Design and methods of the AUTOMATED-WCT trial: evaluating machine learning–based ECG support for WCT interpretation

Abstract

Background

Distinguishing wide complex tachycardia (WCT) as ventricular tachycardia (VT) or supraventricular WCT (SWCT) is critical yet challenging. Manual ECG algorithms require substantial expertise and are inconsistently applied, and contemporary computerized ECG interpretation (CEI) systems often return only a generic “wide complex tachycardia” label. Novel machine learning–based ECG models (Solo Model, Paired Model) can provide a VT probability or a direct VT/SWCT classification, but they have not yet been evaluated in a prospective, randomized, workflow-integrated trial.

Design

We will conduct a prospective, multicenter, investigator-initiated, open-label, four-arm randomized reader trial. Physicians (attendings and fellows in cardiology, emergency medicine, critical care) will be randomized 1:1:1:1 to: (1) Control #1—WCT ECG only; (2) Control #2—WCT ECG + baseline ECG; (3) Solo Model—WCT ECG + model output (no baseline ECG); (4) Paired Model—WCT ECG + baseline ECG + model output. Each participant will interpret 20 adjudicated WCT ECGs on a secure virtual platform, classify rhythm, rate confidence and percieved usefulness, and indicate likely next steps in clinical management. Primary endpoint: WCT classification accuracy. Secondary endpoints: sensitivity, specificity, PPV, NPV, F1 score, time to diagnosis, interpreation confidence, perceived usefulness, and intended management after diagnosis.

Conclusion

The AUTOMATED-WCT Trial will be the first randomized, multicenter evidence on machine learning–based ECG decision support for WCT differentiation.