783 patients (mean age, 59 years; 52% men) who had at least one 10-second 12-lead ECG to train a convolutional neural network. Of these patients, 399,750 were in the training set, 99,977 were in the internal validation set and 275,056 were in the holdout testing set. In addition, 100 patients with several ECGs were randomly selected to assess the ability of the convolutional neural network to estimate physiologic age.

The model was able to classify sex with 90.4% accuracy and an area under the curve of 0.97. The estimation of age was performed as a continuous variable with an average error of 6.9 years (R2 = 0.7).

Of the 100 patients with multiple ECGs, 51% had an average error of less than 7 years between their real age and what the convolutional neural network predicted. Patients with a predicted physiologic age greater than 7 years from their chronologic age often had hypertension, low ejection fraction and coronary disease (P < .01).

There were 27% of the patients in whom the correlation was greater than 0.8 between the physiologic age predicted with the convolutional neural network and their chronological age. These patients had no incident events during 33 years of follow-up.

“It could be hypothesized that the [convolutional neural network]-predicted age may not just reflect preexisting comorbidities or the impact of incident events, but that temporal progression of [convolutional neural network]-predicted age relative to chronologic age per individual ECGs obtained over time could serve as a predictive biomarker for future events,” Kapa and colleagues wrote. “Thus, future population-based research focusing on the potential for risk stratification based on the relative progression of [convolutional neural network]-predicted age compared with chronologic age is also necessary.” – by Darlene Dobkowski

Disclosures: The authors report no relevant financial disclosures.