Cardiac output measurements via echocardiography versus thermodilution: A systematic review and meta-analysis

Echocardiography, as a noninvasive hemodynamic evaluation technique, is frequently used in critically ill patients. Different opinions exist regarding whether it can be interchanged with traditional invasive means, such as the pulmonary artery catheter thermodilution (TD) technique. This systematic review aimed to analyze the consistency and interchangeability of cardiac output measurements by ultrasound (US) and TD. Five electronic databases were searched for studies including clinical trials conducted up to June 2019 in which patients’ cardiac output was measured by ultrasound techniques (echocardiography) and TD. The methodological quality of the included studies was evaluated by two independent reviewers who used the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2), which was tailored according to our systematic review in Review Manager 5.3. A total of 68 studies with 1996 patients were identified as eligible. Meta-analysis and subgroup analysis were used to compare the cardiac output (CO) measured using the different types of echocardiography and different sites of Doppler use with TD. No significant differences were found between US and TD (random effects model: mean difference [MD], -0.14; 95% confidence interval, -0.30 to 0.02; P = 0.08). No significant differences were observed in the subgroup analyses using different types of echocardiography and different sites except for ascending aorta (AA) (random effects model: mean difference [MD], -0.37; 95% confidence interval, -0.74 to -0.01; P = 0.05) of Doppler use. The median of bias and limits of agreement were -0.12 and ±0.94 L/min, respectively; the median of correlation coefficient was 0.827 (range, 0.140–0.998). Although the difference in CO between echocardiography by different types or sites and TD was not entirely consistent, the overall effect of meta-analysis showed that no significant differences were observed between US and TD. The techniques may be interchangeable under certain conditions.

Introduction Continuous or dynamic cardiac function monitoring plays a crucial role in the diagnosis, assessment, treatment, and prognosis of critically ill patients. Cardiac output (CO) measurement is one of the most important parameters in cardiac function monitoring. The commonly used CO measurement methods include indirect Fick methods, thermodilution (TD), Doppler ultrasound (US) or echocardiography, partial carbon dioxide (CO 2 ) rebreathing, thoracic electrical bioimpedance, and magnetic resonance imaging [1,2]. TD via the pulmonary artery (PA) catheter is still considered to be the gold standard method in the clinical setting. However, this method has disadvantages because it is invasive and can lead to severe complications [3]. Echocardiography, as a noninvasive or semi-invasive method for the assessment of cardiac anatomy and function, is favored in clinical practice. The methods commonly used for echocardiography include transthoracic echocardiography (TTE), transesophageal echocardiography (TEE), ultrasonic CO monitor (USCOM), noninvasive continuous CO system (NICO), and ultrasound dilution (UD). Several sites can be used for CO measuring. The velocity time integral (VTI) and cross-sectional area (CSA) of the ascending aorta (AA) [4], PA [5], aortic valve (AOV) [6], mitral valve [7], or left ventricular outflow tract (LVOT) [8]can be used to calculate the stroke volume (SV) using CSA×VTI and the CO = SV×heart rate. Simpson's rule [9] was the first method to delineate the innermost endocardial border of the left ventricle using the trackball at end systolic and end diastolic and then to calculate the left ventricular end-diastolic volumes (LVEDV) and left ventricular end-systolic volumes (LVESV): CO = (LVEDV-LVESV)×heart rate. Some studies used the common carotid artery in point-of-care US to estimate the CO [10]. UD [11] technology is also used to measure hemodynamic variables based on the Stewart-Hamilton principle. This method utilizes an extracorporeal arteriovenous tubing loop (AV loop) inserted between existing arterial and venous catheters and isotonic saline as an indicator [12].
Whether echocardiography can replace TD method in CO measurement remains controversial. Some studies [13][14][15] have revealed that echocardiography is a rapid, accurate, and noninvasive monitoring technology suitable for patients in ICU. Although differences were observed, some studies [10,16] showed the correlation was good. However, some studies [17,18] suggested that echocardiography is not interchangeable with TD for measuring CO. Therefore, this study aimed to evaluate the consistency and interchangeability of cardiac output measurements in US and TD and to find the most optimal types or sites used of echocardiography for CO measuring if possible.

Materials and methods
This review was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis of diagnostic test accuracy (PRISMA-DTA) statement. Ethics committee approval was not required, as it was a review of published data.

Study selection
The inclusion criteria were (1) critically ill patients, (2) clinical trials, (3) studies that used echocardiography to measure the CO, (4) studies that used TD technique as the reference technique, and (5) studies in which outcomes of interest included the data of CO or all the differences between the techniques (bias) and standard deviations (SDs) or bias and limits of agreement (LOA). The exclusion criteria were (1) reviews or case reports, (2) animal studies, (3) studies published in languages other than English, (4) studies only published as an abstract, and (5) studies with no mean and SD of CO and without bias and LOA/SD between two techniques.
Study selection and data extraction were performed by two independent reviewers (YZ and JS). Disagreements were resolved through consensus.

