5,089 Patients Study Shows Signal Changes Predict Severe Arrhythmias

A study of 5,089 patients across 50 Spanish hospitals found tiny electrical variations in cardiac signals recorded by implantable devices can anticipate severe arrhythmias in heart failure patients.

Tiny electrical variations in cardiac signals recorded by implantable devices can now anticipate severe arrhythmias in heart failure patients.

The Centro Nacional de Investigaciones Cardiovasculares Carlos III (CNIC) developed a cardiac electrical signal analysis model. Thousands of heart failure patients currently carry an implantable cardioverter-defibrillator to detect and treat dangerous arrhythmias, though the shocks they produce carry consequences.

The research, published in Circulation: Arrhythmia and Electrophysiology on November 20, 2026, shows that progressive electrical signal changes transmitted remotely to hospitals make it possible to dynamically determine the risk of ventricular fibrillation.

“Unlike other algorithms, our model is grounded in the pathophysiology of heart failure, so its predictions carry a clinical meaning that physicians can interpret,” explained Dr. David Filgueiras-Rama, a cardiologist at the Hospital Clínico San Carlos in Madrid, head of the CNIC Advanced Development in Arrhythmia Mechanisms and Therapy group and coordinator of the study.

Neural network risk scoring

Researchers tested their initial hypothesis in experimental animal models fitted with an implantable defibrillator. As heart failure progressed, the local electrical signal weakened and ventricular activation slowed. In the experimental phase, hemodynamic and electrical alterations of heart failure decompensation significantly increased the risk of inducible ventricular fibrillation episodes compared with sham-operated controls at 87.5% versus 0%, respectively.

External validation was performed in a second cohort of 2,040 patients from 7 centers using devices from a different manufacturer, analyzing 369,845 tracings.

The model generates a risk score from 0 to 100 that updates with each transmission by tracking signal evolution and premature ventricular complexes. In patients who experienced ventricular fibrillation, the score rose in the months before the event while remaining stable in others.

Computational human ventricle models

Computational-clinical approaches have also been utilized to elucidate ventricular tachycardia and fibrillation mechanisms in heart failure. That study evaluated 53 heart failure patients and 18 control patients.

For patients with low-amplitude action potential voltage alternans k-scores above 1.7, complex action potential duration oscillations accurately predicted ventricular tachycardia and fibrillation during long-term follow-up (+82%; −90% predictive values).

In the failing human ventricular model, abnormal sarcoplasmic reticulum calcium handling caused action potential voltage alternans greater than 1 millivolt during pacing with a 550-millisecond cycle length, that transitioned into large magnitude greater than 100-millisecond discordant repolarization time alternans at faster rates, initiating ventricular tachycardia and fibrillation at cycle lengths under 400 milliseconds by steepening apicobasal repolarization at 189 milliseconds per millimeter until unidirectional conduction block and reentry.

Cohort Group Patient Count Participating Centers
Training Cohort 5,089 50 Spanish hospitals
Computational Study 53 HF / 18 control Clinical-computational models

Clinical monitoring integration

“The tool uses information that defibrillators already send to the hospital, so it could be incorporated into remote monitoring platforms without the need to implant another device, which could help bring forward checkups or adjust treatment during periods of higher risk,” Dr. Jesús Diz Díaz, a CNIC researcher and joint first author on the paper together with Dr. José María Lillo Castellano of the Interhospital Foundation for Cardiovascular Research, said.

Specialists could receive automatic alerts when a patient profile shows a marked increase in risk, enabling earlier clinical decision-making. The study confirms that necessary predictive information is already present in daily remote transmissions.

Frequently asked questions about cardiac electrical monitoring

How do implantable defibrillators transmit data?

Devices transmit cardiac electrical activity recordings periodically and remotely to hospital monitoring platforms without requiring additional surgical procedures, utilizing data signals already generated daily.

What range does the risk scoring model use?

The convolutional neural network generates a risk score ranging from 0 to 100 that automatically updates with each new transmission received by the hospital.

Founded in 1999 by the Spanish Ministry of Health, the Centro Nacional de Investigaciones Cardiovasculares Carlos III (CNIC) is an autonomous center within the Instituto de Salud Carlos III. A 2005 agreement with leading private companies and foundations created the Pro-CNIC Foundation to finance its activities, and purpose-built facilities were inaugurated in February 2006.

Its research is organized into three departments — Vascular Biology and Inflammation, Cardiovascular Development and Repair, and Epidemiology, Atherothrombosis and Imaging — working closely with a Translational Platform. The center was recognized by the Severo Ochoa Centers of Excellence Program in 2011.

How many hospital centers participated in the validation cohorts?

The training model utilized data from 5,089 patients across 50 Spanish hospitals, while independent validation analyzed 2,040 patients across seven centers.