Data resources and patient characteristics.

This study was performed using the standard 12-lead surface ECG data of one South Korean tertiary teaching hospital with 1,103 beds. The study protocol was approved by the Ajou University Hospital Institutional Review Board. All ECGs performed from 1 June 1994 to 31 July 2013 were included in the database. All numerical parameters were calculated using Marquette™ 12SL algorithms (versions 7, 13, and 22) developed by GE Healthcare. There were no restrictions with respect to comorbidities or prescribed drugs. The database contained 979,273 ECGs from 461,178 patients (Table 1). An average of 2.1 ECGs per patient were recorded in the database (S2 Table). A total of 188,823 patients received sequential ECGs; of those, 119,768 underwent more than two ECGs in 2 years (S1 Fig). The median and mean durations between the ECG recordings were 340 and 633 days, respectively (S3 Table). Among those patients, 119,768 patients had more than 2 ECG records within 2 years (S1 Fig). The median and mean number of drug prescriptions given to patients who had had at least two ECGs in a 2-year period were 9 and 20.18, respectively (S4 Table and S2 Fig). The average age of the patients was 42.6 ± 19.2 years, and male patients comprised 50.1% of all patients. The proportion of patients with South Korean ethnicity was 98.8%.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Demographic and clinical characteristics of the ECG-ViEW II database population.

https://doi.org/10.1371/journal.pone.0176222.t001

ECG data extraction.

The RR interval, QT/QTc interval, PR interval, QRS duration, P wave axis, QRS axis, and T wave axis values were extracted from each ECG data source. There were three sources from which the ECG data could be extracted: paper ECGs, digitalized ECG records in the ECG management system (MUSE; GE Marquette, Milwaukee, WI), and the EHR system. First, numeric ECG values on printed ECGs were copied from the previous ECG-ViEW database, which was originally extracted using optical character recognition. Second, we extracted PDF files from the ECG management system via the web-parsing software that was used in our previous work (but which was modified to extract additional parameters beyond those included in the previous version) [10]. All files in the system from 1 June 1994 to 31 July 2013 could be extracted automatically. Using the Java platform, the ECG parameters in the extracted PDF files were saved as text files. Finally, from 4 March 2010 to 31 July 2013, ECGs were recorded in a digitalized format in the EHR system of the target hospital. These records were simply transferred to our database (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Schematic illustration of ECG-ViEW II construction.

An overview of the database development process. All numeric ECG parameters were extracted from three ECG sources (ECG management system, EHR system, and ECG printouts) from the subject hospital. After remove duplicated ECGs, the ECG data were integrated with clinical data, validated with QTc data and de-identified. EHR, electronic health record; ECG, electrocardiogram; ECG-ViEW II, Electrocardiogram Vigilance with Electronic Data Warehouse II; QTc, heart-rate-corrected QT

https://doi.org/10.1371/journal.pone.0176222.g001

The ECG data extracted by these three methods were merged into a single database after removal of duplicate data. The paper ECG data contained only the RR interval and QT/QTc interval. ECGs from both the ECG management system and the EHR contained the RR interval, QT/QTc interval, PR interval, QRS duration, P wave axis, QRS axis, and T wave axis. Thus, when the ECG management system data were duplicated with the EHR ECG data or paper ECG data, the ECG management system data remained and the other data were removed. Additionally, when the EHR data were duplicated with the paper ECG data, the EHR data remained and the paper ECG data were removed.

The ECG measuring devices in target hospital used Bazett’s formula (QTc = QT / [RR^0.5]) to calculate the QTc interval. For this reason, the database contains QTc interval values calculated with Bazett’s formula. Other QTc data, based on Fridericia’s formula (QTc = QT / [RR^0.33]) or Framingham’s formula [QTc = QT + 0.154 (1 − RR)], could be easily calculated using the QT and RR intervals provided in the database.

Extraction of demographic characteristics and clinical information.

Using the EHR of the target hospital, we integrated the following demographic and clinical data into the ECG database: age, sex, ethnicity (Korean or non-Korean), prescribed medications, diagnoses, and selected serum electrolyte concentrations (potassium, magnesium, and calcium). The count of diagnosis included in the database is presented in the S5 Table. The observation period of a given patient was defined as the period from 1 year before the first ECG examination to 1 month after the last ECG examination. The age-adjusted Charlson comorbidity index was calculated with ICD-10 diagnosis codes, dating from 1994 to the date of the ECG.

De-identification.

All unique identifiers were excluded to meet the US Health Insurance Portability and Accountability Act privacy rule. Outliers in the laboratory test results were replaced with a value at the 99.5^th percentile (top-coded). Highly stigmatized diagnoses (S6 Table), such as infertility, congenital malformation, sexually transmitted diseases (including HIV infection), and chromosomal abnormalities, were removed (215,725 diagnoses [2.7%] from 5,642 patients [1.2%]). Specific drugs that could be associated with, and used to identify, a specific person (especially antiviral agents for AIDS/HIV) were also removed. The examination date, drug prescription date, and diagnosis date were shifted by adding randomly assigned value within specific range (range: −90 to 90 days; mean, 0.0; standard deviation, 51.9). Thus, although an individual patient would not be identifiable using this information, the intervals between the dates for each individual patient were conserved. The birth years were grouped according to 5-year intervals (S7 Table).

Discovery of QT interval prolongation.

