Novel disease-electrocardiogram associations in inherited arrhythmia syndromes

点击翻译按钮获取中文摘要

ABSTRACT Electrocardiograms (ECGs) have been used for more than a century to detect the electrical activity of the heart. ECGs are used to screen and diagnose patients with inherited arrhythmia syndromes, diseases that can result in cardiac arrhythmias and sudden cardiac death. The ECG is considered an important part of the screening and diagnostic armamentarium of IAS, because it is inexpensive, portable, provides point-of-care results and does not require highly skilled personnel to perform. However, from the standpoint of interpretation, the ECG does not yield a sensitive and specific result and therefore it fails to serve as an accurate screening or diagnostic tool for IAS. Part of this inaccuracy derives from the assessment of too few individuals to generate the normal reference ECG values, with more than 100 age and gender-dependent variables and cut-off values to memorize, all resulting in foundational deficiencies and a very high inter-observer interpretation variability. We have created a novel ECG database from the largest historical cohort of normal individuals of more than 27,000 subjects. We then transformed the data of 102 ECG variables to express the values as Z-scores. Z- scores by definition facilitate an immediate and objective distinction of normality and abnormality across all measures. Expressing the ECG values in Z-scores eliminates inter-observer variability in the interpretation of ECG values. In addition, we developed sophisticated computer algorithms enhanced by artificial intelligence (AI) to detect characteristic traits of ECG variables attributable to a group of subjects. In this study we will collect ECGs from patients with IAS. Next, we will compare these ECGs to our ECG database of normal individuals utilizing the Z-score based nomograms. We will use statistical analysis to detect differences in the 102 ECG variables between the affected (IAS) and unaffected (normal) subjects. We will identify the ECG variables that show the most promising distinction characteristics for an IAS disease entity. Next, we will use AI algorithms to detect highly sensitive and specific combinations of ECG variables. We will apply three different models on the digitized ECG data. First, we will quantify dependencies between ECG variables with a combination of principal components regression and graphical LASSO algorithms. This approach will automatically identify the best combination of ECG variables to differentiate between affected and normal individuals, and will develop a set of variables that can be used to provide the most sensitive and specific disease-ECG associations for specific IAS to date. We will then use two distinct machine learning models to detect anomalies and pattern of novelties in the ECG of subjects with IAS. With the combination of traditional statistical analysis and the AI based algorithms, we will be able to identify specific ECG variables or groups of ECG variables and their Z-score values to serve as predictive tools for the diagnosis of IAS. Our long-term goal is to utilize this model for large scale screening efforts to detect IAS in the young and thereby prevent catastrophic complications, such as sudden cardiac death.