Noninvasive Continuous Arterial Blood Pressure Monitoring with Nexfin®

• Recent studies have shown the feasibility of noninvasive and continuous blood pressure measurement during surgical procedures. However, accuracy and precision are still debated.

• This study in cardiothoracic patients shows that intraoperative noninvasive blood pressure measurements using a recently developed device (Nexfin®) incorporating physiologic reconstructive modeling are closely correlated to invasive blood pressure monitoring over a wide range of pressure changes.

DURING surgical procedures, an arterial access is regularly required for continuous blood pressure monitoring and blood gas sampling. If no arterial line is required, blood pressure can be measured noninvasively with an upper arm cuff device, but only on an intermittent basis. Moreover, upper arm cuff devices have a low accuracy1and do not reliably track large changes in arterial pressure.2In critically ill patients, oscillometric measurements are consistently lower than direct blood pressure measurements.3,4Noninvasive and continuous monitoring of blood pressure is of potential benefit during nonmajor surgical procedures. With the finger cuff technology,5blood pressure can be measured noninvasively and continuously. Devices using this technology, such as the Finapres (Ohmeda, Englewood, CO), have been clinically evaluated and used in a variety of settings.5,–,10The ability to track changes in blood pressure was considered adequate5but concerns about accuracy or precision were raised.11 

The arterial pressure waveform changes gradually from the brachial artery to the finger arteries with a decrease in diastolic pressure and an occasional increase in systolic pressure because of the narrowing of the arteries. These effects explain the scatter that was sometimes seen with the Finapres11,12and which thus far limited the clinical application of finger cuff technology-based devices. The Nexfin® (BMEYE B.V., Amsterdam, The Netherlands) is a new device using finger cuff technology. As opposed to the Finapres, brachial arterial blood pressure is reconstructed from the measured finger arterial blood pressure using previously published waveform filtering to approximate a brachial pressure wave, together with pressure level correction compensating for the proximal-to-distal pressure drop.13,14Appendix 1 describes the principles of finger arterial pressure measurement technology and finger-to-brachial pressure reconstruction in detail. We tested the hypothesis that with these implemented methodologies arterial pressure measured at the finger approaches more proximally measured pressures. To that purpose we compared Nexfin arterial pressure (NAP) with intra-arterial pressure (IAP) in cardiothoracic surgery patients, quantifying accuracy, precision, and tracking capability.

This study was approved by the Medical Ethics Committee of the Academic Medical Center of the University of Amsterdam (Amsterdam, The Netherlands) and written informed consent was obtained from patients scheduled for cardiac surgery. Patients undergoing coronary artery bypass grafting or valve replacement or reconstruction were considered eligible for inclusion (see flow diagram, fig. 1). All patients received standard monitoring (electrocardiogram, pulse oximetry, temperature, end-tidal carbon dioxide partial pressure). Anesthesia was induced with propofol and continued with morphine and sufentanil. Sedation was achieved with midazolam and neuromuscular blockade with pancuronium. Blood pressure was supported by pharmacological vasodilatation (nitroglycerine) or vasoconstriction (ephedrine and metaraminol).

IAP was measured at the radial artery through a 20-G catheter (Ref RA-04020; Arrow International Inc., Reading, PA) and pressure transducer (pressure monitoring set; Edwards Lifescience, Irvine, CA) connected to a module (HPM1006A; Hewlett Packard, Palo Alto, CA) mounted in a Philips monitor system (Philips Medical Systems, Andover, MA). The catheter system was pressurized and a small continuous flow prevented clotting. However, gas dissolved in the fluid may form into microscopic bubbles, or air may enter the system after turning an air-fluid interface stopcock. These bubbles, even though still very small, may introduce a resonant system, resulting in either overdamped or underdamped pressure recordings. Therefore the arterial lining was regularly flushed and maintenance of an adequate resonance frequency (12–25 Hz) was checked by the fast flush technique.15 

