RITHMI is a prospective, nonrandomized study conducted at Hospital Universitario y Politécnico La Fe in Valencia, Spain between January and June 2019 to analyze the ability of the RITHMI wrist-wearable heart rhythm monitor to detect AF in patients older than 18 years referred to a cardiology outpatient clinic for cardioversion of AF or for a general consultation (patients with sinus rhythm or AF). We excluded patients with atrial flutter or an implanted pacemaker.

The study was carried out in accordance with the principles of the Declaration of Helsinki and was approved by the ethics committee of our hospital. All the patients provided signed informed consent agreeing to participate. A reference standard 12-lead ECG interpreted by 2 electrophysiologists was performed in all patients to determine whether they had sinus rhythm or AF.

After the 12-lead ECG, the RITHMI heart rhythm monitor was placed on the patients’ wrists and used to record their heart rhythm for 3minutes while they were seated. Patients seen for a general cardiology consultation continued to wear the wrist device, while those who underwent cardioversion did not.

The wrist-wearable RITHMI heart rhythm monitor is a not yet commercially available prototype that is equipped with a dual-detection system involving 2 types of signals: heart rhythm signals detected by PPG in the distal region of the forearm/wrist and a single-lead ECG signal incorporated into the device that is equivalent to the I lead in 12-lead ECG. The PPG sensor (ADPD107) consists of 2 light-emitting diodes and a receiver on the underside of the device. The ECG sensor (AD8233) has 3 electrodes: 2 on the upper side of the device, which are placed on the index finger and thumb, and 1 on the underside of the device that is in contact with the skin. The position of the individual is designed to close the electrical circuit between the arms and provide an ECG recording equivalent to that provided by a standard I lead (figure 1). One of the electrodes on the upper side of the ECG sensor provides earthing to the circuit and helps eliminate electrical interference. The monitor has a low-consumption Bluetooth microcontroller and a data storage memory device.

Figure 1.

Position of heart rhythm monitor and placement of thumb and index finger to record the electrocardiogram.

(0.03MB).

The RITHMI monitor has 2 automated algorithms—a PPG and an ECG algorithm—programmed to detect AF by statistically analyzing signal irregularities. The PPG sensor is designed for continuous, passive heart rhythm monitoring, while the ECG sensor is used to confirm the presence of arrhythmia and generate an exportable graph for further analysis.

Continuous variables are expressed as mean±SD and categorical variables as absolute values and percentages. The chi-square test was used to compare categorical variables and the t test to compare continuous quantitative variables.

The main aim of this study was to evaluate the ability of the RITHMI monitor to detect AF by comparing the diagnoses provided by the PPG and ECG algorithms with that provided by the reference standard 12-lead ECG interpreted by the 2 electrophysiologists. The evaluation was double-blinded: electrophysiologists were unaware of the results offered by the RITHMI heart rhythm monitor, while the engineer responsible for entering these results into the database was unaware of the reference standard diagnosis. At the end of the study, the diagnosis provided by each of the 2 algorithms was compared with the 12-lead ECG diagnosis.

The following variables were calculated for the PPG and the ECG algorithm to assess the ability of the RITHMI monitor to detect AF: sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (sensitivity/(1-specificity), and negative likelihood ratio (1-sensitivity/specificity). We also calculated the percentage of correctly diagnosed cases, defined as the number of cases correctly diagnosed as AF plus the number of cases correctly diagnosed as sinus rhythm divided by the total number of cases and multiplied by 100.

The κ statistic was used to measure the level of agreement between the 2 automated AF detection systems (PPG and ECG) and between each of these systems and the reference standard 12-lead ECG.

The PPG signal was uninterpretable in 5 of the 167 patients due to poor signal quality. In the remaining 162 patients, the algorithm diagnosed AF in 95 patients (figure 2A) and sinus rhythm in 67 (figure 2B).

Figure 2.

RITHMI heart rhythm monitor output for a patient with atrial fibrillation (A) and sinus rhythm (B). ECG, electrocardiogram; PPG, photoplethysmography.

(0.1MB).

The PPG algorithm showed a sensitivity of 91% and a specificity of 96% for the detection of AF compared with the reference standard 12-lead ECG. In addition, it had a PPV of 97%, an NPV of 88%, a diagnostic accuracy of 93%, a positive likelihood ratio of 20.22, a negative likelihood ratio of 0.05, a false-positive rate of 3%, and a false-negative rate of 9%. Subsequent analysis showed that the false positives were due to poor signal quality, with artifacts caused by sudden movement. The false negatives were attributed to variations in heart rate during arrhythmia that were below the threshold to be classified as an arrhythmic pulse. These variations were usually due to a high heart rate (mean, 121±32 vs. 68±14 bpm for patients with AF correctly identified by the device, P=.015).

