Pragmatic screening for heart failure in the general population using an electrocardiogram-based neural network

Area of Science:

• Cardiology
• Artificial Intelligence
• Public Health

Background:

• Heart failure remains a significant public health concern with a high burden of underdiagnosis.
• Current screening methods often rely on multiple risk factors and clinical assessments, which can be time-consuming and complex.
• There is a need for a pragmatic and efficient screening tool for heart failure applicable at a general population level.

Purpose of the Study:

• To develop and evaluate a novel screening tool for heart failure using only digital electrocardiograms (ECGs).
• To assess the performance of a convolutional neural network (CNN) in identifying heart failure compared to traditional risk variables and machine learning models.
• To create a tool that can be easily integrated into routine clinical practice for early heart failure detection.

Main Methods:

• A prospective, observational study (Hamburg-City-Health-Study) enrolled 5299 individuals aged 45-75 years.
• A convolutional neural network (CNN) was developed using only digital ECGs to differentiate between individuals with and without heart failure.
• The CNN's performance was compared against logistic regression and random forest models utilizing known heart failure risk variables.

Main Results:

• The study included 5299 participants, with 318 (6.0%) diagnosed with heart failure.
• The CNN achieved an area under the curve (AUC) of 0.75, with a sensitivity of 0.67 and specificity of 0.69.
• The CNN demonstrated comparable predictive accuracy to logistic regression (AUC 0.77) and random forest (AUC 0.79) models, using only ECG data.

Conclusions:

• A screening tool for heart failure was successfully developed using only digital ECGs and a CNN.
• This AI-driven tool requires no additional input beyond an ECG, facilitating seamless integration into general practitioner workflows.
• The developed screening tool has the potential to significantly reduce the underdiagnosis of heart failure in the general population.