NAP was measured by the Nexfin monitor. The measurement method is based on the volume-clamp method using a finger cuff, proposed by Peňáz.16Measurements are regularly and automatically calibrated during measurement with “Physiocal,” the physiologic calibration developed by Wesseling.17A detailed description of methods of measurement and the published finger-to-brachial pressure reconstruction, i.e. , waveform filtering and level correction,13,14can be found in appendix 1. An appropriately sized finger cuff was applied to the mid-phalanx of the middle finger18ipsilaterally to the intra-arterial catheter. The “heart reference system” measured and corrected the hydrostatic difference between the finger and the heart. The “finger side” was fixated next to the measurement finger and the “heart side” at the arterial pressure transducer level.18 

The analog signals of IAP, the unreconstructed noninvasive finger arterial pressure (FAP) and the reconstructed arterial pressure (NAP; i.e. , following application of finger-to-brachial pressure reconstruction) measured by the Nexfin were sampled at 200 Hz and stored on a hard disk.

Blood pressure measurement commenced before the patient was sedated and continued throughout surgery (see appendix 2 for details). Measurements have been obtained during induction and (off-pump) maintenance. The investigator was blinded to the pharmacological agents used, and pharmacological management was not considered in the data analysis.

FAP and NAP were compared with IAP on a beat-to-beat basis. Beats were matched (analysis software developed in-house) allowing a maximum time window of 80 ms between the start of the upstroke in the respective pressure waveforms. Matched signals that were recorded during Physiocal calibration of the NAP were not taken into account. Signals were visually inspected for any artifacts, such as beats detected during transients just after a fast flush. These artifacts were manually removed. The sets of matched IAP, FAP, and NAP beats, the values for systolic, mean (obtained by integration of the pressure curve divided by the duration of the cardiac cycle), diastolic, and pulse pressure were saved in a data file.

A detailed description and a graphical representation of the statistical procedures can be found in appendix 2. For each patient recording, matched beats over a 30-min period were extracted. The starting point for including data in the analysis was chosen on the basis of the following criteria: the Nexfin monitor should have minimally reached a Physiocal interval of 50 beats and the arterial line must have been flushed. To compare systolic, diastolic, mean, and pulse pressure from FAP and NAP with those from IAP, in each patient a Bland and Altman19approach was followed: the values of each pair of FAP and IAP beats, and of NAP and IAP beats, were averaged and their FAP − IAP and NAP − IAP differences were computed. Next, 10-s averages were calculated, resulting in 180 consecutive data sets per patient. The 10-s averages of systolic and diastolic NAP were plotted against the systolic and diastolic IAP as scatter plots for each individual patient. Also for these data, the correlation coefficients (Pearson product moment correlation, separately for systolic and diastolic values) were calculated. Consistency was assessed by dividing the 180-point data sets into three sets of 60 points each. From the NAP − IAP differences in these sets, Cronbach's α and intraclass correlation coefficients were calculated. The differences of these three sets were compared with their average difference in each patient.

For each patient, the mean and SD of the 180 averages of NAP and IAP are shown in Bland-Altman19plots as horizontal coordinates where the SDs express the range of pressures for each patient (“within-subject variability”). Similarly, the mean and SD of the 180 NAP − IAP differences of each patient represent the vertical coordinate in the Bland-Altman plot, where the SDs are measures of the individual consistency (“within-subject precision”). These means and SDs of the 180 data sets for each patient were used in the group statistics.

Mean ± SD of the FAP − IAP and NAP − IAP differences over the group were defined as accuracy and precision. Differences were compared with the 5 ± 8 mmHg criterion formulated by the Association for the Advancement of Medical Instrumentation.20To investigate whether the differences are related to mean arterial blood pressure (MAP), respectively, heart rate (see appendix 1) linear regression analysis was performed. Pressure dependency was evaluated by comparison of systolic, diastolic, mean, and pulse NAP − IAP differences between the quartiles based on their corresponding pressures, MAP, and heart rate.

