Abstract
Blood pressure (BP) is a critical vital sign for assessing individual health. While oscillometric techniques are widely used to estimate mean arterial pressure, establishing reliable confidence intervals (CIs) for systolic (SBP) and diastolic (DBP) pressure estimates remains a challenge, particularly with limited data. This paper proposes a nonparametric bootstrap (NPB) technique to derive CIs from a small number of BP measurements. By generating pseudo-measurements to derive pseudo maximum amplitudes (PMA) and pseudo envelopes (PE), the method establishes new relationships for CI estimation. Experimental results from 85 patients demonstrate that this approach yields significantly narrower CIs than the conventional Student’s t-method.
Introduction
Most modern home blood pressure monitors utilize the oscillometric method, which measures pressure fluctuations in the arterial wall as a cuff deflates. While convenient, these devices typically provide only point estimates for SBP and DBP, offering no measure of the underlying uncertainty. In statistical analysis, calculating a robust Confidence Interval (CI) generally requires a large sample size. However, in clinical or home settings, obtaining the 100+ readings typically required for traditional CI estimation is physically uncomfortable and impractical. This study addresses the "small data" challenge by applying bootstrapping techniques to quantify uncertainty using only five measurements.
Methodology
The proposed algorithm employs a Nonparametric Bootstrap (NPB) approach. The fundamental concept is to treat a small sample of five measurements as a representative population and resample from it with replacement to create a large synthetic dataset (e.g., $B=1000$ iterations).
As shown in Figure 1, the oscillometric waveform (OMW) is used to construct an envelope. The SBP and DBP are then determined using heuristic ratios—specifically 0.70 for SBP and 0.45 for DBP—relative to the maximum amplitude (MA). The NPB process generates Pseudo Maximum Amplitudes (PMA) and Pseudo Envelopes (PE). By analyzing the distribution of these synthetic envelopes, the researchers can determine the upper and lower bounds of the pressure estimates, effectively "stretching" the utility of a limited physical dataset to approximate the true distribution of the estimator.
Experimental Setup and Datasets
The methodology was validated using an experimental dataset comprising 425 measurements from 85 healthy subjects (37 females, 48 males) aged 12 to 80. Each volunteer provided five sets of measurements using a wrist-worn UFIT TEN-10 device. To establish a ground truth, two trained nurses performed simultaneous auscultatory (stethoscope-based) readings for each patient, serving as the reference standard.
Key Results and Discussion
The proposed PMAE (Pseudo Maximum Amplitude Ensemble) method demonstrated superior precision compared to the conventional Student t-method (MAAST) and the Guide to the Expression of Uncertainty in Measurement (MAAGUM).
As illustrated in the distributions in Figure 4, the bootstrap method allows for a granular view of measurement variability. Key performance metrics include:
- CI Width Reduction: For SBP, the NPB method yielded a CI width of 2.7 mmHg, compared to 13.3 mmHg for the t-method and 14.0 mmHg for the GUM method.
- Improved Stability: The standard deviation of the CI was significantly lower (3.2 for SBP) than that of the t-method (8.1).
- Accuracy Maintenance: The Mean Absolute Error (MAE) compared to nurse readings was approximately 6.50 mmHg for SBP and 5.64 mmHg for DBP, which is comparable to standard point-estimation algorithms.
While traditional methods are often overly conservative with small samples—leading to wide, less useful intervals—the NPB approach provides a much tighter and more reliable window for the patient's true blood pressure.
Conclusion and Future Work
This research demonstrates that nonparametric bootstrapping can effectively provide precise confidence intervals for oscillometric BP measurements using minimal data. While the study focused on healthy subjects, future work should evaluate the algorithm's performance in populations with cardiovascular conditions like arrhythmias or hypertension, which may introduce additional noise into the oscillometric envelope. Additionally, further refinement is needed to bring the Mean Absolute Error closer to the 5 mmHg threshold recommended by AAMI standards.
Summary updated