Abstract TH121: Evaluating the Performance of AI-Driven photoplethysmography(PPG) Models for Cuffless Blood Pressure Monitoring: A Meta-Analysis

Introduction:

Cuff-based manometry has been widely used to obtain continuous monitoring of blood pressure (BP). However, it is associated with some inaccuracies like white-coat hypertension, masked hypertension etc. Therefore a desire for continuous and accurate BP measurement technique is required. One of the recent cuffless BP measurement technique using machine-learning and artificial intelligence (AI)-based technology can estimate BP by extracting BP-related features from photoplethysmography (PPG)-based waveforms. However, the accuracy of cuffless PPG-based BP methods have not been explored. Through this meta-analysis, we aim to compare the PPG-based BP models accuracy with the conventional methods.

Methods:

We systematically searched on PubMed, Cochrane library, and IEEE Xplore for studies providing data on continuous cuffless PPG-based BP monitoring for variations. Included studies provided data on performance metrics like mean absolute difference (MAE), Mean difference (ME) for systolic BP (SBP) and diastolic BP (DBP). The AI-driven technique using PPG was compared with either Medical Information Mart for Intensive Care (MIMIC) or Ambulatory Blood Pressure (ABP) methods. The Mantel-Haenszel method was used to assess the continuous variables at a random effect model to calculate the pooled mean difference value at a 95% CI using OpenMeta Analyst software for analysis. P<0.05 was considered to be significant.

Results:

A total of 15 studies providing data of 20,562 patients was included to evalute the perfomance of PPG-based cuffless BP estimation model compared with the conventional techniques. The pooled MAE for systolic BP (SBP) was 7.557 mmHg (95% CI: 6.421 to 8.693; P<0.001), while the MAE for diastolic BP (DBP) was significantly lower at 3.052 mmHg (95% CI: 2.548 to 3.556; P<0.001). In terms of directional bias, the pooled ME for SBP was 1.277 mmHg (95% CI: –0.194 to 2.749; P=0.089) and the pooled ME for DBP was 0.264 mmHg (95% CI: 0.039 to 0.490; P=0.021).

Conclusion:

The findings suggest that the PPG-based BP model performed better for DBP estimation than SBP which could be due to the result of less beat-to-beat variability for DBP measurement. ME slightly overestimated SBP although was statistically insignificant. While ME did overestimate DBP, the estimated ME was <1 mmHg, hence the clinical bias is negligible according to the Association for the Advancement of Medical Instrumentation (AAMI) standards.