Agreement Between Transesophageal Echocardiography and Thermodilution-Based Cardiac Output

Accurate, noninvasive assessment of cardiac output (CO) has remained an elusive goal in the fields of critical care medicine and cardiac anesthesia. Almost a half century after its development,1,2 pulmonary artery catheter-based thermodilution remains the most extensively validated, clinically relevant method of cardiac output measurement.3 However, the invasive nature and associated risks of this technique lead to a decrease in utilization,4,5 as well as the search for an accurate, noninvasive method of measuring CO. Transesophageal echocardiography (TEE), which is safe6 and increasingly ubiquitous,7 is a leading candidate to replace the PAC. But, is it accurate? Two-dimensional (2D) TEE estimates of cardiac output neglect left ventricular outflow tract (LVOT) eccentricity, and in the 3-dimensional (3D) TEE era are thought to be a correctable source of error.8 In this issue of Anesthesia & Analgesia, Graeser et al9 compared the use of thermodilution to Doppler-derived CO using both 2D and 3D TEE. Presumably, if 3D TEE provides a better estimate of the LVOT cross-sectional area than 2D TEE, Doppler-based estimates of CO would more closely agree with thermodilution. Graeser et al’s9 study is notable for its rigorous methodology. For the thermodilution measurements, 2 sets of 4 consecutive injections of 10 mL of ice cold saline were utilized (mean bias between replicates was −0.04 L/min). For the echocardiographic measurements, the velocity time integral of the LVOT was averaged over 6 consecutive cardiac cycles, and in both the 2D and 3D examinations, 2 sets of 10 LVOT diameter (2D) and area (3D) measurements were performed. Mean bias between replicates for the 2D and 3D measurements were −0.05 and 0.07 L/min, respectively. Yet, despite this attention to detail (exhibited by excellent repeatability), when using thermodilution as the reference standard, 3D TEE produced what appeared to be wider limits of agreement when compared to 2D TEE. The authors shed some light on this surprising result by randomly selecting 19 patients to undergo cardiac computed tomography angiography (CTA). As expected, they found that 3D measurements of LVOT area agreed with CTA more closely than 2D estimates. How then is it possible that 3D TEE estimates of cross-sectional area agreed more closely with CTA, yet 3D TEE estimates of CO agreed less with thermodilution cardiac output than the 2D estimates? Logic suggests that this conundrum must be the result of errors in either thermodilution or the CTA measurement. As mentioned by the authors, cardiac CTA has not been rigorously established in the literature as a reference method for determining LVOT cross-sectional area, making it possible that both 3D TEE and CTA are inaccurate methods for this measurement. Limitations with Doppler flow were also mentioned as a possible source of disagreement, but these errors would affect 2D and 3D TEE equally. Not mentioned, but worth considering, is that thermodilution is not a perfect measure of cardiac output. While the authors used iced cold saline (higher signal to noise ratio) and 2 sets of 4 repetitions, injections were made irrespective of the ventilator cycle. Yet maximal precision and accuracy of thermodilution requires choosing a consistent time point in the ventilatory cycle, as right ventricular output varies by approximately 30% depending on when in the ventilatory cycle the bolus is initiated.10 The repeatability of the reported measurements implies that any variation due to the influence of ventilator cycling was effectively averaged out. Additionally, while thermodilution is a clinical gold standard, that is, the most accurate device that clinicians use routinely, it is not a reference gold standard. Reference standards, like transit time flow probes, are highly accurate, but invasive and not used clinically. Analysis of 3 way comparisons between thermodilution, Doppler-based techniques, and true reference standards in both animals and human suggests there is minimal difference in accuracy between thermodilution and Doppler-based techniques.3 It is important to recognize that agreement studies only determine how well 2 forms of measurement agree—by their nature (2 way study), they are incapable of determining which of 2 measurements is more accurate. This is increasingly important as device studies that lack a validated goal standard have been published.11,12 A third possibility, which deserves mentioning, is that the authors, by no fault of their own, have made the “limits of agreement equivalent” of a type II statistical error. A key, under-appreciated feature of limits of agreement analysis is that there is a confidence interval around the reported limits of agreement. Examination of Figure 3 reveals relatively wide confidence intervals around the limits of agreement for both 2D and 3D TEE. In fact, based on visual inspection of these confidence intervals, it is possible that the true limits of agreement for 2D TEE could be as large as −3.1 to 1.9 L/min, and that the true limits of agreement for 3D TEE could be as small as −1.3 to 1.9 L/min, in which case we would actually conclude that 3D TEE exhibited better agreement with thermodilution (the opposite can also be said). Graeser et al’s9 study leaves much to be answered. Because of the methodology and high repeatability, we can say with some confidence that 3D TEE-based measures improve agreement with CTA-based LVOT area, but do not appear to improve agreement with thermodilution-based cardiac output. Whether or not this is due to limitations in thermodilution-based cardiac output, to limitations in the accuracy of CTA, or due to an underpowered study cannot be determined. While this “nonconclusion” is somewhat unsatisfying, it is what the data tell us. It is our hope (and expectation) that the readers of Anesthesia & Analgesia will help us make this important determination. DISCLOSURES Name: Patrick D. Millan, MD. Contribution: This author helped prepare the article. Name: Robert H. Thiele, MD. Contribution: This author helped prepare the article. This manuscript was handled by: Maxime Cannesson, MD, PhD.