Determination of cardiac output from pulse pressure contour during intra-aortic balloon pumping in patients with low ejection fraction

The IAB catheters (Narrowflex 8F LightWAVE; Arrow Intl, Reading, PA) equipped with a fiber-optic high-fidelity pressure transducer at the tip were positioned under transesophageal echo control in the descending thoracic aorta at 2 cm distal to the left subclavian artery. The IAB catheters were connected to an IABP device (AutoCat 2 Wave, Arrow Intl). Standard procedures were followed to time inflation and deflation of the IAB. Thermodilution catheters (Edwards Lifesciences, Irvine, CA) were inserted into the pulmonary artery. Conductance volume catheters equipped with a high-fidelity micromanometer (F7; CD Leycom, Hengelo, The Netherlands) were positioned into the LV along the long axis via a pulmonary vein. The conductance pressure–volume catheters were connected to a cardiac function analyzer (CD Leycom) to measure ventricular volumes [8, 9, 14] using dedicated software (Conduct NT v3.18, CD Leycom). The conductance volume method is based on measuring time-varying electrical conductance of 5 to 7 ventricular blood segments, delineated by selected catheter electrodes. Correct positioning of the pressure–volume catheter was verified by transesophageal echocardiography and by inspection of the segmental volume signals. Parallel conductance was assessed by injection of 10 ml hypertonic saline (6%) into the pulmonary artery [8, 9, 14]. Absolute LV volumes were calculated by matching effective conductance SV with simultaneously measured thermodilution SV and by subtracting parallel conductance from total conductance volume.

The WKflow method uses a nonlinear, time varying three element Windkessel model, similar to the Modelflow method [15]. The model includes aortic characteristic impedance (Z0), arterial compliance (Cw), and systemic vascular resistance (Rp) (Fig. 1). Time varying Z0 and Cw depend on the aortic cross-sectional area, which can be estimated from time varying arterial pressure, age, and gender by means of the arctangent model of Langewouters et al. [16]. To obtain the correct value of Rp for the beat under study a number of steps occur. First, model aortic blood flow is simulated with Rp from the former beat. Simulated blood flow during diastole is set to zero (assuming no aortic valvular leakage in this phase). A time varying aortic pressure is calculated with the simulated flow and the Windkessel parameters Rp, Cw and Z0. Next to the beat under study, Rp is adjusted by an optimization procedure in which the difference between simulated and measure aortic pressure at the end of the beat is minimized. So the optimization does not include aortic pressure during augmentation. With this scheme, the model can follow changes in systemic peripheral resistances that occur with a time constant which is typically about 10 s. After optimization the dicrotic notch is detected at the first local minimum in the aortic flow signal after peak flow [8, 9, 17, 18] (Fig. 1). Integrating the flow from pressure upstroke to dicrotic notch gives stroke volume (area under flow curve). Cardiac output is found by multiplying stroke volume by heart rate. Furthermore, the simulated aortic pressure can be used to calculate mean, diastolic and systolic aortic pressure.

In the absence of mitral regurgitation the amount of blood ejected by the left ventricle, i.e. stroke volume is equal to the change in LV volume during systole. With the WKflow method we accurately define begin of systole by the aortic pressure upstroke and the end of systole by detecting aortic valve closure from the simulated aortic flow pattern [8, 9, 17, 18].

Measurements were performed during vena cava occlusions before or after cardio-pulmonary bypass during IABP at a 1:2 assist ratio and in a subgroup of patients with IABP not pumping. ECG, LV pressure (Plv), aortic pressure (Pao), intra-aortic balloon pressure (Piabp) and LV volume (Vlv) signals were measured and digitally sampled at a rate of 250 Hz and stored on computer disk. Data recorded during marked arrhythmia, was used to evaluate the robustness of the WKflow model.

After confirming a normal distribution of data with the Kolmogorov–Smirnov test, agreement between SVlv, SVwf, as well as agreement in changes in stroke volume was evaluated with Bland–Altman statistics. The agreement between SVlv and SVwf was computed as the bias (i.e. low bias is high accuracy) and standard deviation (SD) (i.e. low SD high precision), with the limits of agreement (LOA) computed as the bias ± 2SD. Percentage precision or coefficient of variation (COV) was computed as [COV = 100 * (SD/mean)]. Low COV represents high precision. In addition linear regression between SVlv and SVwf is performed. The slope and goodness of fit by assessing R² are calculated. After confirming equality of variance by Levene’s statistical test of the results of VCO with the IABP in 1:2 mode and the results of VCO without IABP were merged.

