1 Introduction

Measurement of blood pressure is one of the most common and important clinical and diagnostic measurements made by family doctors and hospital physicians [3]. Accurate blood pressure measurement is very important, with published data estimating that with a 5 mmHg overestimation, millions of people would receive an inaccurate diagnosis of hypertension and take unnecessary anti-hypertension medication, as well as increase unnecessary costs to healthcare providers [13]. A systematic error of 5 mmHg is also the absolute maximum systematic error allowed in assessing blood pressure devices [12].

Systolic and diastolic blood pressures (SBP and DBP) are important in assessing cardiovascular status, but there is also clinical interest in mean arterial pressure (MAP), defined as the average pressure throughout the cardiac cycle. MAP is considered to be the driving pressure for perfusion of most vital organs, which should be at least 60 mmHg [4]. MAP has physiological and clinical importance, and can be used as a predictor of cardiovascular risk [21]. For women in pregnancy, MAP has been reported as a better predictor for pre-eclampsia than SBP and DBP [7]. MAP is also useful in intensive care, where it can often be directly measured from invasive arterial pressure, for assessing hemodynamic variables to guide treatment [10].

The true MAP is calculated from the invasive blood pressure curve, but invasive measurements are generally not clinically acceptable. Traditionally, non-invasive MAP is calculated by using an empirical formula, in which MAP approximately equals DBP plus 33% of the pressure difference between SBP and DBP [11]. The value of ‘33%’ is referred to here as the manual mean weight for calculating MAP. The gold standard for non-invasive SBP and DBP measurement has been readings taken by a trained observer using a mercury sphygmomanometer and the Korotkoff sound technique [17]. MAP derived from this classic manual technique is regarded as a useful measurement.

Mean arterial pressure is also determined in most automatic, non-invasive blood pressure measurement devices (NIBP) using the oscillometric technique [8, 18], even where it is not displayed. Those automated devices are used frequently in many health care institutions [3]. Technically, automated oscillometric devices analyse the pressure pulse changes (oscillometric pulses) induced in a pressurized cuff wrapped round the limb during deflation. These changes are caused by the pulse radiating down the artery producing pressure changes in the cuff, which are expected to be greatest at MAP [18]. This feature has been used to determine MAP in these oscillometric blood pressure devices. Manufacturers of automated devices then devise their own algorithms by adding additional information, such as characteristic ratios of the pulse amplitude to the maximum pulse amplitude, to estimate SBP and DBP after using MAP in the calculation [8, 24].

A characteristic of the oscillometric pulses (typically amplitude) is used to construct the oscillometric waveform envelope which characterizes the variation in oscillometric pulse amplitude with cuff pressure. Ideally, the oscillometric waveform envelope is a smooth curve with a distinct peak from which the MAP can be determined. However, the oscillometric waveform envelope is constructed from samples (at the pulse rate) of the changing cuff pressure. In the majority of oscillometric NIBP devices, often with fast deflation rates up to 10 mmHg/s, the number of detectable oscillations is reduced in comparison with that from the recommended deflation rate of 2–3 mmHg/s. Interpolation is then necessary to compensate for long periods between oscillometric pulses. Furthermore, the oscillometric waveform envelope is not always a neat bell-shaped curve with a distinct peak [2]. Rather than a distinct peak the oscillometric waveform envelope may have a plateau. To further add to the challenge of determining the MAP, the oscillometric measurement is compromised by movement and other forms of artefact, such as respiratory disturbance [1, 14, 19, 22]. These disturbances are associated with changes in the oscillometric pulses in the cuff pressure, and hence destroy the smoothness of the oscillometric waveform envelope, leading to difficulties in accurate BP determination. Therefore, developing a technique which can reliably determine MAP is important.

The aim of this study was to develop different techniques to estimate automated MAPs from oscillometric waveforms and to compare these estimations of MAP with the auscultatory derived method.

2 Methods

2.1 Subjects

Fifty-five normal healthy subjects, with no known cardiovascular disease, were studied. There were 39 male and 16 female subjects, with ages in the range 20–74 years. The detailed subject information including age, sex, height, weight and arm circumference are summarized in Table 1. This study received ethical permission, and all subjects gave their written informed consent.

Table 1 General data information for the subjects studied
Full size table

2.2 Blood pressure measurement protocol

Blood pressure measurements were undertaken in a quiet room. The subject was seated in a chair with their feet on the floor and with their arm supported at heart level. There was a 5-min rest period before formal recording to allow cardiovascular stabilization.

Figure 1 shows the schematic representation of the measurement protocol. Auscultatory SBP and DBP were recorded under controlled conditions with a reliable, accurate and clinically validated electronic sphygmomanometer (Accoson Greenlight 300) [9], simultaneously by two trained observers using a dual-headed stethoscope. The whole auscultatory measurement procedure followed the guidelines recommended by the American Heart Association [17]. All measurement pairs agreed within 4 mmHg and their average values were used for further analysis. On each subject, repeat recordings were performed after the cuff pressure was released for 2 min. There was no significant difference on the auscultatory SBP and DBP between repeat recordings (both P > 0.15). The average value of auscultatory SBP and DBP from the repeat recordings was calculated as the reference SBP and DBP for that subject. In total, 148 recordings were included from all subjects, with three recordings on 38 subjects and two recordings on 17 subjects. The overall mean and standard deviation (SD) of auscultatory SBP and DBP over all subjects are also given in Table 1.

