Machine learning and feature engineering for predicting pulse presence during chest compressions.

Area of Science:

• Emergency Medicine
• Biomedical Engineering
• Cardiology

Background:

• Current cardiopulmonary resuscitation (CPR) protocols necessitate pausing chest compressions to check for a pulse, which can negatively impact patient survival rates.
• Minimizing interruptions in chest compressions is crucial for effective resuscitation during cardiac arrest.

Purpose of the Study:

• To design and validate an electrocardiogram (ECG)-based algorithm for predicting pulse presence during CPR without requiring pauses in chest compressions.
• To evaluate the algorithm's accuracy in real-time clinical scenarios.

Main Methods:

• Utilized real-time ECG, impedance, and audio recordings from 383 out-of-hospital cardiac arrest patients.
• Developed a predictive algorithm using wavelet transform and principal component analysis on ECG data.
• Trained and tested a linear discriminant model on 60% training and 40% test data splits.

Main Results:

• The algorithm achieved an Area Under the Curve (AUC) of 0.84 for predicting pulse presence with CPR and 0.89 without CPR in the test dataset.
• Approximately 38% of pulse checks revealed a spontaneous pulse.
• Demonstrated moderate accuracy in predicting pulse presence using ECG signals.

Conclusions:

• An ECG-based algorithm shows promise for improving resuscitation efforts by enabling pulse prediction without interrupting chest compressions.
• Further development and validation could lead to enhanced CPR protocols and better patient outcomes.