IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 60, NO. 10, OCTOBER 2011 3405 Confidence Interval Estimation for Oscillometric Blood Pressure Measurements Using Bootstrap Approaches Soojeong Lee, Miodrag Bolic, Senior Member, IEEE, Voicu Z. Groza, Fellow, IEEE, Hilmi R. Dajani, Member, IEEE, and Sreeraman Rajan, Senior Member, IEEE Abstract—Although estimation of average blood pressure is commonly done with oscillometric measurements, confidence in- tervals (CIs) for systolic blood pressure (SBP) and diastolic blood pressure (DBP) are not usually estimated. This paper adopts bootstrap methodologies to build CI from a small sample set of measurements, which is a situation commonly encountered in practice. Three bootstrap methodologies, namely, nonparametric percentile bootstrap, standard bootstrap, and bias-corrected and accelerated bootstrap are investigated. A two-step methodology is proposed based on pseudomeasurements using bootstrap princi- ples to first derive the pseudomaximum amplitudes and then the pseudoenvelopes (PEs). The SBP and DBP are estimated using the new relationships between mean cuff pressure and PE and then the CIs for such estimates are obtained. In order to reduce the amount of processing, a single-step methodology that directly derives PE using bootstrap principles is also presented. Applica- tion of the proposed methodology on an experimental data set of 85 patients with five sets of measurements for each patient has yielded a narrower CI than the currently available conventional methods such as Student’s t-distribution method. Index Terms—Blood pressure (BP), bootstrap, confidence inter- val (CI), oscillometric method. I. I NTRODUCTION D EVICES that perform blood pressure (BP) measurements based on oscillometric methods have become ubiquitous in home-based and personal health care due to their ease of use [1]–[6]. These devices typically only provide single-point estimates for both systolic blood pressure (SBP) and diastolic blood pressure (DBP) and do not provide a confidence interval (CI). Although standards exist for the measurement procedure, there are no standards for the estimation of the SBP and DBP. The SBP and DBP estimates from oscillometric waveforms are most commonly obtained using the maximum amplitude algorithm (MAA) [3]. This algorithm determines the maximum Manuscript received July 20, 2010; revised May 26, 2011; accepted June 5, 2011. Date of current version September 14, 2011. This work was supported by collaborative research funding from the Ontario Centres of Excellence and Biosign Technologies Inc. The Associate Editor coordinating the review process for this paper was Dr. Domenico Grimaldi. S. Lee was with the School of Information Technology and Engineering, University of Ottawa, Ottawa, ON K1N 6N5, Canada. He is now with the Department of Electronic Engineering, Sogang University, Seoul 121-742, Korea (e-mail: leesoo86@sogang.ac.kr). M. Bolic, V. Z. Groza, H. R. Dajani, and S. Rajan are with the School of Information Technology and Engineering, University of Ottawa, Ottawa, ON K1N 6N5, Canada (e-mail: mbolic@site.uottawa.ca; groza@site.uottawa.ca; hdajani@site.uottawa.ca; sreeraman@leee.org). Digital Object Identifier 10.1109/TIM.2011.2161926 amplitude (MA) of the envelope of pressure pulses detected using a piezoelectric sensor embedded in a deflating pressure cuff. The algorithm then uses empirically derived ratios to estimate the SBP and DBP relative to the MA of the envelope. There has been no work reported in the literature on estimation of CI for BP measurements until very recently. In [7], CIs were calculated for SBP, DBP, heart rate, and pulse rate obtained from an Omron HEM725CIC (Omron Healthcare Inc., Vernon Hills, Illinois, USA) device over a period of 7 days, with four measurements per day leading to a total of 28 measurements per subject, which is not considered large; hence, the conven- tional Student’s t-distribution (T), instead of asymptotic normal distribution, was used to obtain the CI of SBP and DBP. The asymptotic normal approximation is commonly used to produce CIs, but it performs well only when the sample size is large. Even though oscillometric methods for BP measurements are noninvasive, faster, and simpler than invasive measurements, it is not practically feasible to obtain a large number of measure- ments for each subject or guarantee repeatable conditions for reproducible measurements. Therefore, one has to be content with fewer measurements, and this rules out the use of the standard normal distribution method for CI or even the use of Student’s t-distribution method for CI such as the one presented in [7]. In general, there is a need to develop a methodology for obtaining CI for BP estimates obtained through an oscillometric method as only a very few measurements are available; hence, in this paper, the bootstrap method for obtaining CI is proposed. The bootstrap method was originally introduced by Efron in [8] and used in [9] to address CI estimate for statistics based on independent and identically distributed (i.i.d.) random variable from some unknown distribution F μ,σ . The CI obtained through bootstrap method uses a limited data set for improving accuracy of the estimates and is used in places where the conventional method such as application of the Student’s t-distribution for improving accuracy are not valid [10]. The advantage of the bootstrap method is that it does not require any modeling or assumptions on the data, except that the data should be i.i.d. [10]. The idea of using bootstrap approach with oscillometric BP measurements was originally presented in the conference paper [11] by us. This paper is an expanded version of the conference paper with the following enhancements. 1) Additional bootstrap techniques are introduced, and a comparison of the different bootstrap variants is pre- sented. The effect of number of resamples on the CIs is also verified. 0018-9456/$26.00 © 2011 IEEE