Ensuring Equality in Blood Pressure Monitoring

Think back to the last time you had your blood pressure measured. You may recall the cuff (also known as a sphygmomanometer) squeezing around your upper arm as it inflated. The results would’ve indicated your blood pressure while you were wearing the cuff, but wouldn’t have been able to detect how your blood pressure changed on your drive home, later that evening, or the following week.

In recent years, interest has grown in using photoplethysmography (PPG) waveform analysis, a noninvasive method to estimating blood pressure that can be done intermittently or continuously with no cuff required. For people with hypertension, this kind of continuous monitoring can be beneficial.

PPG waveform analysis is a promising method, but results can be impacted by individual characteristics such as age and physiology, requiring a demographic-aware approach. In a recent Public Library of Science paper, a group of Tufts researchers examined disparities in blood pressure prediction models that use PPG signals. “Demographic Subgroup-Aware Analysis of Blood Pressure Predictive Models” identifies key factors that contribute to inconsistencies and makes suggestions for enhancing model performance across diverse populations.

The paper was a collaboration between two labs in the Department of Electrical and Computer Engineering at Tufts. Blessing Kolawole, a Ph.D. student and the paper’s first author, worked with Associate Professor Valencia Koomson, Associate Professor Yingjie Lao, and Ph.D. student Sinuo Fan to improve three models for blood pressure prediction based on PPG waveform analysis.

The Tufts team trained three machine learning models using publicly available data from bedside patient monitors in intensive care units. They conducted a subgroup-aware analysis to evaluate the accuracy of their models across different demographic groups.

Ultimately, their predictive methods were competitive with state-of-the-art techniques in PPG waveform analysis. Despite variation across demographic groups, their data remained within the acceptable clinical thresholds established by the Association for the Advancement of Medical Instrumentation for all racial categories tested.

“Our results suggest that signal quality and demographic representation during model training play a crucial role in the accuracy and reliability of blood pressure predictions,” the researchers share in their conclusion. With this knowledge, the team hopes that future researchers can prioritize these features when developing and testing blood pressure prediction technology.