Our previous ECG-ViEW database was developed to provide information on QT interval prolongation and associated factors [10]. The database is already being used to evaluate adverse drug reactions that result in QT interval prolongation. After examining the association of the QTc interval with certain drugs, Yun et al. [11] suggested that famotidine administration can prolong the QTc interval and increase the proarrhythmic potential. Additionally, Park et al. [13] concluded that selective serotonin reuptake inhibitors are less likely to be associated with QTc interval prolongation. The ECG-ViEW II contains previous ECG-ViEW data and an additional 270 thousand ECGs, allowing researchers to study the relationship between the QT interval and drug administration. According to Bazett’s formula, a QTc of > 450 ms in men, and > 470 ms in women, in considered abnormal[1]. Using these criteria, QTc prolongation was seen in 69,647 ECGs in men (n = 33,551) and 39,478 ECGs in women (n = 21,009) in the database (Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Number of electrocardiograms and patients with abnormal ECG parameters.

https://doi.org/10.1371/journal.pone.0176222.t003

Discovery of conduction abnormality.

The PR interval represents the time from the onset of atrial depolarization to the onset of ventricular depolarization. A shortened PR interval (< 120 ms) may be associated with Wolff-Parkinson-White syndrome or junctional rhythms. A prolonged PR interval (> 200 ms) may indicate a first-degree heart block, which is associated with a significant risk of atrial fibrillation (relative risk = 1.45), heart failure with left ventricular dysfunction (relative risk = 1.39), and mortality (relative risk = 1.24) [12]. The ECG-ViEW II contains data from 28,798 patients with a shortened PR interval and 21,381 patients with a prolonged PR interval (Table 3). The database also contains data on serum electrolytes, abnormalities of which can deteriorate PR interval prolongation. Analysis of PR interval data with electrolyte values allows researchers to study conduction abnormalities, heart failure, etc.

Discovery of diseases associated with cardiac axis deviation.

ECG-ViEW II contains more information on the axis of P wave, QRS, and T wave than does our previous database. QRS axis deviation is the most important axis information. A normal QRS axis ranges from −30 to 90 degrees. A QRS axis of > 90 degrees is defined as right axis deviation. Right ventricular hypertrophy, chronic lung disease, lateral wall myocardial infarction, dextrocardia, and left posterior fascicular block can be associated with right axis deviation. A QRS axis less than −30 degrees is defined as left axis deviation, which can be associated with left ventricular hypertrophy, left bundle branch block, inferior wall myocardial infarction, Wolff-Parkinson-White syndrome, ventricular pacing/ectopy, and ostium primum atrial septal defect. The ECG-ViEW II database includes 39,637 ECGs showing right axis deviation (25,528 patients), and 27,173 showing left axis deviation (12,222 patients) (Table 3).

The normal P wave axis ranges from 0 to 75 degrees. Various studies have evaluated P wave axis deviations. Rangel et al. [14] showed that an abnormal P wave axis is correlated with atrial fibrillation. In addition, Acar et al. [15] showed that the P wave axis was significantly increased in patients with systemic lupus erythematosus and positively correlated with the SELENA-SLEDAI score. Another study showed that an abnormal P wave axis was associated with a 55% increased risk of all-cause mortality [16]. The ECG-ViEW II contains 88,819 ECGs with an abnormal P wave axis (61,467 patients) (Table 3). The database also contains all diagnoses of patients during the observation period. Researchers could evaluate the correlation of the P wave axis and diagnosis history.

The T wave axis was categorized as normal (15° to 75°), borderline (> 75° to 105° or −15° to < 15°), or abnormal (−180° to < −15° or > 105° to 180°). There was a nearly two-fold increased risk of coronary heart disease death, and an approximately 50% increased risk of incident coronary artery disease and all-cause mortality, for those with marked T wave axis deviation [17]. Moreover, T wave axis deviation is correlated with metabolic syndrome, low-grade systemic inflammation, and left ventricular hypertrophy in patient with diabetes mellitus or pregnancy [18–21]. ECG-ViEW II contains 41,258 patients with an abnormal T wave axis and provides an opportunity to evaluate the correlation of many diseases with the T wave axis parameter.

Discovery of cardiac disease in patients with a wide QRS complex.

The QRS complex represents depolarization of the ventricles. A normal QRS complex duration is ≤ 120 ms in adults and ≤ 80 ms in children. An abnormal QRS complex may result from one of the following: an interventricular conduction disturbance, aberrant ventricular conduction, ventricular pre-excitation, and ventricular dysrhythmia. The ECG-ViEW II contains 27,680 ECGs with a wide QRS complex, from 12,096 patients (Table 3), thereby providing valuable information for researchers to investigate diseases associated with a wide QRS complex.

Discovery of correlation of ECG parameters with echocardiography parameters.

An echocardiography database (EchoDB) was made after the previous ECG-ViEW database was opened to the public. It includes approximately 100,000 echocardiography data parameters from 72,399 patients. Among the parameters included in the EchoDB are the ejection fraction, left ventricular end diastolic and systolic dimensions, left atrium size, interventricular septal end diastolic and end systolic thicknesses, left ventricular mass index, ratio of mitral velocity to early diastolic velocity of the mitral annulus, and right ventricle systolic pressure. The ECG-ViEW II can be studied more thoroughly when combined with the EchoDB. If any researchers wish to evaluate the ECG data along with echocardiographic parameters, they can contact the ECG-ViEW II team in the Department of Biomedical Informatics in Ajou University (abmi@ajou.ac.kr). Researchers are required to send a study proposal (containing the study design) and complete the application form on the ECG-ViEW II website (send by email). After examining the submitted proposal and obtaining permission from the institutional review board of the target hospital, members of the ECG-ViEW II team analyze the data according to the researcher’s proposal.