VFRisk clinical algorithm developed using artificial intelligence and machine learning

VFRisk clinical algorithm developed using artificial intelligence and machine learning

Currently left ventricular ejection fraction is the only widely used parameter for prediction of sudden cardiac arrest manifesting as ventricular fibrillation or pulseless ventricular tachycardia. These are also known as shockable rhythms as they are amenable to revival with a direct current shock, unlike asystole and pulseless electrical activity. It has been shown that ventricular fibrillation constitutes 93% of shockable rhythms while pulseless VT contributes only 7%. A new clinical algorithm called VFRisk has been derived by Chugh SS et all using machine learning and artificial intelligence techniques using a large community database of about 1 million persons [1].

Authors analysed lifetime clinical records, ECG and echocardiogram which were available from hospital databases. Genetic analysis was also done using blood samples drawn during venous or intraosseous access by emergency medical personnel at the time of resuscitation of out-of-hospital cardiac arrest. Those with documented VF or pulseless VT constituted 33% of the total cases of out-of-hospital sudden cardiac arrest. Total number of cases was 1374 during the period 2002-2019. There was also a control group of 1600 persons with about 70% coronary artery disease prevalence for comparison.

Prediction models were constructed from a training dataset using backwards stepwise logistic regression. It was then applied to an internal validation dataset. External validation was performed in a geographically distinct population of around 850,000. VFRisk algorithm constructed with 13 clinical, electrocardiogram, and echocardiographic variables had very good discrimination in the training dataset. It was successfully validated in the internal and external datasets. The algorithm substantially outperformed reduced left ventricular ejection fraction of 35% or less as a risk predictor. It also performed well across the entire spectrum of left ventricular ejection fraction.

Authors opined that these findings could enhance primary prevention of sudden cardiac arrest, especially in patients with mid-range or preserved left ventricular ejection fraction. It is a common observation that primary prevention ICDs are giving diminishing returns over the years with increasing number needed to treat due to better guideline directed medical therapy. VFRisk may be considered as a cumulative risk prediction tool specific for avertable sudden cardiac death. If more external validation randomized controlled studies can re-confirm the findings of the authors, this new risk prediction algorithm will be a boon for prevention of SCD.

The important predictors included in the VFRisk algorithm can be divided into various subgroups. Diabetes mellitus was the predictor considered among the cardiovascular risk factors. Factors from prevalent cardiovascular disease were myocardial infarction, atrial fibrillation, stroke and heart failure. Important comorbidities considered were chronic obstructive pulmonary disease, seizure disorder and syncope. ECG parameters were heart rate above 75 per minute, QTc 460 ms or more in females and QTc 450 ms or more in male, Tpeak-end in V5 of 90 ms or more and delayed intrinsicoid deflection of 50 ms or more. Echocardiographic parameter was left ventricular hypertrophy. The parameters were assigned scores from 1.4 to 3.5, with a total score of 25.6. Interestingly, the highest score of 3.5 was for seizure disorder!

References

1. Chugh SS, Reinier K, Uy-Evanado A, Chugh HS, Elashoff D, Young C, Salvucci A, Jui J. Prediction of Sudden Cardiac Death Manifesting With Documented Ventricular Fibrillation or Pulseless Ventricular Tachycardia. JACC Clin Electrophysiol. 2022 Apr;8(4):411-423. doi: 10.1016/j.jacep.2022.02.004. Epub 2022 Mar 30. PMID: 35450595; PMCID: PMC9034059.