Cardiac arrhythmias frequently occur in patients admitted to the Coronary Care Unit (CCU). The majority of these patients are treated for an acute myocardial infarction, which carries an increased risk of life-threatening arrhythmias such as ventricular tachycardia (VT) or ventricular fibrillation (VF). This risk is one of the reasons these patients are monitored for 48 hours after a myocardial infarction, in accordance with the guidelines of the European Society of Cardiology (ESC) for acute coronary syndrome. Other arrhythmias, such as asystole, atrial fibrillation, or atrioventricular block, also occur in CCU patients. These arrhythmias are recorded on the electrocardiogram (ECG) monitor in the CCU and trigger an alarm for healthcare staff. However, in order to apply this alarming with sufficient sensitivity, many false positive alarms are also produced, which increases the workload for nurses (alarm fatigue) and undermines patient well-being. This study will investigate whether Artificial Intelligence (AI) models, using continuous ECG data, can help improve the prediction of patients at risk of a life-threatening cardiac arrhythmia. Firstly, this study will aim to predict patients at risk of VT/VF in both the short term (30 minutes) and long term (1 day) in patients under continuous telemetric monitoring. This prediction facilitates timely intervention by the team in the short term, and in the long term, the safe transfer of a patient to a lower-complexity ward or earlier safe discharge of a patient. Secondly, this study will aim for improved detection to reduce the number of false negative alarms and thereby reduce alarm fatigue. The performance of these AI models can be evaluated through this retrospective observational study. Patients aged 18 years or older who have been admitted with acute cardiac disease will be included. The primary objective of this study will be to evaluate the performance of AI models that detect and predict critical arrhythmias in the short and long term, using ECG data obtained via the monitoring system.
Primary objective: Assessment of the performances of AI models in identifying patients at risk of sustained VT and VF from bedside monitor ECG in different timeframes: * 30-minute prediction model * 1-day prediction model Secondary objectives: • Assessment of potential healthcare savings if the AI model in would be used in clinical practice, such as CCU length-of-stay (CCU-LOS), hospital length-of-stay and associated costs Exploratory objectives: * Real time and continuous detection of events for alarming * Prediction of other types of arrhythmias (e.g., Atrial fibrillation (AF), atrioventricular block, severe brady-arrhythmia) using in hospital ECG monitoring. * Identification of clinical risk factors for sustained VT and/or VF * Exploration of development and assessment of new AI models using new (clinical) input
Study Type
OBSERVATIONAL
Enrollment
3,000
Catharina Hospital Eindhoven
Eindhoven, North Brabant, Netherlands
RECRUITINGOccurrence of sustained ventricular tachycardia or ventricular fibrillation
The primary outcome of the study is the occurrence of sustained ventricular tachycardia (VT) (monomorphic and polymorphic with a heartrate \> 100 bpm and duration \> 30 seconds or with hemodynamic compromise such as fainting or need for resuscitation) or ventricular fibrillation. (Binary outcome measure 0 = no event during admission, 1 = event during admission)
Time frame: During admission
Secondary outcome measure
\- A 'textbook' outcome (no adverse events) (Binary outcome measure 0 = no event during admission, 1 = event during admission)
Time frame: During admission
Secondary Outcome Measure
\- In-hospital onset and offset of cardiac arrhythmias (e.g. atrial fibrillation, atrio-ventricular block or severe tachy- or bradyarrhythmia, non-sustained VT). (Binary outcome measure 0 = no event during admission, 1 = event during admission)
Time frame: during admission
Secondary outcome measure
\- In hospital death/cardiovascular in-hospital death (include cause if available) (Binary outcome measure 0 = no event during admission, 1 = event during admission)
Time frame: During admission
Secondary outcome measure
\- Pulseless electrical activity (PEA) and asystole (Binary outcome measure 0 = no event during admission, 1 = event during admission)
Time frame: During admission
Performance of AI prediction model
Discrimination of AI prediction model expressed with Area Under the Receiver Operating Characteristic curve (AUROC), Area Under the Precision-Recall Curve (AUPRC), sensitivity, specificity, (Positive Predictive Value) PPV and (Negative Predictive Value) NPV
Time frame: During admission
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.