This study included patients with PAH/CTEPH diagnosed according to international guidelines from 2015 (3). Patients without PH or PH classified to other diagnostic groups were excluded. In all patients, the right heart catheterization (RHC) examination was clinically indicated, and patients provided written informed consent to participate in a prospective randomized, placebo-controlled, double-blinded, triple-phase, and crossover trial evaluating acute hemodynamic effects of acetazolamide at rest and end-exercise. The study design, methods, and results on the acute hemodynamic effect of acetazolamide on PVR and other measures at rest have been published (10). This study focuses on the comparison of the CO at rest and end-exercise measured by different methods, and these results have not been previously published.

The study complies with the declaration of Helsinki and was approved by the local ethics committee (BASEC 2016-00089, and the trial was registered at ClinicalTrials.gov (NCT02755259).

Assessed data were inspected visually in the LabChart program for plausibility, and artifacts were deleted. Continuously registered data were averaged over periods of 15 seconds. Where appropriate, i.e., during a steady state, missing data points were carried forward. The results are expressed as means ± SD unless indicated otherwise. The normality assumption was tested using the Shapiro-Wilk test. Heteroscedasticity was assessed using the studentized Breusch-Pagan test. The agreement and bias were assessed by the method described by Bland and Altman (15). Bias was expressed as the mean of the differences obtained by different techniques (e.g., oDF-DF and TD-DF) with its 95% CI. The limits of agreement are expressed as mean ± 1.96 SD. Percentage error was derived by dividing the limits of the agreement by the mean of both methods. The coefficient of variation was obtained by dividing the SD of a method by its respective mean. The repeatability coefficient was calculated according to Carstensen et al. (16) for linked or paired replicates under the assumption of identical conditions. The paired data structure also had an influence on further statistical parameters. All statistical analyses were performed using RStudio (Version 1.2.1578, R Studio Inc., San Francisco, CA, USA). Acceptance criteria were defined as limits of agreement within 2 L/min and a percentage error of 30% (17).

We included 24 patients (7 female) with PH (CTEPH, n = 17; PAH, n = 7). Patient characteristics and baseline resting hemodynamics are shown in Table 1. The mean age was 59 ± 14 years, body mass index (BMI) 27.9 ± 4.6 kg/m², heart rate (HR) 75 ± 11 bpm, and mPAP 37 ± 12 mmHg.

The average CO of both methods at rest, end-exercise, and both combined, and further statistical parameters are summarized in Table 2. The regression equation including data at rest and end-exercise had a slope of 1.07 with an intercept of −0.50 L/min and an R² of 0.92 (Figure 1A). The normality assumption was met (p = 0.64).

Resting values: The distribution of CO values is illustrated in Figure 2A. Bland-Altman analysis showed a bias of −0.26 L/min (95% CI, −0.62 to 0.09 L/min) and lower and upper limits of agreement of −1.92 and 1.39 L/min, respectively (Figure 3A). The percentage error was 28.6%. The coefficient of variation and the repeatability coefficient were 25.3% and 2.03 L for DF, respectively, 25.9% and <0.01 L for DFp.

End-exercise values: The distribution of CO values is illustrated in Figure 2B. The bias was 0.29 L/min (95% CI, −0.08 to 0.67 L/min). Likewise, the limits of agreement showed no drastic changes, with the lower limit at −1.25 L/min and the upper limit at 1.83 L/min (Figure 3B). The percentage error decreased to 16.1%. Compared to rest measurements, the coefficient of variation was increased marginally while the repeatability coefficient was completely inversed, being 0.99 L for DF and 1.72 L for DFp.

Resting and end-exercise values: Including both resting and end-exercise measurements, the bias showed no noteworthy change [0.01 L/min (95% CI, −0.35 to 0.37 L/min)]. Similarly, limits of the agreement remained comparable (−1.67 to 1.69 L/min). The percentage error was 22.3%, which equals a decrease of 6.3% compared to the resting value.

The average COs of both methods at rest, end-exercise, and both combined, and further statistical parameters are summarized in Table 2. The regression equation for TD and DF, i.e., rest and end-exercise, had a slope of 1.07 with an intercept of −0.20 L/min and an R² of 0.72 (Figure 1B). However, neither normality (p < 0.05) nor homoscedasticity (p < 0.05) were met.

Resting values: The distribution of CO values is illustrated in Figure 2C. The bias was −0.45 L/min (95% CI, −0.98 to 0.08 L/min). The limits of agreement were −2.93 and 2.03 L/min, respectively (Figure 4A). The percentage error was 44.6%, which equals an increase of 16.0% compared to DF vs. DFp. The coefficient of variation and the repeatability coefficient for TD were lower than that for DF.

Exercise values: The distribution of CO values is illustrated in Figure 2D. A bias significantly different from zero was detected with 1.39 L/min (95% CI, 0.37–2.42 L/min. The lower and upper limits of agreement were −2.84 and 5.62 L/min, respectively (Figure 4B). The percentage error was 41.6%, which equals a decrease of 3.0% compared to resting measurements. The coefficient of variation and the coefficient of repeatability were increased and were comparable to that of DF.