For data collection, statistical analyses and plotting, Microsoft Office Excel 2007 (Microsoft Corporation, Redmond, WA), SPSS 19 (IBM SPSS Statistics 19, IBM Corporation, Somers, NY), and Sigmaplot 11 (Systat Software Inc., Chicago, IL) were used.

A total of 53 patients were included; data of three patients were excluded from analysis because of technical or operational problems (fig. 1). In one patient, the heart reference system was dislocated, causing an unknown offset in the NAP; in another patient the finger cuff was malfunctioning; and in the third patient finger blood flow was insufficient. Thus, data of 50 patients were available for analysis (table 1). The ranges of IAP were 37–182 mmHg for systolic pressure, 25–110 mmHg for diastolic pressure, 33–133 mmHg for MAP, and 4–92 mmHg for pulse pressure. In total, 2.3% of the data were deleted owing to artifacts, of which more than two-thirds were related to the invasive measurement (damping and flushing of the catheter). Noninvasive measurements showed occasional oscillations in the finger cuff pressure apart from loss of data because of Physiocal calibration.

Figure 2shows a representative example of varying IAP and NAP during surgery. The effect of the brachial pressure reconstruction by waveform filtering (reducing systolic peaking) and level correction (reducing pressure drop) is illustrated in figure 3.

Figure 4shows the scatter plots of all patients. Correlation coefficients, within-subject variability, and within-subject precision are summarized in table 2. Table 3gives measures of consistency of three consecutive periods of the pressure differences, showing good to excellent Cronbach's alphas (see appendix 2 for interpretation). When the differences of these three sets were compared with their average difference in each patient, group average ± SD were within 2 ± 4 mmHg for systolic pressure, within 1 ± 3 mmHg for MAP, and within 1 ± 2 mmHg for diastolic and pulse pressure. Figure 5shows Bland-Altman plots of the individual patient averages plus SDs of systolic, diastolic, mean, and pulse pressure. Table 4lists all IAP and NAP pressures and the NAP − IAP differences. NAP tracked changes in IAP over a wide range of pressures during cardiothoracic surgery (fig. 4). This is quantified in the Bland-Altman plots (fig. 5) and in tables 2and 3, demonstrating that, although individual measurements may have a bias, the differences with IAP are constant.

The group's averaged NAP − IAP differences (n = 50) remained within the 5 ± 8 mmHg Association for the Advancement of Medical Instrumentation criterion. Regression analysis revealed no significant relation between the differences and MAP and heart rate, respectively. When systolic, diastolic, mean, and pulse pressures were divided in quartiles, their respective NAP − IAP differences were not significantly different, nor did quartiles based on MAP and heart rate show any differences.

Values for unreconstructed finger arterial pressures, FAP and FAP − IAP are given in table 5. For unreconstructed pressures the bias in systolic and MAP is too large for the Association for the Advancement of Medical Instrumentation criteria.

The main finding of this study is that implementation of physiologic models together with new finger cuff technologies renders noninvasive continuous blood pressure measurement feasible. Changes in IAP were tracked, maintaining precision over the wide ranges of pressure during cardiothoracic surgery.

The arterial pressure waveform changes when traveling from the brachial artery to the finger arteries. Systolic peaking results in higher pulse pressures, which was noticed in earlier devices such as the Finapres. The transfer function in the Nexfin was able to reciprocate this. In the smaller arteries, such as those of the hand and fingers, resistance to flow becomes manifest, causing an overall decrease in pressure. The level correction method brings the pressure levels close to brachial values.13,14Reconstructed brachial arterial pressure, here referred to as NAP, was compared with radial pressure, which usually has a somewhat larger pulse pressure than brachial pressure. This explains the underestimation of the pulse pressure of NAP. Nonetheless, NAP − IAP differences remained small; the differences between unreconstructed finger arterial pressures and IAP were larger. The present study did not focus on the influence of pharmacological hemodynamic management on the NAP. However, pharmacological interventions, although not specifically identified, were included in the analysis.