Fig. 1
figure 1

Overview of the blood pressure measurement protocol

Full size image

During each recording, the cuff pressure was deflated at 2–3 mmHg/s and recorded by a pressure sensor in a separate recording system. As the subject was asked to keep still during the whole measurement procedure, movement artefact and its effect to the cuff air system was minimized. The cuff pressure was then digitally recorded to a data capture computer at a sample rate of 2000 Hz for subsequent off-line analysis. The oscillometric pulses were then extracted from the cuff pressure after segmenting each pulse and removing the baseline cuff pressure by using the software previously developed by the PTB in Berlin, Germany, and checking the pulse extraction manually [16]. The segmentation borders were at the feet of oscillometric pulses. The peak of each oscillometric pulse was used for further analysis. Figure 2 shows typical examples of deflating cuff pressures and the corresponding oscillometric pulses as the cuff pressure was reduced. The oscillometric waveform in Fig. 2a has a distinct peak, whilst there is a plateau in Fig. 2b.

Fig. 2
figure 2

Typical examples of cuff pressure and extracted oscillometric pulse waveforms with a distinct peak (a) and with a plateau (b). Oscillometric pressure was measured from the oscillometric pulses, which were extracted from the cuff pressure after the removal of the cuff deflation pressure

Full size image

2.3 Mean arterial blood pressure determination

Auscultatory MAP was estimated in each recording from auscultatory DBP plus 33% of the pressure change from DBP to SBP, with supplementary data for the use of 40% in the formula [5]. Six different techniques were then developed using Matlab 7.1. (MathWorks Inc., USA) to determine the corresponding automated MAPs from the cuff pressure at a specific time determined from oscillometric pulse pressure. Details are illustrated in Fig. 3 and described below:

Fig. 3
figure 3

Illustration of the calculation of MAP from six different techniques. See text for the detailed definition. In this example, MAP 5 overlaps MAP4

Full size image

  • MAP1 and MAP2 were the cuff pressures at the peak and foot, respectively of the largest oscillometric pulse [25]. They are the traditional oscillometric techniques for MAP estimation, but can be influenced by the variation of oscillometric waveform caused by artefact or noise.

  • MAP3 and MAP4 were the cuff pressures at the peak of the pulse which was closest to the peak of the 4th and 6th order fitted polynomial curve, respectively. The polynomial curve fitting technique was used here because it has the power to interpolate between the sampled pulses and to smooth out noise in the oscillometric waveform. Furthermore, because the oscillometric waveform envelope is similar to a bell-shaped curve, an even number of orders above 2 is required to construct a curve with flat features at the beginning and the end.

  • MAP5 was obtained from MAP4 after correcting for the amplitude difference between the 6th order polynomial curve and the recorded oscillometric pulse, such that when the polynomial curve lay above the oscillometric pulse, the correction added was positive. This amplitude difference can be caused by respiratory effects.

  • MAP6 was the baseline cuff pressure associated with the peak of the 6th order fitted polynomial curve, where the baseline was drawn through the foot of all pulses. This technique may be able to correct the effect of fast cuff deflation rate, with a limited number of detectable oscillometric pulses.

2.4 Data analysis

For all the blood pressure recordings, SPSS software package (SPSS Inc., USA) was employed to determine the effect of different techniques on MAP estimation, and also to perform regression analysis between the auscultatory MAPs and automated MAPs from the six different oscillometric techniques, obtaining the square of the correlation coefficients (R 2). Bland–Altman analysis (Systat Software Inc., USA) was also performed to assess the agreement between the automated MAPs and auscultatory MAPs, with their paired differences and SD of differences calculated. A value of P < 0.05 was considered statistically significant. Finally, the automated mean weights calculated from the automated MAPs were compared with the classical manual mean value of 33%.

3 Results

The overall mean MAPs from all the recordings for the auscultatory and oscillometric techniques are given in Table 2. ANOVA analysis showed that different techniques resulted in significant differences in MAP estimation (P < 0.001). The mean paired differences between the oscillometric techniques and the auscultatory technique were significant (all P < 0.01). Technique 6 with curve fitting improved the paired differences from −1.3 and −3.4 mmHg (classic technique, without curve fitting) to −1.0 mmHg. These paired differences were associated with differences between the manual and automated mean weights of 4% (33–29%) and 9% (33–24%) for the classic techniques to 2% (33–31%) for technique 6.

