# Can eHealth programs for cardiac arrhythmias be scaled-up by using the KardiaMobile algorithm?

## TL;DR

This study evaluates the KardiaMobile algorithm's accuracy in detecting heart rhythms in a telemonitoring program, finding it highly reliable for identifying normal rhythm.

## Contribution

The study provides new evidence on the algorithm's diagnostic accuracy in a real-world telemonitoring program for cardiac arrhythmias.

## Key findings

- The algorithm showed high sensitivity (0.956) and specificity (0.985) for detecting sinus rhythm.
- It also demonstrated strong performance for atrial fibrillation detection, though with slightly lower PPV.
- Only a small number of false outcomes remained uncorrected, indicating high reliability.

## Abstract

Remote monitoring devices for atrial fibrillation are known to positively contribute to the diagnostic process and therapy compliance. However, automatic algorithms within devices show varying sensitivity and specificity, so manual double-checking of electrocardiographic (ECG) recordings remains necessary.

The purpose of this study was to investigate the validity of the KardiaMobile algorithm within the Dutch telemonitoring program (HartWacht).

This retrospective study determined the diagnostic accuracy of the algorithm using assessments by a telemonitoring team as reference. The sensitivity, specificity, negative predictive value (NPV), positive predictive value (PPV), and F1 scores were determined.

A total of 2298 patients (59.5% female; median age 57 ± 15 years) recorded 86,816 ECGs between April 2019 and January 2021. The algorithm showed sensitivity of 0.956, specificity 0.985, PPV 0.996, NPV 0.847, and F1 score 0.976 for the detection of sinus rhythm. A total of 29 false-positive outcomes remained uncorrected within the same patients. The algorithm showed sensitivity of 0.989, specificity 0.953, PPV 0.835, NPV 0.997, and F1 score 0.906 for detection of atrial fibrillation. A total of 2 false-negative outcomes remained uncorrected.

Our research showed high validity of the algorithm for the detection of both sinus rhythm and, to a lesser extent, atrial fibrillation. This finding suggests that the algorithm could function as a standalone instrument particularly for detection of sinus rhythm.