Moreover, the NAP − IAP differences did not depend on blood pressure and heart rate level, indicating that NAP is comparable with IAP in the presence of hemodynamic changes. This is supported by the close correlation over a wide range of pressures.

The Association for the Advancement of Medical Instrumentation20has suggested an average bias of maximal 5 ± 8 mmHg. These standards are applied when comparing measurements from noninvasive intermittent blood pressure devices with invasively determined arterial pressure measurements. However, there has been no agreement regarding accuracy criteria of noninvasive devices that assess the entire arterial pressure waveform compared with invasive blood pressure or noninvasive standards as Riva-Rocci/Korotkoff.21Nonetheless, reconstructed NAP was within these limits, whereas the unreconstructed finger arterial pressure was not. During only one measurement the plethysmographical signals were insufficient to obtain adequate blood pressure readings. Thus, in most patients measurements were possible, and we attribute this to the high intensity/high sensitivity plethysmograph. Generally, use of Nexfin may be beneficial in cases when an arterial line is not indicated or feasible, whereas the surgical procedures still involve major fluid shifts or require monitoring for rapid hemodynamic changes. Nexfin application is not possible in case of insufficient or absent blood flow to the finger as in vascular disease, e.g. , M. Raynaud, and conditions with extreme vasoconstriction.

Recently, Nexfin performance was assessed in a small group of critically ill patients.22Although the authors suggested the opposite, MAP appeared reliable as it was within the Association for the Advancement of Medical Instrumentation criteria. Measurements with the Nexfin do not need external calibration, whereas other devices using finger cuff technology, such as the Finometer (FMS, Amsterdam, The Netherlands)23,24and the CNAP (CNSystems Medizintechnik AG, Graz, Austria), require calibration with an upper arm cuff.25 

In conclusion, the Nexfin measures arterial blood pressure noninvasively and continuously using physiologic pressure reconstruction. This reconstruction results in values comparable with those invasively obtained, in the presence of the considerable hemodynamic perturbations during cardiac surgery. This supports the feasibility of noninvasive continuous arterial pressure monitoring when invasive measurements are not indicated.

The Nexfin (BMEYE B.V., Amsterdam, The Netherlands, fig. 6) is a device using finger cuff technology with high sensitivity optical components and digital control systems. The measurement method builds on the volume-clamp method as introduced by Peňáz.16Arterial blood volume of the finger is measured by an optical plethysmograph, which is mounted inside the cuff. The plethysmograph uses a light-emitting diode and a sensitive photo diode with a high signal-to-noise ratio. At measurement startup, the finger cuff pressure is increased in a staircase-like fashion (fig. 2) and an initial assessment of the pressure-diameter relation of the artery is made. Then blood pressure measurement starts and a fast feedback system controls finger cuff pressure, keeping arterial blood volume at a constant level (clamped). During measurement, the pressure in the cuff should equal real-time finger blood pressure so that the arterial wall is “unloaded.” The unloaded state is ascertained by applying a physiologic calibration (“Physiocal”) as developed by Wesseling.17A Physiocal is regularly and automatically performed during measurement to define and maintain the diameter at which the finger artery should be clamped, thus compensating for vasomotor-related changes in finger arteries. The Physiocal algorithm includes the search procedure and criterion for the automated determination and periodic adjustment of the arterial unloaded volume. It explores part of the pressure-diameter relation by analyzing the plethysmogram at a number of steady pressure levels, and is able to track the unloaded diameter of a finger artery even if smooth muscle tone changes. To maintain the correct unloaded diameter of the finger artery, Physiocals are performed at regular intervals.17,24A consequence is that the measurement of blood pressure is temporarily interrupted for two or more beats. The interval between successive Physiocal periods of the NAP signal depends on the stability of the pressure-diameter characteristics of the finger artery, and normally an interval of 30 beats or higher is considered to indicate stability of the NAP measurement,18although shorter intervals do not necessarily indicate that a measurement is not reliable.

