Evaluation of physiology-based models for noninvasive blood pressure estimation using pulse arrival time

Q: What is the clinical evidence from this study?

Study design: Observational. Population: Blood pressure estimation (n=15). Intervention: Generalized physiology-based model incorporating PAT, RR, and respiration vs. Classical models using only PAT. Primary outcome: Root mean square (RMS) error for diastolic blood pressure estimation.

Artur Poliński; J. Rosell

doi:10.1038/s41598-026-50073-5

April 27, 2026Scientific ReportsOpen Access

Evaluation of physiology-based models for noninvasive blood pressure estimation using pulse arrival time

View Full Paper

Ask AI

Bookmark

Share

Key result

A generalized physiology-based model reduces diastolic BP estimation error ~26% versus classical models.

n=15

Figures

Why the study?

Noninvasive, continuous BP measurement methods using pulse wave velocity are needed for wearable monitoring, but existing approaches face limitations including accuracy.

Does a nonlinear model utilizing pulse arrival time, pulse period, and respiration improve the accuracy of noninvasive blood pressure estimation compared to simple models?

Key Points

To evaluate generalized, nonlinear physiology-based models that estimate continuous blood pressure using pulse arrival time, pulse period, and respiratory phase against invasive reference data.
Evaluated deterministic models (1) through (10) using a publicly available database containing invasive arterial blood pressure catheter recordings.
Computed pulse arrival time (PAT) using four fiducial point algorithms, combined with pulse period (RR) and respiratory activity phase (resp).
Assessed model performance and parameter configurations using correlation, root mean square (RMS) error, Akaike information criterion (AIC), Bayesian information criterion (BIC), and minimum description length (MDL) across a 70%–30% data split.
Winning nonlinear models required 33 to 274 parameters, reaching a maximum correlation of 0.73 with an RMS of 3.81 mmHg for diastolic blood pressure and a correlation of 0.75 with an RMS of 6.37 mmHg for systolic blood pressure.
Parameter ablation revealed that pulse arrival time had the largest impact on estimation error (12.32% to 31.44%), followed by pulse period (4.51% to 17.28%) and respiration (0.34% to 2.95%).
Pure autoregressive models of pressure alone outperformed all physiological models, achieving correlations near 0.93 and reducing RMS error by approximately twofold.

Study Design

Type

Observational (n=15)

Structured PICO

Does a nonlinear model utilizing pulse arrival time, pulse period, and respiration improve the accuracy of noninvasive blood pressure estimation compared to simple models?

P

Population

15 records from the publicly available MIT-BIH Polysomnographic Database (PhysioNet) containing simultaneous blood pressure, ECG, and respiration signals (originally 18 records, 3 excluded due to missing data).

I

Intervention

Generalized, nonlinear models for blood pressure estimation utilizing pulse arrival time (PAT), pulse period (RR), and respiratory activity phase (resp).

C

Comparator

Simple models presented in the literature utilizing only pulse arrival time (PAT) or PAT and pulse period (RR).

O

Outcome

Accuracy of blood pressure estimation measured by correlation, root mean square (RMS) error, Akaike information criterion (AIC), Bayesian information criterion (BIC), and minimum description length (MDL).surrogate

Main Result

Absolute Event Rate: 3.81% vs 5.18%

Limitations

Original signals were corrupted by noise and artefacts, requiring data fragmentation
Blood pressure change range was relatively small as measurements were performed on lying persons
Stationarity of the signals was assumed
Invasive blood pressure was used for PAT estimation instead of noninvasive plethysmographic measurements
Models may not capture physiological differences between individuals such as arterial stiffness
The considered signals (PAT, RR, and resp) contain incomplete information about the current pressure value, leading to relatively large errors.
The number of significant coefficients at the 5% significance level depends on the person, with no common set of parameters for all data.

Implication

Complex nonlinear models incorporating pulse arrival time, heart rate, and respiration improve noninvasive blood pressure estimation compared to simple models, but still lack sufficient accuracy for reliable continuous monitoring.

Abstract

A noninvasive and nonocclusive blood pressure (BP) measurement method is essential for ambulatory and long-term monitoring. It appears that in the case of wearable systems for continuous pressure measurement, the technique utilizing the dependence of pulse wave velocity on pressure is particularly useful. However, it has some limitations, e.g., accuracy. A generalized, nonlinear model, with respect to physiological parameters, for BP estimation, utilizing information about pulse arrival time (PAT), pulse period (RR), and respiratory activity phase (resp), is proposed and analysed. Analyses have been conducted using a publicly available database. The models are compared against each other using various measures such as correlation, root mean square, Akaike information criterion, Bayesian information criterion, and minimum description length. An optimal model, superior to those presented in the literature, is recommended for each measure. In addition, the influence of individual signals on the pressure estimation error was analysed. The results show that simple models generate large errors in BP estimation. Including more parameters improves the results, but the errors are still relatively large. The presented results suggest that the considered signals, i.e., PAT, RR, and resp, contain incomplete information about the current pressure value.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Poliński et al. (2026) conducted an observational in Blood pressure estimation (n=15). Generalized physiology-based model incorporating PAT, RR, and respiration vs. Classical models using only PAT was evaluated on Root mean square (RMS) error for diastolic blood pressure estimation. A generalized physiology-based model incorporating pulse arrival time, RR interval, and respiratory activity improved blood pressure estimation, achieving an RMS error of 3.81 mmHg for diastolic pressure compared to 5.18 mmHg with classical models.

synapsesocial.com/papers/69f2f0e31e5f7920c6386d71 https://doi.org/10.1038/s41598-026-50073-5

View Full Paper

Ask AI

Bookmark

Share

Evaluation of physiology-based models for noninvasive blood pressure estimation using pulse arrival time

Figures

Key Points

Study Design

Structured PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Evaluation of physiology-based models for noninvasive blood pressure estimation using pulse arrival time

Figures

Key Points

Study Design

Structured PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

More related papers

More from Scientific Reports

Keep reading

More related papers

More from Scientific Reports