Abstract Body (Do not enter title and authors here): Background: Cuffless blood pressure (BP) monitoring devices with the ability to continuously and noninvasively collect data could transform the screening and management of hypertension. While some wearable devices can estimate BP, their accuracy and reliability remain limited.

Objective: To evaluate the feasibility of using a wrist-worn photoplethysmography (PPG) device (Lifeleaf®) in combination with a supervised learning algorithm to infer clinically useful BP measurements.

Methods: A total of 87 patients undergoing electrophysiology procedures were enrolled (Table 1). Time-synchronized BP data from an intra-arterial line served as the reference. These data were divided into (1) a training set to develop a convolutional neural network (CNN) and (2) a testing set to assess the algorithm’s predictive performance. Data from 60 patients were used to train the model, and the remaining 27 patients were used for validation.

Results: Data from 81 of 87 patients were included in the final analysis. The mean age was 63.6 ± 10.5 years, and 69.1% were male. A total of 6,977 paired PPG and arterial BP readings were analyzed. The mean absolute deviation for systolic BP was 13.08 ± 17.22 mm Hg, and for diastolic BP, 8.42 ± 10.56 mm Hg. Subgroup analysis revealed reduced correlation in patients with BP >140/90 mm Hg, likely due to fewer training samples in this range.

Conclusion:
This study demonstrates the feasibility of extracting clinically meaningful BP measurements from a PPG device using a supervised learning algorithm. The use of arterial line BP as the reference standard adds methodological rigor. Future work will focus on enhancing model performance, particularly in the hypertensive cohort.