A “heart reference system” measures the hydrostatic difference between the finger and the heart and corrects for any changes in the height of the hand with respect to heart level.18 

When traveling from the brachial artery to the finger arteries, the arterial pressure waveform changes gradually in shape and in absolute levels. Specifically, diastolic pressure decreases from the brachial to the finger measurement site because of flow resistance in the smaller arteries.14In addition, when traveling from the brachial to the radial measurement site, the narrowing of arteries enhances arterial pressure wave reflection, which results in peaking of systolic and thus pulse pressure (fig. 7). This can be described as a resonance frequency in the transfer function (fig. 8) and thus can be accounted for with an inverse model.14This is referred to as waveform filtering. The waveform filtering is done with a digital filter, which works on a series of the samples of the digitized signal, each sample with its own multiplication factor. In this way the filtering can be performed in the same rate as the samples come in. The development of a digital filter should be done with several precautions, notably sampling frequency exceeding the Nyquist frequency, removing the transient response before the filter assuming the desired steady state frequency response, and reduction of unwanted side lobes in the filter frequency response by the use of Hamming weighting. Drifts in the baseline of the signals should be eliminated.14 

Beyond the level of the radial artery down to the finger arteries, the frictional losses prevail, resulting in reduced pressure levels. A population-based level correction formula can rectify this pressure reduction, as described by Gizdulich et al.  14Systolic (Psys) and diastolic (Pdia) values of the waveform filtered pressure are used as input to formula 1:

in which ΔP is the pressure over which the waveform needs to be shifted (ΔP should be subtracted). When large pulsations are superimposed on a low diastolic pressure, the waveform is shifted in an upward direction. This can be interpreted as a high flow state (as indicated by large pulse pressure) and low vascular resistance (low diastolic values) associated with a large pressure decrement. Conversely, a small pulse pressure superimposed on a high diastolic pressure can be associated with a low flow state with high vascular resistance. Thus, by establishing the physiologic state of the arterial hemodynamics, a correction for the pressure decrement can be made.14Each beat is corrected with its own “level correction,” which is based on the last 30 beats.26 

In the past, we constructed a model, representing the human subclavian, axillary, and brachial arteries, based on Womersley's theory for an artery under stiff longitudinal constraint and including viscous fluid damping.27The vascular wall was taken to be linear and viscoelastic, and wall viscosity was modeled with a second order polynomial; the constants were taken from an earlier study.28The tube system was divided into seven segments, with length, radius, and wall thickness based on biologic data.29Wave propagation and damping followed directly from longitudinal impedance based on Womersley's theory and from transverse impedance that is based on the (viscoelastic) wall properties. The distal part of the arm was modeled as a three-element Windkessel. The model was linear, allowing Fourier analysis and treatment per frequency. In this model we performed a sensitivity analysis to determine which arterial parameters have the largest effect of the pressure transformation when changed. We found that in particular those parameters which affect travel time have a large impact, i.e. , arterial stiffness and radius of the trajectory over which the wave travels.30The effect of changes in local peripheral resistance was negligible. This corresponds with earlier findings by Bos et al.  31that local infusion of vasoactive substances has little effect, whereas systemic infusion, affecting mean arterial pressure, does have an effect. This most likely results from different radius and stiffness because of changed distending pressures.

The relation between proximal and distal pressure can be described by a relatively simple patient-averaged (generalized) transfer function. Application of such transfer functions gives good results, and although extensive effort has been put in finding ways to achieve individualization, so far improvement has been limited.32Therefore the use of a generalized transfer function in the Nexfin seems an adequate first approach. However, the transfer function will change in response to vasoactive medication; although, as mentioned above, as a result of systemic rather than local changes. Thus, not so much peripheral vasoconstriction in the fingers but total vasoconstriction will change the transfer function. This explains why the level correction formula is effective by using the relation between systolic and diastolic values.