The diagnostic ability of the RITHMI ECG algorithm was evaluated in 133 patients. It diagnosed AF in 87 patients and sinus rhythm in 46. Based on the reference standard diagnosis, it had a sensitivity of 94%, a specificity of 96%, a PPV of 98%, an NPV of 90%, a diagnostic accuracy of 95%, a positive likelihood ratio of 21.68, a negative likelihood ratio of 0.01, a false-positive rate of 2%, and a false-negative rate of 6%. The false positives were explained by signal artifacts due to poor contact between the electrodes and the skin resulting in abrupt changes in the electrical signal, while the false negatives were attributed to variations in heart beat during arrhythmia that were below the threshold to be classified as an arrhythmic pulse.

Overall, the PPG and ECG algorithms concurred in their diagnosis of AF or sinus rhythm in 96% of cases (κ=0.91). They produced discrepant results for either AF or sinus rhythm in 5 cases due to signal artifacts caused by movement of the patient during PPG recording or by poor contact between the sensor and the skin during ECG recording.

Overall (both algorithms provided the same diagnosis), the RITHMI monitor had a sensitivity and a specificity of 95% for the diagnosis of AF, a PPV of 99%, an NPV of 100%, a diagnostic accuracy of 95%, a positive likelihood ratio of 19.09, and a negative likelihood ratio of <0.1. The level of agreement between the diagnosis provided by the RITHMI monitor and the reference standard 12-lead ECG diagnosis was κ=0.98.

The European Society of Cardiology guideline recommends opportunistic pulse-taking to detect silent AF in people older than 65 years (class I recommendation, level of evidence B) and systematic ECG screening in people older than 75 years or patients with a high risk of stroke (class IIa recommendation, level of evidence B).1

The technology described in this study could improve the diagnosis of AF in at-risk populations as it allows for continuous, passive monitoring via a wrist-wearable device. Before it can be widely used, however, it is first necessary to overcome the limitations associated with poor signal quality (noise artifacts, use of single-lead ECG recordings) and, above all, to confirm our findings in the general population, where there is a lower prevalence of AF and a higher likelihood of false positives. One recent study of over 400 000 people without a history of AF who had a smartphone and smartwatch with a dedicated AF application reported irregular pulse notifications consistent with AF in 0.52% of the overall population and in 3.2% of people older than 65 years.16 An ECG performed a mean of 13 days after the detection of irregularities confirmed the presence of AF in 34% of cases.

The possibility of using smartphone or smartwatch technology to detect AF could revolutionize the conventional model used to detect this disorder, blurring the distinction between patient and consumer and almost certainly leading to an increase in the demand for healthcare services and probably the number of people diagnosed with AF. At the same time, accessibility issues make it difficult to predict the role that this technology will play in the future and will also change the profile of people who will benefit from monitoring. Finally, studies are needed to investigate the usefulness of anticoagulant treatment in patients with AF detected by smart technology.

The main limitation of the study is linked to the conditions in which the PPG and ECG signals were obtained, as it is easier to obtain an adequate signal (particularly in the case of the PPG sensor) in patients at rest and in environments free of external stimuli. The limitations associated with measuring electrical rhythms captured by PPG sensors are described in the literature, and include the influence of tremor or motion on the signal-to-noise ratio.10–13 All the patients in our series were white, so we were unable to evaluate the potential influence of skin pigmentation on signal quality and algorithm performance. The PPG algorithm was associated with a false-negative rate of 9% and most of the false negatives were due to a high heart rate. Other studies have not described heart rate or have excluded patients with higher rates13,14 because of their effect on automated AF detection algorithms.

Overall, we were unable to analyze the performance of the PPG algorithm in 3% of patients due to poor signal quality signal. This rate, however, might have been higher in a less controlled environment. Patients with atrial flutter or a pacemaker were excluded from our study, as they may have a regular pulse (the RITHMI algorithms work by analyzing irregularities).

The ECG algorithm was analyzed in just 133 patients as it was incorporated into the RITHMI monitor at a later data than the PPG algorithm. It is unlikely, however, that earlier incorporation would have significantly modified our results.

Another limitation of our study is our small sample size. It would be desirable to confirm our results in a larger population exposed to different conditions (ie, not only in patients at rest or in a controlled environment). Further studies are needed before the RITHMI monitor can be validated for use in an AF detection campaign. Nonetheless, our findings do confirm proof of concept and serve as a starting point for designing a study to assess the usefulness of this wrist-wearable device in the general population.