Table 2 The mean MAP calculated from the manual auscultatory method and from six different oscillometric techniques. Their mean and SD of paired differences, and the mean weights calculated from the MAPs are also presented
Full size table

Figure 4 shows the comparison of MAPs obtained from the auscultatory and different oscillometric techniques. The linear regression analysis showed that the correlations between the six calculated automated MAPs and the auscultatory MAP were significant (all P < 0.001), with the R 2 of 0.71, 0.73, 0.84, 0.84, 0.84 and 0.86, respectively, increasing from techniques 1 to 6. Bland–Altman analysis showed that technique 6 improved the SD of paired differences best from 6.2 and 5.9 mmHg (classic technique, without curve fitting) to 3.7 mmHg.

Fig. 4
figure 4

Regression analysis results of MAP from the auscultatory and six oscillometric techniques (top six sub-figures). Bland–Altman plot of MAP differences ±2SD between the six oscillometric techniques and the auscultatory method (bottom six sub-figures)

Full size image

When the mean weight of 40% rather than 33% was used to calculate the auscultatory MAP, all mean paired difference in comparison with the auscultatory MAP fell by an identical 2.9 mmHg with no change to the differences between oscillometric techniques. There were small differences in the SD of paired differences, but not to the order between oscillometric techniques, with MAP6 still having the lowest variability.

4 Discussion

We have shown that the automated MAPs calculated from the oscillometric waveform using different measurement algorithms were different. The technique associated with the peak of the 6th order polynomial model of the oscillometric pulse waveform yielded the smallest mean paired difference in comparison with the auscultatory technique. Technique 1 yielded a higher mean than that from technique 2. This difference reflects the amplitude of the oscillometric pulses (see Fig. 3) with technique 2 associated with the lower cuff pressure as it is the pressure at the foot, rather than the peak of the oscillometric pulse. Similarly, the MAPs from technique 3, 4 and 5 were higher than from technique 6 although the polynomial curve fitting technique was applied to all of them. This difference again reflects the amplitude of oscillometric pulse with technique 3–5, whilst technique 6 used the cuff pressure at the baseline (see Fig. 3). Although the mean MAP differences compared with the auscultatory technique were less than 5 mmHg, the systematic mean difference between different oscillometric techniques was up to 5.5 mmHg. The International Standard requires a maximum systemic difference of 5 mmHg [12]. Furthermore, in a patient population, the mean differences between the oscillometric techniques and the auscultatory technique can be expected to be much higher.

Next, the significant differences between MAPs from the oscillometric techniques and the auscultatory technique resulted in the automated mean weights calculated from automated MAPs being different for all the six techniques studied. Technique 6 produced the smallest mean difference (−2%) against the classically assumed mean value of 33%. It is generally accepted that the automated MAP from the oscillometric waveform gives a fairly accurate estimation in comparison with the intra-arterial MAP [24], our finding allows us to question whether 33% is a useful mean weight for achieving accurate MAP estimation. It has been reported that a weight of 41% is more accurate than the traditional 33% for the calculation of MAP from invasive SBP and DBP after investigating the pulses measured invasively from 150 patients during cardiac catheterization [15]. Bos et al. [5] also found that, in comparison with the intra-arterially measured MAP, using the mean weight of 40% avoided the underestimation of MAP calculated from the empirical formula. Razminia et al. [20] reported the use of the traditional 33%, but with an added correction for heart rate (HR), where MAP was expressed as: MAP = DBP + [0.33 + (HR × 0.0012)] × (SBP-DBP). It is accepted that there is still uncertainty with calls for more invasive clinical studies to evaluate the mean weight for MAP estimation [6, 23], and whether the mean weight changes with different clinical conditions, such as the age and blood pressure range, which could influence the shape of the oscillometric waveform [2]. These values would provide useful information for validating automated blood pressure devices for clinical use.

More interestingly, we have also shown that the variability of MAP measurement differences from different oscillometric techniques in comparison with the auscultatory technique can be improved with curve fitting algorithms. The technique, associated with the peak of the 6th order polynomial model of the oscillometric pulse waveform, was shown to result in the lowest variability. It yielded the smallest SD of paired differences (3.7 mmHg) when referenced to those calculated from the classic empirical formula. The smoothing of the polynomial curve fitting of the peaks accounted for its reduced SD. The fitted curve smoothes out random and noisy variations in the oscillometric waveform. Having a good quality extracted oscillometric waveform is important for improving the estimation of automated MAP. In addition, although it is recognized that the interpolation errors would be greater with fewer oscillometric peaks detected in a single measurement, this model-based technique with interpolation takes account of the peak of the fitted curve lying between two oscillometric pulses, hence correcting for the effect of cuff deflation (or inflation) between heartbeats. This is increasingly important because clinical demands lead to reduced determination time and hence increased cuff deflation rate, with a reduced number of detectable oscillometric pulses.

In conclusion, different techniques for automated MAP estimation from the oscillometric waveform produced different results, and the technique associated with the peak of the 6th order polynomial curve was shown to have the smallest differences in comparison with the manual auscultatory technique.