Deep learning analysis of 12-lead electrocardiograms for bloodstream infection prediction: a multi-center validation...

Area of Science:

• Artificial Intelligence in Medicine
• Cardiology
• Infectious Diseases

Background:

• Prompt diagnosis of bloodstream infections (BSIs) is crucial for effective antimicrobial stewardship.
• Current diagnostic methods, relying on blood cultures, often involve delays of 48 hours or more.
• This delay hinders timely treatment and contributes to antimicrobial resistance.

Purpose of the Study:

• To develop and validate a deep learning model for predicting the risk of BSIs.
• To utilize standard 12-lead electrocardiograms (ECGs) as the input data for the prediction model.
• To assess the model's performance in identifying clinically significant BSIs.

Main Methods:

• A retrospective, multi-center study trained a deep learning model on 163,826 ECGs from adult emergency department patients.
• The model was validated in three independent cohorts: internal (22,205), external (22,709), and MIMIC-IV (66,937).
• Study adhered to Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) guidelines.

Main Results:

• The deep learning model achieved areas under the curve (AUCs) of 0.821, 0.800, and 0.756 in the test cohorts.
• At a high-sensitivity threshold, the model identified 19-25% of patients as low risk with a 99.8% negative predictive value (NPV).
• The model demonstrated superior discrimination compared to XGBoost and logistic regression baselines and stratified 90-day mortality risk.

Conclusions:

• Deep learning analysis of standard 12-lead ECGs can predict clinically significant BSIs with high accuracy.
• The model offers discrimination superior to conventional ECG parameters, aiding early risk stratification.
• Prospective validation is recommended prior to clinical implementation.