Photoplethysmography (PPG) finds new applications in cardiology far beyond heart rate and Spo2 measurements, especially in atrial fibrillation classification. Nevertheless, the impact of other hemodynamic and structural heart parameters on PPG signal morphology remains unclear.
This observational study investigated the application of PPG for clustering and phenotyping cardiovascular patients. A cohort of 108 patients (mean age 68.9+/-12.6, 47 females) admitted to a cardiology ward between 2021 and 2023 was analysed. Ten-minute PPG measurements were collected at rest under standardised conditions using red and infrared light with the MAX30102 fingertip sensor. Routine clinical data, including echocardiography were collected. The Symmetric Projection Attractor Reconstruction (SPAR) method was applied to analyse the nonlinear dynamics of PPG signals, with hierarchical clustering used to group signals based on theta density profiles.
Among five PPG-derived clusters, four were selected for further analysis, due to clusters balanced size. Cluster 2 (n=13) had a 69.2% prevalence of atrial fibrillation (AF), an average AcT of 90.75, TAPSE of 16.9 mm, the lowest ejection fraction (EF) of 36.23%, and the highest average age of 76 years. Cluster 3 (n=10) had 60% AF prevalence, AcT of 100.25, TAPSE of 17.4 mm, EF of 44%, and an average age of 73.7 years. Clusters 4 (n=35) and 5 (n=44) were healthier, with AF prevalence of 25.7% and 11.4%, higher AcT (118 and 124), TAPSE (21.2 mm and 22.1 mm), EF (50.3% and 50.9%), and younger ages (68.4 and 65.3 years). All inter-cluster differences were statistically significant (p<0.05).
The exploratory analysis revealed distinct cardiovascular phenotypes, with Cluster 2 showing high AF prevalence, low EF, and older age and Clusters 4 and 5 representing healthier profiles, suggesting PPG's potential for advanced hemodynamic phenotyping.