Furthermore,
Towards predictive ambulatory ai ventricular:
Retain – Towards predictive ambulatory ai ventricular
- A deep learning algorithm developed from data from an outpatient ECG to a derivation makes it possible to predict the risk of ventricular arrhythmia In the next 13 days.
- Its performance (area under curve, specificity and negative predictive value) are high, on the other hand its positive predictive value remains low.
- Also, if this work opens “ invaluable perspectives “, As indicated the editorial Supporting the publication, there are still important challenges that must be met before seeing a monitoring tool for the ambulatory ECG based on artificial intelligence (AI) integrate practices.
The prevention of sudden death of cardiac origin is traditionally based on the medium. For example, long -term prediction towards predictive ambulatory ai ventricular of potentially fatal arrhythmias, the ejection fraction of the left ventricle being the fundamental parameter used in clinical practice. However, these predictions remain perfectible, in particular to assess the short -term risk. Furthermore, Does the analysis of subtle modifications of the registration of the electrocardiogram (ECG) by AI could help detect signals. However, not identified by humans? French researchers have evaluated it by retrospectively using data from ECG routes to a single derivation recorded in ambulatory in subjects of 18 years. more for 14 consecutive days.
In total. the cohort formed included 247,254 patients (of which 1,104 had a sustained ventricular arrhythmia), who were randomly separated into two groups, one to develop the predictive tool (181,177 subjects), the other to validate it. Finally, an external validation was carried out from a second retrospective cohort comprising 20,497 subjects.
Concretely. the researchers used the first 24 -hour recordings to predict towards predictive ambulatory ai ventricular the risk of sustained ventricular arrhythmia incurred over the following 13 consecutive days. The model was fueled from three data groups: the patient’s socio -demographic data. Additionally, the quantitative measurements calculated from the ECG, a time graph of cardiac frequency density and a selection of ECG wave extracts.
Still perfectible performance
The developed model made it possible to reach an area under the curve (Auroc) of 0.957 [0,943-0,971] and 0.948 [0,926-0,967] in the cohorts of internal and external validation respectively. For a specificity set at 97.0 %, the sensitivity was 70.6 % and 66.1 % respectively. Thus, the model predicted precisely the future occurrence of sustained tachycardia in 80.7 % and 81.1 %, respectively, knowing that 9 out of 10 cases were translated by the occurrence of ventricular fibrillation. No difference was observed by sex, but a slight loss of performance was highlighted with aging.
However. the towards predictive ambulatory ai ventricular value of sensitivity suggests that “ 3 to 4 out of 10 patients with subsequent tachycardia have not been reported as high risk »Underlines the editorial. As for the positive predictive value, it was low due to the scarcity of the event, or 12.3 % [10,6-14,4] et 10,1 % [7,9-12,6] For internal and external validations respectively. That ” constitutes the Achilles heel of this approach “Comments the editorial. false positives in this clinical context inducing a significant risk of” associated anxiety of anticipation and potentially useless interventions ».
Study authors report that reducing the predictive period improves the model: the auroc reaches 0.96 [0,948-0,972] et 0,952 [0,933-0,974] When the risk is predicted at 3 days, and no longer at 13 days, and even 0.970 [0,960-0,982] et 0,961 [0,946-0,966] For a prediction at 1 hour. Ventricular complexes occurring prematurely, towards predictive ambulatory ai ventricular early depolarizations appeared as the most predictive elements within the elements that have fed the model.
It is therefore necessary to consider this deep learning model as a tool allowing to glimpse the development of ” New real -time surveillance tools. which could be integrated into “smart monitoring” systems based on artificial intelligence “And who” could integrate portable devices such as connected watches or implantable recorders ».
Further reading: Made myself up on my menstrual cycle – World Hepatitis Day: Luxembourg intensifies its mobilization against a silent epidemic – Flu epidemic in Reunion: emergencies, medical office and pharmacies taken by storm – Update of the pharmacy medicinal list: Virpax registration – The questions we ask about our diet in summer.
