Diagnostic Accuracy of a Smartphone-Operated, Single-Lead Electrocardiography Device for Detection of Rhythm and Conduction Abnormalities in Primary Care

Jelle C.L. Himmelreich, MD, MSc; Evert P.M. Karregat, MD, MSc; Wim A.M. Lucassen, MD, PhD; Henk C.P.M. van Weert, MD, PhD; Joris R. de Groot, MD, PhD; M. Louis Handoko, MD, PhD; Robin Nijveldt, MD, PhD; Ralf E. Harskamp, MD, PhD

Disclosures

Ann Fam Med. 2019;17(5):403-411. 

In This Article

Methods

We reported this diagnostic accuracy study in accordance with the Standards for Reporting of Diagnostic Accuracy Studies (STARD) 2015 statement.[13] The study protocol was approved by our institution's Medical Ethical Review Committee. All participants provided written informed consent.

Study Design

We enrolled consecutive patients as part of the Validation of a mobile bedside ECG Screening and diagnostic Tool for Arrhythmias in general practice (VESTA) study. Eligible patients were aged 18 years or older who were assigned to 12L-ECG for any nonacute indication as ordered by the local primary care physician in 1 of 10 participating general practices across the Netherlands. The practices were in possession of a 12L-ECG device and had qualified and skilled personnel to perform 12L-ECG. Exclusion criteria were a clinically acute indication for ECG as defined by the local primary care physician (eg, suspicion of acute coronary syndrome) and presence of a pacemaker rhythm on 12L-ECG. We categorized patients according to indication for 12L-ECG either because of presentation with new symptoms (symptom-driven ECG) or as an integral part of protocolized care for primary or secondary prevention of cardiovascular disease (protocol-driven ECG). For each participant, the study design involved 3 different readings as follows: (1) the 1L-ECG read by the AF detection algorithm of the smartphone application, (2) the 1L-ECG read by cardiologists, and (3) the standard 12L-ECG read by cardiologists.

Index Test

The KardiaMobile (AliveCor, Inc) is a smartphone-connected, 1L-ECG device that displays ECG recordings in real time (30 seconds) via a smartphone application with a built-in AF detection algorithm (Figure 1). The 1L-ECG recordings were assessed in 2 ways as follows:

Figure 1.

The KardiaMobile and Kardia smartphone application.
Photograph by Jelle Himmelreich.

1. The AF detection algorithm assessed all 1L-ECG recordings. It classified recordings as either possible AF, normal, or unreadable, or provided no classification. We marked all recordings classified as possible AF as positive for AF. We marked all other algorithm classifications, or when no classification was provided, as negative for AF. The algorithm did not provide a classification for when a 1L-ECG recording was truncated (<30 seconds).

2. Cardiologists (M.L.H., R.N., J.R.dG.) assessed all 1L-ECG recordings in randomized order. The evaluation consisted of scoring each recording for the presence of arrhythmias, ectopic beats, and conduction abnormalities according to a scoring template designed for this study (see Supplemental Appendix, http://www.AnnFamMed.org/content/17/5/403/suppl/DC1/, for exact definitions).

Reference Standard

All 12L-ECG recordings were independently evaluated by 2 cardiologists, and in case of disagreement, by a third cardiologist (M.L.H., R.N., J.R.dG.). We presented 12L-ECG recordings to the cardiologists in randomized order. The evaluation consisted of scoring each recording for the presence of arrhythmias, ectopic beats, and conduction abnormalities according to a scoring template designed for this study (see Supplemental Appendix, http://www.AnnFamMed.org/content/17/5/403/suppl/DC1/, for exact definitions).

Rhythm Measurement

Personnel instructed each patient to commence the KardiaMobile recording by holding the device loosely with both hands (corresponding with lead I for 12L-ECG). We advised patients who used hand lotion or had sweaty hands to wash their hands with soap or to use alcohol wipes on the fingertips to optimize electrical conduction quality. When a steady 1L-ECG signal was visible on the smartphone, the local investigator started a 10-second 12L-ECG recording. We thereby obtained 10 seconds of simultaneous recording. We excluded patients for whom 1 or both ECG types were not available or when there was no 10-second overlap between recording types. The 1L-ECG recordings were not used for clinical decision making.

Data Collection

Three investigators (J.C.L.H., E.P.M.K., R.E.H.) visited participating practices to collect the 12L-ECG recordings (as PDF file or photocopy of paper original) as well as patient data at the time of index ECG from the practice's electronic health records. We collected the corresponding 1L-ECG recordings (PDF files) from the secure online platform that is part of the KardiaMobile software package. Baseline data included sex, age, indication for undergoing 12L-ECG, use of relevant antiarrhythmic drugs, and relevant medical history. We pseudonymized all data before storing it in a secured electronic case report form (Castor EDC).

Statistical Analysis

We expressed diagnostic accuracy for all analyses as sensitivity, specificity, positive and negative likelihood ratios, and positive and negative predictive values, with 95% CI. The primary analyses of this study were (1) the diagnostic accuracy of 1L-ECG as assessed by cardiologists in detecting AF or AFL with 12L-ECG as reference and (2) the diagnostic accuracy of the AF detection algorithm for AF/AFL with 12L-ECG as reference. Secondary analyses were (1) the diagnostic accuracy of 1L-ECG as assessed by cardiologists in detecting any rhythm abnormality, defined as any nonsinus rhythm including AF/AFL and/or presence of any ectopic beat, with 12L-ECG as reference and (2) the diagnostic accuracy of 1L-ECG as assessed by cardiologists in detecting any conduction abnormality, defined as presence of atrioventricular (AV) block, bundle branch block (BBB), and/or left axis deviation and/or left anterior fascicular block, with 12L-ECG as reference. We counted the cardiologists' generic assessment of BBB on 1L-ECG as true positive even if specification of subtype of BBB (left BBB or right BBB) was provided by the corresponding 12L-ECG.

We performed an exploratory analysis of the primary and secondary outcomes stratified by whether ECG was performed based on symptoms or as part of protocol-driven care. We performed a sensitivity analysis on the comparison of the AF detection algorithm vs 12L-ECG for the outcome AF/AFL, in which we excluded all patients with a truncated 1L-ECG recording.

We presented discrete variables as number and percentage and continuous variables as mean ± standard deviation. We compared continuous variables using the Student t test and proportions using the Fisher exact test or Pearson χ 2 test and used 2-tailed tests. We evaluated statistical significance in all analyses at the .05 level and performed analyses using IBM SPSS Statistics version 24.0 (IBM Corp) and MedCalc version 18.10.2 (MedCalc Software).

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....