Resting and exercise values: Including both rest and end-exercise, the bias was 0.36 L/min (95% CI, −0.42 to 1.13 L/min). The limits of agreement were −3.28 and 4.00 L/min, respectively. The percentage error was 47.2%. Both, the coefficient of variation and the repeatability coefficient were increased more in relation to their counterparts of DF.

Our comparison between the DF and DFp methods showed no displacement of the regression line over the line of identity (Figure 1A), and the mean of the differences was not significantly different from zero (Figures 3A,B), which indicates that adequate accuracy/bias of DF is compared to DFp. Also, the percentage error and the limits of agreement were within predefined acceptance criteria of 30% and 2 L/min (Table 2).

At the end-exercise, the percentage error between DFp and DF was decreased drastically while the absolute limits of the agreement remained the same (Table 2), reflecting an even better precision for the DFp method under more strenuous work and therefore higher CO values. These end-exercise results may even better reflect the precision of DFp to DF, as in our study, setting arterial and mixed venous blood gases were sampled in the supine resting position, whereas VO₂ was obtained starting 10–15 min after the patient was set in a stable semi-supine position. This might have influenced the limits of agreement between those two methods at rest. However, this time lag was inevitable to avoid too many blood samples.

Overall, the bias, limits of agreement, and percentage error between CO measurements by DFp and DF were within an acceptable range at rest and good at end-exercise measurements. The coefficient of variation remained approximately the same at rest, end-exercise, and both combined, with the reference method (DF) always showed a smaller value and allowed a rational conclusion of the Bland-Altman plot (22). It suggests that both methods contributed almost equally to agreement/precision in the comparison. Potential oximetry pitfalls, such as hypoperfusion, shivering, or movement artifacts during exercise, cannot be excluded having influenced the coefficient of variation (27, 28). The course of the repeatability coefficient and coefficient of variation were found slightly discordant, and the coefficient of repeatability varied during exercise. This may be explained by the crossover study design and our assumptions of interchangeability of the replicates between the three phases and the resulting long pauses between them. The DF as a gold standard method to assess CO requires simultaneous blood samples from an artery and the catheter-tip, which requires trained staff that should handle the samples without delay under avoidance of air admixture. Given the fact that DFp does not have these drawbacks and allows for a simultaneous mPAP and therefore pressure/flow measurement, it is a promising alternative to DF and maybe even less prone to errors in case of repetitive assessments during exercise (24). Studies compared other non-invasive methods for CO measurements, such as impedance cardiography to DF, which showed acceptable agreement and precision compared to DF or TD in patients with PH within a CO range of 3–7 L/min (29, 30). Further studies focused on the indirect Fick method to assess CO, which does not measure VO₂ but assumes it from nomograms (31). These studies concluded that the LaFarge method was the most precise, but the indirect Fick method was associated with large errors and cannot be recommended (3, 32–36). However, those nomograms are not intended for exercise protocols and would possibly be associated with misdiagnosis of PH.

Thermodilution requires repetitive similar injections of cold saline, which needs time and precludes simultaneous measurement with mPAP, which may introduce bias to pressure/flow measurements in a biologically variable system, especially during exercise (37–39). Additionally, in comparison to DF, TD is more imprecise for lower CO values and when valvular regurgitation is present (40)—conditions that are frequent in patients with PH (9, 41). As simultaneous assessments are challenging, there have been only a few reports on the agreement between TD and DF in patients with PH, (36) especially during exercise.

The additionally performed CO measurement by intermittent TD did not reveal an acceptable overall agreement with DF during high CO levels, as obtained at the end-exercise. Neither limits of the agreement nor percentage errors were within predefined criteria (Table 2, Figures 4A,B). Although the present study revealed that the limits of agreement between TD and DF were relatively wide at rest, the regression line and its confidence interval showed no displacement above the line of identity over a broad range of CO values (Figure 1B). Furthermore, the coefficient of variation was lower than that of the reference method at rest, which renders the rational interpretation of the Bland-Altman plot questionable. A potential explanation for the discrepancies in precision and accuracy would include the response time of the measurement technique, which is aggravated by the time needed for triplicate cold saline injection. Additionally, the limitation of time lag as discussed in the comparison of DFp to DF may also have influenced the measurements.

Measurement of CO is even more challenging during stepwise incremental exercise, where steps are suggested to be at least 3 min to reach a steady state (42). Our study shows that agreement of TD to DF is lower at end-exercise than at rest, suggesting a systematic error at high CO values (43). Additionally, this finding indicates that patients with PH presumably need more than 3 min to reach a steady state. This would support the findings of Lador et al., who investigated “CO kinetics at the onset of exercise in patients with PH” and observed that up to 5 min are warranted to reach a steady state (44). Another study showed that mPAP/CO slopes calculated from CO by TD lead to significant overdiagnosis of exercise PH compared to CO measured by DF, (45) potentially related to exercise-induced right-to-left shunting, which is commonly found in patients with PH (46). However, in the present study, CO was rather overestimated by TD compared to DF at end-exercise (Figure 4B). Our and other studies' results thus suggest that accuracy and precision of the CO measurement during stepwise incremental exercise protocols by the TD method compared to the DF method is lower and thus particularly absolute CO values measured by this method must be interpreted with caution and in the context of clinical practice and research (47).