In each individual patient there is a strong relation between SVlv and un-calibrated SVwf (Fig. 3). During IABP in 1:2 assist the slope of the augmented beats, 0.994 ± 0.107, is not significantly different from the following assisted beats 0.994 ± 0.053 (p = 0.852). Combining all beats during IABP the slope is 0.997 ± 0.088, correlation coefficient R² is 0.891 ± 0.070 and not different (p = 0.912) from unity. For the observation made without IABP the averaged slope is 0.978 ± 0.128 and R² is 0.899 ± 0.108. This slope is not different from unity, p = 0.661.

The results of Bland–Altman analysis comparing SVlv and non-calibrated SVwf during vena cava occlusion with IABP in 1:2 mode are shown in Table 2. Bias between SVlv and SVwf is 1.04 ± 3.99 ml (small but significantly different from zero, p = 0.007). The limits of agreement (LOA) range from − 6.94 to 9.02 ml. The COV or precision is 10.9%. The results of Bland–Altman analysis during vena cava occlusion without IABP indicate an accuracy of 0.48 ± 4.36 ml (not significantly different from zero, p = 0.667) and precision is 11.6%. The LOA ranged from − 8.24 to 9.20 ml.

Between the group with IABP in 1:2 mode and the group without IABP we could not detect a difference in homogeneity of variance (Levine statistics p = 0.647) and in mean difference SVlv–SVwf, allowing to group all results together. The results of all VCO show a good overall agreement between SVlv and SVwf (Fig. 4 and Table 2).

We performed one calibration per patient of SVwf by thermodilution. During IABP support bias or accuracy improved to 0.14 ± 3.07 ml, COV or precision to 8.3%, and LOA to − 6.00 to 6.28 ml. After calibration bias is not different from zero now (p = 0.630).

Without IABP support, calibration by thermodilution results in a markedly improvement of the agreement between SVlv and SVwf (bias 0.01 ± 2.71 ml, precision 7.5% and LOA − 5.41 to 5.43 ml). Bias is not different from zero (p = 0.331). Between the group with IABP in 1:2 mode and the group without IABP we could not detect a difference in homogeneity of variance (Levine statistics p = 0.242) and in mean difference SVlv–SVwf, allowing to group all results together. The results of Bland–Altman analysis for all observation grouped is given in Fig. 4. With no significant bias (p = 0.242) precision is 7.8% and percentage limits of agreement for the difference is 2  × 7.8 = 15.6%.

The results of all beat-to-beat changes in SVlv and changes in SVwf normalized to the start of vena cave occlusion are given in Fig. 5. We found no difference in the change of SVlv and change of SVwf (p = 0.820). Regression analysis of the data for all VCO’s showed a slope of 1.031 and a goodness of fit R² = 0.869.

In Fig. 5 the data points in the lower left and the upper right quadrants represent equal direction of change (correct). The data points in the upper right and lower right quadrants indicate opposite changes (incorrect). Most data points are positioned in the left under quadrant indicating that vena cava occlusion diminishes both SVlv and SVwf. Counting the number of data points with an equal direction of change in percentage of the total number of data point delivers the level of concordance between the methods. For the data presented in Fig. 5 we found concordance in 96% of the cases. This score improves if clinically irrelevant changes in absolute values < 5% are excluded from counting to 99%.

During irregular heart beats and IABP assist at a 1:1 ratio we observed marked beat-to-beat changes in SVlv and SVwf varying from 20 to 85 ml, Fig. 6. Nine recordings from six different patients were analyzed with the two methods. Bland–Altman analysis for agreement between the two methods showed an accuracy or bias of 0.828 ± 3.287 ml, precision 7.6% and LOA from − 5.75 to 7.40 ml, see Fig. 7.

Thermodilution is the reference CO (COtd) method for steady state circumstances. A single thermodilution estimate of CO or SV has a COV or precision of 15–20%, while a triplicate injection technique has an error of 5% when the injections are made at different phases of the ventilatory cycle [19, 20]. It is against these levels of inherent error that any new method of CO or SV measurement should be judged [21]. However, the clinical circumstances of the critically ill patients with or without arrhythmia and in need for IABP often preclude extended hemodynamic steady state episodes, and exclude therefore a direct comparison between stroke volume by WKflow and thermodilution. Moreover during these rapid changing hemodynamics, mostly due to arrhythmia, there is a strong need for an actual continuous insight, which can only be given then by a beat-to-beat method. Therefore, we have chosen for a beat-to-beat comparison between WKflow method and the beat-to-beat left ventricular conductance volume method. The conductance method was per patient once calibrated by a triplicate phase spreading thermodilution technique [19, 20] during a relative stable episode. With this approach the effect of hemodynamic instability and arrhythmia is expected to be similar for SVlv and SVwf [14], which was uniformly observed see Fig. 6.

