I left my smartphone at home and can’t tell if I’m in atrial fibrillation

Abstract

The enormous impact of atrial fibrillation (AF) on human suffering, health-care utilization, and physician resources is clear to the readers of HeartRhythm. Recent publications have developed 3 alarming trends: (1) the incidence of AF is increasing, perhaps fueled by the obesity epidemic; (2) our ability to treat AF is not improving, at least as assessed by ageadjusted mortality rates postdiagnosis; and (3) health-care costs for treatment of AF are growing exponentially. In reaction to these depressing facts, the tendency to want to learn more about our enemy is only natural. Improvements in bioengineering technology and computer science have allowed the development of ever more comprehensive monitoring strategies for AF detection. These improvements have demonstrated the axiom “the more you look, the more you will find” in many different clinical situations—cryptogenic stroke—in patients with sick sinus syndrome requiring pacemaker therapy and following AF ablation. In patients with cryptogenic stroke, the clinical value of more intensive monitoring, although apparently selfevident, has not been established; the relevance of detecting short duration episodes of AF in other clinical scenarios is even less clear. In addition, as will be discussed below, the diagnostic accuracy of many available monitoring techniques is far from ideal. Into this fray comes the present work from McManus et al. These investigators collected mobile phone camerabased recordings of the pulse waveform from the subjects’ index finger in 76 patients with persistent AF before and after cardioversion. The recordings were analyzed by using a peak detection algorithm and subjected to 2 statistical analyses— the root mean square of successive RR intervals (RRSSD/ mean) and Shannon entropy (ShE)—both of which express the unpredictable nature of the ventricular response during AF. Threshold values for these analyses were determined from the authors’ previous work. This study demonstrated that an algorithm combining RRSSD/mean and ShE showed “excellent sensitivity (0.962), specificity (0.975), and