Performance and feasibility of three different approaches for computer based semi-automated analysis of ventricular arrhythmias in telemetric long-term ECG in cynomolgus monkeys

Computer-based analysis of long-term electrocardiogram (ECG) monitoring in animal models represents a cost and time-consuming process as manual supervision is often performed to ensure accuracy in arrhythmia detection. Here, we investigate the performance and feasibility of three ECG interval analysis approaches A) attribute-based, B) attribute- and pattern recognition-based and C) combined approach with additional manual beat-to-beat analysis (gold standard) with regard to subsequent detection of ventricular arrhythmias (VA) and time consumption. ECG analysis was performed on ECG raw data of 5 male cynomolgus monkeys (1000 h total, 2 x 100 h per animal). Both approaches A and B overestimated the total number of arrhythmias compared to gold standard (+8.92% vs. +6.47%). With regard to correct classification of detected VA event numbers (accelerated idioventricular rhythms [AIVR], ventricular tachycardia [VT]) approach B revealed higher accuracy compared to approach A. Importantly, VA burden (% of time) was precisely depicted when using approach B (-1.13%), whereas approach A resulted in relevant undersensing of ventricular arrhythmias (-11.76%). Of note, approach A and B could be performed with significant less working time (-95% and - 91% working time) compared to gold standard. In sum, we show that a combination of attribute-based and pattern recognition analysis (approach B) can reproduce VA burden with acceptable accuracy without using manual supervision. Since this approach allowed analyses to be performed with distinct time saving it represents a valuable approach for cost and time efficient analysis of large preclinical ECG datasets.

Automated summary

Computer-based analysis of long-term ECG monitoring in animal models is a time-consuming process due to the need for manual supervision for accurate arrhythmia detection. This study investigated the performance and feasibility of three ECG interval analysis approaches in detecting ventricular arrhythmias. Results showed that the combined attribute and pattern recognition analysis approach had higher accuracy in classifying detected events and accurately depicting arrhythmia burden. Additionally, this approach significantly reduced working time compared to manual supervision. Therefore, it is a valuable method for cost-effective and time-efficient analysis of large preclinical ECG datasets.