The first device using finger cuff technology made commercially available by Wesseling's research group BMI-TNO was the Finapres Model-5 in 1984. In 1996 Ohmeda introduced the Ohmeda Finapres 2300 (Ohmeda, Englewood, CO), based on the TNO Model-5. These devices reported the finger arterial pressure, without finger-to-brachial pressure reconstruction. As described above, finger arterial pressure differs from more proximal pressure such as brachial and even radial pressure, depending on the hemodynamic status of the patient. Developments in the modeling of the pressure transfer over the arterial system have resulted in improved pressure values by the Nexfin. Another improvement is measurement of the hydrostatic difference between the heart and the finger. A 13-cm height difference corresponds with a pressure difference of 10 mmHg. Before this hydrostatic difference was automatically corrected, such an error was easily introduced, even in a supine patient with the hand alongside the body.

Besides the progress in modeling, improvements in hardware have been implemented. Since the introduction of the Finapres, the efficiency of infra-red LEDs has increased. Whereas in the past LEDs were driven to their maximum to obtain sufficient intensities, the large dynamic range of modern LEDs, together with the digital control system, provides more flexibility in measuring. Since the LED driving current is smaller for the high-efficiency components, the cross-talk between LED current and plethysmographic signals is also less. Further, the new cuffs benefit from the smaller size of the optical components: the effective contact area between cuff and finger is larger, resulting in a better pressure transfer. After 8 h operation the Nexfin automatically stops measuring and closes the recording file. Monitoring can be resumed, but a different finger should be selected for measurement. On theoretical grounds the reliability may be influenced by procedures involving temporary artery distortion by extreme upper limb anteflexion.

Preparations for the recording of invasive and noninvasive pressures were made shortly after the patient had taken place on the operation table. A Nexfin cuff was applied to the finger and the radial artery was cannulated; figure 9shows a schematic overview. Recording began as soon as the Nexfin was started. Of the simultaneous recordings, 30-min periods of matched beats were selected for each patient. The same length of data were used for all patients, so that data from all patients carried the same weight in the analyses. This period commenced after the noninvasive pressure recording had sufficiently stabilized, as indicated by a Physiocal interval of 50 beats or higher, and after the arterial line had been flushed. Depending on the achievement of these two requirements, recordings could comprise induction of anesthesia, pharmacological hemodynamic management, and surgery, until the start of the extracorporeal circulation (fig. 9). Figure 10illustrates the outlined period with actual recordings from one patient.

Data typically excluded from the analysis is illustrated in figure 11, where damping (before catheter flushing) or even repetitive damping of the invasive pressure recording does not allow a meaningful comparison. In figure 12, an example is given of erratic noninvasive arterial pressure with unrealistically low pulse pressure because of a faulty cuff (see fig. 1).

We also calculated the accuracy and precision of NAP compared with IAP over all available data. This approach is scientifically objectionable since artifacts were not excluded, so two methods were not compared per se , but also the effect of errors present in either method. It should be acknowledged that IAP, which serves as a reference method, is commonly fraught with errors (fig. 11). Further, the included data per patient depended on the length of their respective recording. With these caveats, the findings for NAP − IAP were −1.2 ± 7.9, 3.3 ± 5.4, 2.4 ± 6.3, and −4.5 ± 3.7 mmHg for systolic, diastolic, mean, and pulse pressures, respectively, still meeting the criteria of the Association for the Advancement of Medical Instrumentation criteria.

