Smartphone-based photoplethysmographic measurements for diagnosis of cardiac arrhythmia

<p dir="ltr">Background</p><p dir="ltr">Atrial fibrillation (AF) is the most common clinically significant cardiac arrhythmia and is associated with an increased risk of stroke, heart failure, dementia, and all-cause mortality. Early diagnosis and treatment are essential but often delayed due to the intermittent nature of the arrhythmia and limited access to electrocardiogram (ECG)-based diagnostics. Smartphone-based photoplethysmography (PPG) is a non-invasive optical technique using the phone's built-in camera as measurement sensor, with the potential to serve as a scalable and accessible tool for heart rhythm and heart rate assessment in ambulatory settings. This thesis investigated the diagnostic performance and clinical utility of a smartphone-PPG system when used by patients with AF and atrial flutter (AFL) in their home environment.</p><p dir="ltr">Methods and Results</p><p dir="ltr">The studies were prospective, single-centre, and conducted with patients at Danderyd University Hospital in Stockholm. In the studies, the participants performed ambulatory, unsupervised smartphone-PPG recordings (CORAI) simultaneously with single-lead ECG (KardiaMobile) as reference.</p><p dir="ltr">Study I - Prospective validation of arrhythmia diagnostics by manual reading of smartphone PPG. Adults undergoing elective or acute direct current cardioversion (DCCV) for AF/AFL performed simultaneous heart rhythm recordings with PPG and ECG daily for 30 days after DCCV. The reference heart rhythm from ECGs was read by experienced cardiologists (gold standard), and the heart rhythm from PPG reports was read by trained physicians blinded to the heart rhythm from ECGs. Among 280 participants providing 18 005 paired recordings, manual heart rhythm reading from PPG achieved a sensitivity of 99.0% and specificity of 99.7% for diagnosing AF. With AFL recordings included, sensitivity and specificity were 97.7% and 99.4%, respectively.</p><p dir="ltr">Study II - External validation of automatic arrhythmia diagnostics from PPG using a machine learning (ML) algorithm. A support vector machine classifier, trained on 16 092 PPG recordings made by 180 patients in an independent training cohort was externally validated on the recordings from the 280 participants in the cohort in Study I. The ML model for rhythm classification had higher accuracy compared with manual reading in diagnosing AF, with both a sensitivity and specificity of 99.7%. With AFL recordings included, the diagnostic performance decreased slightly but remained high (sensitivity 99.3%, specificity 99.1%). The PPG recordings had a high proportion with sufficient quality for diagnosis, and only a small fraction of recordings was excluded due to either insufficient quality or low algorithmic certainty.</p><p dir="ltr">Study III - Randomised clinical trial (RCT) investigating efficacy and feasibility of pre-cardioversion heart rhythm monitoring with smartphone PPG. In this RCT with patient-blinded allocation, participants scheduled for DCCV were randomised 1:1 to either intervention or control. The heart rhythm from PPG recordings pre-cardioversion was read daily by the investigators. In case of spontaneous conversion to sinus rhythm (SR) or if non-adherence to oral anticoagulation (OAC) treatment protocol were detected in the intervention group, their scheduled DCCV were cancelled or postponed, respectively. No such action was taken for participants in the control group. Among 104 (intervention) and 99 (control) participants, the monitoring strategy using smartphone PPG reduced same-day DCCV cancellations from 23.2% to 4.8%, a relative risk reduction (RRR) of 79.3% and an absolute risk reduction (ARR) of 18.4%. Same-day cancellations due to spontaneous conversion to sinus rhythm were reduced from 18.2% to 1.0%, with a RRR of 94.7% and an ARR of 17.2%. Adherence to twice-daily recordings was high (median 2.1 recordings/day), and 100% of pre-cardioversion PPGs had a sufficient quality to make a heart rhythm diagnosis, based on automatic signal-quality analysis. No same-day cancellations were attributed to OAC non-adherence.</p><p dir="ltr">Study IV - Validation of heart-rate measurements using smartphone PPG. The recordings from the participants in Study III were used to assess the agreement of heart rate from simultaneous PPG and ECG recordings on a per- measurement basis using automatic heart rate algorithms. In 203 participants with 17 588 paired recordings peri-cardioversion, the overall agreement between heart rate from PPG and ECG was high. The mean absolute error was 2.4 beats per minute (bpm), the root mean square error was 4.6 bpm, and the mean heart rate difference was 0.3 bpm. The heart rate agreement was higher for regular heart rhythms, such as SR and AFL, with regular AV conduction, compared with irregular heart rhythms, such as AF and AFL, with variable AV conduction.</p><p dir="ltr">Conclusions</p><p dir="ltr">The studies in this thesis showed that the smartphone-PPG method can be used by patients in ambulatory, unsupervised settings to achieve highly accurate heart rhythm diagnostics of atrial fibrillation and atrial flutter. A machine learning algorithm demonstrated excellent diagnostic accuracy, reducing the need for manual interpretation. Pre-cardioversion monitoring with smartphone PPG significantly reduced same-day cancellations of DCCV by detecting spontaneous conversion to sinus rhythm. Heart rate assessment from PPG recordings showed excellent overall agreement with heart rate derived from ECG recordings. These results support the use of smartphone PPG as an independent diagnostic tool for heart rhythm and heart rate in ambulatory settings, minimising the need for manual review and ECG confirmation.</p><h3>List of scientific papers</h3><p dir="ltr">I. <b>Fernstad J,</b> Svennberg E, Åberg P, Kemp Gudmundsdottir K, Jansson A, Engdahl J.</p><p dir="ltr">Validation of a novel smartphone-based photoplethysmographic method for ambulatory heart rhythm diagnostics: the SMARTBEATS study.</p><p dir="ltr">EP Europace. 2024 Mar 30;26(4):euae079<br><a href="https://doi.org/10.1093/europace/euae079">https://doi.org/10.1093/europace/euae079</a><br><br></p><p dir="ltr">II. <b>Fernstad J,</b> Svennberg E, Åberg P, Kemp Gudmundsdottir K, Jansson A, Engdahl J.</p><p dir="ltr">External validation of a machine learning-based classification algorithm for ambulatory heart rhythm diagnostics in pericardioversion atrial fibrillation patients using smartphone photoplethysmography: the SMARTBEATS-ALGO study.</p><p dir="ltr">EP Europace. 2025 Mar 28;27(4): euaf031<br><a href="https://doi.org/10.1093/europace/euaf031">https://doi.org/10.1093/europace/euaf031</a><br><br></p><p dir="ltr">III. <b>Fernstad J,</b> Svennberg E, Åberg P, Engdahl J.</p><p dir="ltr">Pre-cardioversion heart rhythm monitoring using smartphone PPG - A Randomized Clinical Trial</p><p dir="ltr">[Submitted]</p><p dir="ltr">IV. <b>Fernstad J,</b> Svennberg E, Åberg P, Engdahl J.</p><p dir="ltr">Validation of a method for ambulatory heart rate measurements using smartphone photoplethysmography: the SMARTBEATS-RATE study</p><p dir="ltr">[Manuscript]</p>