Bland–Altman analysis for the difference between SVlv and non-calibrated SVwf resulted in percentage limits of agreement (LOA) of ± 23.2%. According to Critchley and Critchley [21] this percentage LOA is within the 30% limits of acceptability for clinical use. Because the conductance method and the WKflow method are physically independent of each other the error of the comparison C of two estimations A and B can be computed as Variance (C) = Variance (A) + Variance (B) [19, 20].

Kornet et al. [22], studied the accuracy of the conductance method to determine SV in comparison to thermodilution and estimated an error of around 10% for SVlv. The error we found for the difference between the SVwf and SVlv is on average 12% (Table 2, all). Thus we can conclude that SVwf method is responsible for an error of approximately √144–100 = 7%, which is comparable to the probable error for standard SV by thermodilution and of SVlv and may therefore replace the standard thermodilution method and conductance method. While the performance of SVwf is acceptable, after one thermodilution calibration per patient of WKflow the precision of the difference between SVlv and SVwf decreases to 7.8% (Table 2). Moreover, apart from quantitative estimation of SV, we observed an excellent agreement between SVlv and SVwf measures for directional SV changes, equal for calibrated and uncalibrated values, demonstrating that the SVwf can reliably track trends in SV as well as in CO.

In general pulse contour methods use the radial or femoral artery pressure as input for their calculation. This assumes that the radial or femoral pressure represents the pressure in the aorta and that changes in waveform morphology are related to corresponding changes in blood flow. However, profound differences between aortic and peripheral pressures have been reported especially during periods shortly after return from cardiopulmonary bypass [23, 24]. We previously showed [20] that this resulted in significant outliers in pulse contour CO values. Hatib and colleagues [25] showed that the radial and femoral arterial pressure may decouple from the aortic pressure. Using the WKflow method and solid state aortic pressure measured on tip of the IAB catheter, instead of more peripheral arterial pressure measured with a fluid filled manometer system solves this potential source of error.

A clean and high fidelity pressure signal from the tip-transducers is mandatory for precise detection of the start and end of systole. Both beat-to-beat SVlv and SVwf rely heavily on these beat-to-beat points in time. The start of systole by the WKflow method is defined as the time of the last local minimum in the arterial pressure before the start of the steep increase in arterial pressure. To detect the end of systole from arterial pressure several methods were proposed. The WKflow method which we previously used in the dicrotic notch detection algorithm [8, 9, 17, 18] and used in this evaluation detects the moment of aortic valve closure by identifying the first local minimum after peak in simulated aortic blood flow. Thus the WKflow software accurately detects begin and end of systole both during regular and irregular heartbeats. Therefore, we expect our estimates of both SVlv and SVwf to be reliable.

Because many heart failure patients frequently show arrhythmia, a new method should be tested also in this condition. Based on our analysis during periods of arrhythmia we conclude that if the IAB is properly timed to heart irregularities [9] then a reliable determination of beat-to-beat stroke volume is possible with this new WKflow method.

A potential limitation of the method is the patient selection. Similarly to the standard radial/femoral/finger arterial pulse contour methods, it is required that patients have a competent aortic valve and no aortic aneurysms which are both contraindications to IABP use. All patients studied had no signs of aortic valves leakage and aortic aneurysms. An aortic aneurysm may affect a patient’s arterial compliance. A competent aortic valve is needed because WKflow computes the forward flow into the aorta and ignores backward flow. For the analysis of conductance stroke volume an extra precondition is the absence of mitral valve regurgitation. We cannot exclude the possibility that some patients had small undetected valve leakages or that such leakages occurred during some periods of the observations. If so, this will lead to an overestimation of real stroke volume by WKflow during aortic regurgitation, an overestimation by the conductance volume method during mitral valve regurgitation. Furthermore, in our patients with LV aneurysm, marked LV dyssynchronic segmental volume changes were observed throughout the cardiac cycle, indicating diastolic and systolic dyskinetic and akinetic wall motions. The SV estimation by the conductance catheter is the sum of all segmental volume changes and the paradoxical segmental volume changes are consequently subtracted from total ventricular volume changes. On that way total ventricular SV estimation by including the dyssynchronous aneurysm segments will result in a valid SV estimation [26].

All our observations were carried out during cardiac surgery in patients with severe heart failure in whom a prophylactic use of IABP was advisable. Therefore, our IABP observations could only be done during episodes of minor surgical intervention.