A Study to Evaluate Accuracy and Validity of the "Chang Gung" Ventricular Systolic Dysfunction Screening Software

This is a retrospective study conducted in Taiwan that aimed to test the performance of "Chang Gung" Ventricular Systolic Dysfunction screening software using electrocardiogram (ECG) and echocardiography data. The study data was obtained from a database that includes six hospitals of Chang Gung Memorial Hospital, between January 1, 2006, and December 31, 2019.

The software was developed using a training set of 133,225 data and validated using a set of 57,134 data. For clinical validation, a total of 1,172 test data were randomly selected from the testing set, stratified by hospital classification, age group and gender. The hospital classification, age group and gender ratios were based on the proportion of the testing data. The test data were also stratified by the presence or absence of left ventricular systolic dysfunction(LVSD) for test group and control group, defined as a heart output rate of less than 40% within 14 days before and after the ECG recording.

During the clinical trial, a cardiologist with 15 years of experience in treating cardiovascular disease examined the ECG data without any exclusion criteria. The cardiologist also confirmed the accuracy of the left ventricular ejection fraction (LVEF) measurement, which was defined as LVSD in the echocardiography reports. The LVEF was extracted from legally-binding echocardiography reports, not by the cardiologist during the clinical trial. The ECG data were screened and filtered for quality before being input into the software. The primary outcome was the sensitivity of the software, which was defined as not inferior to 0.86. The study also analyzed secondary outcome measures, including the area under the receiver operating characteristic curve, accuracy, specificity, positive predictive value, negative predictive value, false-positive rate, and false-negative rate.

Inclusion criteria

• Equal or greater than twenty years old.
• Static 12-lead electrocardiogram of General Electric MUSE XML format file.
• The data comes from the static 12-lead electrocardiogram device of General Electric (model MAC5500).
• Echocardiography data within 14 days before or after the electrocardiogram acquisition time.
• The electrocardiogram signal is 500 Hz.
• The Alternating current (AC) filter of the electrocardiogram signal is 60 Hz.
• The length of the electrocardiogram signal is ten seconds (the electrocardiogram output of model MAC5500 is ten seconds, and there is no need to capture a segment).

Exclusion criteria

• Cases used in the model development process.
• Lacks any electrode.
• Contain any electrode lacks a segment.