Of each patient, a 30-min period of IAP, FAP, and NAP were analyzed. For each beat systolic, mean, diastolic and pulse pressures were determined, and also their respective (FAP + IAP)/2 and (NAP + IAP)/2 averages together with their FAP − IAP and NAP − IAP differences. Then, for each patient 10-s averages were computed, resulting in 180 data sets for systolic, mean, diastolic, and pulse pressures of IAP and NAP and similarly for IAP and FAP (fig. 13). The period of the averaging was chosen to be between 1 s (or approximately no beat averaging) and 1 min, which we consider too slow for monitoring. Monitors generally apply such averaging, to avoid jumpy readings, without the risk of missing events. In figure 13(“patient”) an example is given for IAP and NAP; systolic, mean, diastolic, and pulse pressures were all treated in the same manner. The systolic and diastolic NAP 10-s values (column B, b¹… b¹⁸⁰, fig. 13) were plotted versus  the corresponding IAP 10 s values (column A, a¹… a¹⁸⁰) as individual scatter plots (fig. 4in the main text of this study) and Pearson's correlation coefficients were calculated over these columns. Median and interquartile range of the correlation coefficients in the group were represented in table 2of the main text. Consistency of the data were assessed by dividing column D in three equal parts of a length of 60 NAP − IAP differences and calculating Cronbach's α and the intraclass correlation coefficient (ICC). Internal consistency by Cronbach's α is considered excellent for α≥ 0.9, good for 0.9 > α≥ 0.8, acceptable for 0.8 > α≥ 0.7, questionable for 0.7 > α≥ 0.6, poor for 0.6 > α≥ 0.5, and unacceptable for 0.5 > α. Since the ICC calculates contribution of the variability in the differences in a patient relative to the total variability, the ICC is sensitive to the range of values in the group. The variability in the differences is rather low, resulting in relatively low ICC values.

Next, we followed a methodological approach, which is an extension of the traditional Bland and Altman representation of data.19When two methods are compared, the average of each measurement pair is the x value and their difference is the y-value. For each patient the mean of the NAP and IAP averages is accompanied by an SD (in the x direction), and similarly, the mean of the NAP − IAP differences is accompanied by an SD (in the y direction). In this way the individual data can be represented in Bland-Altman fashion,19which allows assessment of both individual differences (patient-related) and of group differences (measurement system-related) including the range over which these were determined. Work of Bland and Altman includes a method for calculating the confidence of the limits of agreement with multiple measurements per individual but did not focus on simultaneously displaying both individual and group statistics.33Thus, each patient contributed one point with horizontal and vertical error bars (fig. 5). The x- and y-coordinates for each patient in the Bland-Altman plot were obtained as follows (fig. 13). The average (I) over column C, thus over the (NAP + IAP)/2 averages, gave the x-coordinate of a patient in the Bland-Altman plot. The average (K) over the column D, thus over the NAP − IAP differences, gave the y-coordinate. The error bars in the Bland-Altman plot were the SDs over the same columns: J over column C for the x-coordinate, and L over column D for the y-coordinate. The values E and G and I, J, K and L of each patient were then used in the calculation of the group statistics.

The 50 patients all contributed their statistics mentioned above; see figure 13(“group”). The average and SD over the IAP, NAP, and NAP − IAP columns, thus over E (e¹… e⁵⁰), G (g¹… g⁵⁰) and K (k¹… k⁵⁰) values of each patient were represented in table 4in the main text. Of column K, the average (U) and SD (V) were used to assess the compliance with the Association for the Advancement of Medical Instrumentation criteria (average and SD over a group of patients smaller than 5 ± 8 mmHg).20 

The average (S) and SD (T) over the “SD (NAP + IAP)/2” column (J) gave the within-subject variability; average (W) and SD (X) over the “SD (NAP − IAP)” column (L) gave the within-subject precision.21These values are the group statistics describing what is visually represented in the Bland-Altman plots (fig. 5) and are represented in table 2in the main text. Comparing tables 2and 3with table 4gives insight into the performance in an individual and in a group, respectively.

To determine pressure-related differences in the NAP − IAP values (column D in fig. 13), systolic, diastolic, mean, and pulse pressures were divided in quartiles according to their respective levels (column C) as well as according to MAP and heart rate and tested by ANOVA. No differences were detected (see Results of the main text).