Years of practice as healthcare professional

Years of ECG interpretation

Number of weekly interpreted ECGs

I currently use artificial intelligence-based ECG algorithms in clinical practice

What is your profession?

What is your age?

What is your gender?

In your opinion, what level of evidence should be required to authorize the use of an artificial intelligence algorithm in clinical practice?

Appropriate modalities for AI-aided analysis

For which modalities would you consider which Artificial Intelligence-aided analysis of ECG signals appropriate?

(Choose one answer per row)

Appropriate scenarios for AI-aided analysis

For which scenarios would you consider which Artificial Intelligence-aided analysis of ECG signals appropriate?

(Choose one answer per row)

Requirements for automatic ECG systems

What type of visualization of the ECG signal after AI-only analysis of an ambulatory Holter ECG would you prefer in your clinical practice, on top of a text summary and heart rate trends? (Choose 1 answer)

AI-powered algorithms for ECG analysis can still missinterpret some ECG signals.

By increasing the ability to detect arrhythmias (higher sensitivity), the probability of false positive alarms for episodes without arrhythmia rises (lower specificity), and vice versa. Which of the following sensitivities and specificities would you still consider adequate for practical use?

Responsibility

In case of an automatic AI algorithm missing a critical arrhythmia, who is currently responsible?

Responsibility

In case of an automatic AI algorithm missing a critical arrhythmia, who should be held responsible?

Case 1

A 30-year-old female patient presents to your outpatient clinic with a history of recurrent palpitations every few days. After about a minute, the symptoms terminate spontaneously (on-off symptoms). The patient wears a 14-day ambulatory ECG monitor. Its signal is interpreted using a FDA/CE approved AI algorithm, with the following result:

“Continuous sinus rhythm during the whole analysis. No arrhythmias or pauses detected. 36 hours (10.2% of the time worn) not analyzable due to artefact.”

The patient is desperate because she had 4 episodes of palpitations at various timepoints during the time that she wore the device. Full-disclosure ECG signal of the analysis is accessible at request.   What are your next steps? (Choose 1 answer)

Case 2

A 75-year-old patient with coronary artery disease, diabetes mellitus, and hypertension undergoes ambulatory ECG monitoring with AI analysis of the signal, due to previous syncope. The left-ventricular function is normal. During the analysis, the patients recalls a few episodes of presyncope and one episode of transient loss of conciousness. The results of the automatic ECG analysis indicate the following:

“Continuous sinus rhythm with a high rate of ventricular ectopic beats (0.2%), no sustained arrhythmia or pauses. 20 hours (5.9% of the time worn) not analyzable due to artefact.”

What are your next steps? (Choose 1 answer)

Case 3

You are the cardiologist on call at a major hospital. An emergency ambulance team contacts you regarding a 50-year-old male patient presenting with chest pain. He has a history of hypertension and smokes. An artificial intelligence ECG tool interpreted the 12-lead ECG and has recommended immediate coronary angiography. You are told that the colleagues did not detect ECG changes indicative of ST elevation myocardial infarction.

What do you do? (Choose 1 answer)