Prediction Algorithms for Blood Pressure Based on Pulse Wave Velocity Using Health Checkup Data in Healthy Korean Men: Algorithm Development and Validation (Preprint)

Dohyun Park, Soo Jin Cho, Kyunga Kim, Hyunki Woo, Jee Eun Kim, Jin-Young Lee, Janghyun Koh, Jeanhyoung Lee, Jong Soo Choi, Dong Kyung Chang, Yoon‐Ho Choi, Ji In Chung, Won Chul, Ok Soon Jeong, Se Yong Jekal, Mira Kang2021 · DOI

TL;DR

This study developed and validated blood pressure (BP) prediction models using pulse wave velocity (PWV) and routine health checkup data in healthy Korean men. By stratifying participants by age (under and over 60), they achieved better prediction accuracy, demonstrating that a simple model with PWV, BMI, and age is efficient and feasible for predicting BP, meeting clinical standards.

What Problem Does This Paper Address?

Accurate blood pressure (BP) measurement is crucial for managing cardiovascular disease (CVD), a leading cause of death worldwide. While wearable devices offer potential for continuous BP monitoring, previous studies using pulse transit time (PTT) and pulse wave velocity (PWV) have been limited by small sample sizes and a lack of evaluation on the relative importance of different clinical variables. This leaves a gap in understanding how effectively PWV can predict BP in a larger, more diverse population and which other factors contribute most significantly.

Methods

The researchers employed multiple linear regression to build BP prediction models. A key methodological innovation was the age-based stratification of the study population into two groups: younger than 60 years and 60 years or older. This approach was motivated by the understanding that cardiovascular system changes with age, particularly aortic stiffness, which directly impacts PWV. They developed and compared several models:

  • Model 1: Used only PWV.
  • Model 2: Added BMI and age to Model 1.
  • Model 3: Included 17 variables (PWV, BMI, age, heart rate, smoke status, white blood cell count, hemoglobin, uric acid, sodium, potassium, LDL, HDL, triglyceride, testosterone, creatinine, CRP, and diabetes).

Model performance was evaluated using mean error, absolute error, and root mean squared error (RMSE). To account for potential partition bias, the analysis was repeated 200 times with random seeds. The relative importance of predictor variables was assessed using Johnson's relative weights.

Datasets

The study utilized data from 1362 healthy Korean men aged 18 and older who underwent health medical examinations at the Samsung Medical Center between January 2014 and December 2015. Data was extracted from the Clinical Data Warehouse. Participants taking antihypertensive medications or alpha-blockers were excluded to avoid confounding effects on PWV. The dataset was split into training (70%) and validation (30%) cohorts. For the validation cohort, 100 patients were randomly selected from each age group (<60 and ≥60 years) after stratification by age and BMI.

Key Insights and Ideas

  • Age Stratification is Key: The most significant finding is that stratifying the population by age (specifically at 60 years) substantially improves BP prediction accuracy compared to a single model for all ages. This highlights the differential impact of aging on cardiovascular mechanics.
  • Simplicity Wins: A parsimonious model incorporating only PWV, BMI, and age performed comparably to a much more complex model with 17 variables. This suggests that these three factors are the most critical drivers of BP in this population.
  • PWV's Dominance: Pulse wave velocity emerged as the most influential predictor of both systolic and diastolic blood pressure, underscoring its potential as a non-invasive BP surrogate.
  • Clinical Feasibility: The final model, using readily available health checkup data (PWV, BMI, age), meets established clinical criteria (AAMI), making it practical for real-world application.

Key Results

  • The age-stratified models (Algorithm 2) demonstrated significantly better performance (lower mean error) for both systolic and diastolic blood pressure compared to the non-stratified model (Algorithm 1).
  • For systolic BP prediction, the age-stratified model showed a significantly smaller mean error. For diastolic BP, the age-stratified model also showed a significantly lower mean error in the younger age group.
  • The performance difference between a simple three-variable model (PWV, BMI, age) and a 17-variable model was not statistically significant.
  • The final model (Model 2 in Algorithm 2) met the American Association for the Advancement of Medical Instrumentation (AAMI) criteria for BP prediction.
  • Prediction errors were within the range of approximately 9 to 12 mmHg, comparable to those of a gold standard mercury sphygmomanometer.
  • PWV was the most significant predictor of BP, followed by BMI and age. The influence of PWV and BMI on BP was more pronounced in the younger age group (<60 years).

Limitations and Future Work

The authors acknowledge several limitations:

  • Healthy Male Population: The study was conducted exclusively on healthy Korean men. This limits the generalizability of the findings to women, individuals with hypertension or hypotension (though some were included), and other ethnic groups.
  • Medication Information: Due to the retrospective nature of the study, detailed medication information beyond antihypertensives was not fully captured, potentially limiting the analysis of drug effects on PWV.
  • Internal Validation: The model was internally validated. External validation is necessary to confirm its robustness and generalizability across different populations and settings.

Future work should focus on:

  • Expanding studies to diverse populations, including women and individuals on various medications.
  • Investigating the performance of PWV in predicting BP across a wider range of BP values.
  • Conducting external validation of the developed prediction model.

Summary updated

Abstract

<sec> <title>BACKGROUND</title> Pulse transit time and pulse wave velocity (PWV) are related to blood pressure (BP), and there were continuous attempts to use these to predict BP through wearable devices. However, previous studies were conducted on a small scale and could not confirm the relative importance of each variable in predicting BP. </sec> <sec> <title>OBJECTIVE</title> This study aims to predict systolic blood pressure and diastolic blood pressure based on PWV and to evaluate the relative importance of each clinical variable used in BP prediction models. </sec> <sec> <title>METHODS</title> This study was conducted on 1362 healthy men older than 18 years who visited the Samsung Medical Center. The systolic blood pressure and diastolic blood pressure were estimated using the multiple linear regression method. Models were divided into two groups based on age: younger than 60 years and 60 years or older; 200 seeds were repeated in consideration of partition bias. Mean of error, absolute error, and root mean square error were used as performance metrics. </sec> <sec> <title>RESULTS</title> The model divided into two age groups (younger than 60 years and 60 years and older) performed better than the model without division. The performance difference between the model using only three variables (PWV, BMI, age) and the model using 17 variables was not significant. Our final model using PWV, BMI, and age met the criteria presented by the American Association for the Advancement of Medical Instrumentation. The prediction errors were within the range of about 9 to 12 mmHg that can occur with a gold standard mercury sphygmomanometer. </sec> <sec> <title>CONCLUSIONS</title> Dividing age based on the age of 60 years showed better BP prediction performance, and it could show good performance even if only PWV, BMI, and age variables were included. Our final model with the minimal number of variables (PWB, BMI, age) would be efficient and feasible for predicting BP. </sec>
BibTeX
@misc{park2021prediction,
  author = {Dohyun Park and Soo Jin Cho and Kyunga Kim and Hyunki Woo and Jee Eun Kim and Jin-Young Lee and Janghyun Koh and Jeanhyoung Lee and Jong Soo Choi and Dong Kyung Chang and Yoon‐Ho Choi and Ji In Chung and Won Chul and Ok Soon Jeong and Se Yong Jekal and Mira Kang},
  title = {{Prediction Algorithms for Blood Pressure Based on Pulse Wave Velocity Using Health Checkup Data in Healthy Korean Men: Algorithm Development and Validation (Preprint)}},
  year = {2021},
  doi = {10.2196/preprints.29212},
}
Figure