Data extraction
A data collection form was developed prior to data extraction. Two authors (YZ and YW) extracted relevant data from included articles. The extracted data included (1) first author and year of publication; (2) number of patients, sex, and age; (3) the data of CO in both groups; (4) the type of ultrasound and sites; (5) the bias, SD, LOA, and percentage error (PE); (6) the Pearson R coefficient and linear equations; and (7) patient population.
When the results of the trial were reported as median and quartile, the Stela Pudar-Hozo method was used to estimate the mean and standard deviation. Bias was defined as the mean of the two measurement differences, and LOA was defined as bias±1.96SD (some studies defined LOA as bias±2SD). The PE was defined as 1.96SD divided by the mean CO of the two methods. Posteriori probability was also calculated.
Ethics approval was waived for this study as patient consent was obtained within the individual trials and all data were anonymized.

Quality assessment
Studies with critically ill patients who needed CO monitoring were included. The CO measured by thermodilution was the reference standard, regardless of other modes of CO monitoring. Study quality was assessed using QUADAS-2, which was tailored for our systematic review (S1 Table). The quality of each paper was evaluated by two authors (YW and JS) independently, and any disagreements were resolved by consensus.

Statistical methods
The systematic reviews were conducted in compliance with the PRISMA guidelines. The Review Manager Software version 5.3 for Windows (The Cochrane Collaboration, 2014) was used to perform the meta-analysis. The STATA version 12.0 (StataCorp, College Station, TX, USA) was used to analyze the publication bias (Egger's test). The Cochrane Q-test was used for heterogeneity analysis. A fixed-effect model was used when P>0.1 and I 2 <50%; otherwise, a random effects model was adopted. If necessary, a sensitivity analysis was performed to reduce the heterogeneity to P>0.1 and I 2 <25% by omission of some studies as few as possible. All Pvalues were two-sided, and P<0.05 was considered significant.

Search results
Of the initial 808 records identified, 676 remained after duplicates were removed. Then, 477 records were excluded based on the title, and 83 records excluded based on the abstract; 116 articles were evaluated in full text. Forty-eight full-length articles were excluded, including 33 articles with missing data and 15 articles that used TD and US techniques, but not CO measurement, in the optimization of cardiac preload. Finally, 68 and 43 studies were included in qualitative synthesis and quantitative synthesis, respectively (meta-analysis) (Fig 1). All studies were published between 1971 and 2018 (Table 1).

CO evaluation using different types of Doppler
Of these included studies, there were 41 studies with 49 CO measured results, and 1522 patients were included in the meta-analysis; no significant differences were observed between

CO evaluation at different sites
In six studies, the AA was used to measure CO by Doppler, and six studies were included in the meta-analysis. Significant differences were observed in the use of US at AA with TD (ran-

Bland-Altman analyses and regression analyses
In all studies, the median of bias between US and TD was -0.12 (ranged from -2.50 to 3.04 L/ min). The median of LOA was 0.94 L/min (ranged from ±0.05 to ±4.72 L/min). Twenty-eight studies reported that the PE and the median were 24.3% (ranged from 7.2% to 53%). The median of R (correlation coefficient) was 0.827 (ranged from 0.140 to 0.998). The slope ranged from 0.126 to 1.58, and the intercept ranged from -1.91 to 3.55 in the 35 regression equations ( Table 1).

Publication bias
The

Discussion
In this systematic review, we included 68 studies, of which 43 studies reported data on CO measurement and were included in the meta-analysis. The overall effect showed that no significant difference was observed between echocardiography and TD in measuring CO; the subgroup analysis showed no significant differences in the different types. In all sites, the difference was founded only in AA. Further, the sensitivity analysis showed no change in the results. However, there was a wide range in bias, LOA, PE, and correlation coefficient of the two technologies and was beyond the clinically acceptable range in some studies.
In these different types of echocardiography, the sensitivity analysis showed that the TEE, USCOM, and UD had hairline bias (�0.1 L/min) with TD; TEE had the lowest standard error and maximum weight followed by TTD. The UD had a small mean difference, but with wide 95% CI. CO could be easily overestimated using TTE instead of using TD, and underestimated using other types of echocardiography (ACM, NICO, point-of-care US, and Doppler PA catheter) despite the differences not having statistical significance. Therefore, we speculate that TEE was the preferred method because it can obtain more accurate results in the measurement of CO. Considering that TEE is semi-invasive and UD can only be used for specific population, USCOM can be the first choice for noninvasive echocardiography for the measurement of CO.
The sites of the ultrasonic probe will also have an effect on the test results. In our meta-analysis, no statistical difference was found except for the measurement of CO in the AA. The lowest mean difference of CO comes from PA, compared with the TD, followed by AV loop. However, the method of measuring PA CO used a PA catheter ultrasound probe and was an invasive procedure [59]. The AV loop used in UD is also based on an invasive procedure [28,33,35,68]. In these studies, more researchers were willing to measure CO in the AOV and LVOT, with mean differences of <0.1 L/min and a narrower LOA. Therefore, the AOV and LVOT as the recommended sampling locations for CO detection are feasible. This finding is also consistent with the recommendation of the American ultrasound guidelines [78].
In all these studies, the largest bias and LOA (bias = 3.04 L/min, LOA = ±9.44 L/min) were found in a study with the ABCOM 1 transtracheal Doppler (TTD) versus TD [34] and with the lowest correlation coefficient (R = 0.14). TTD system requires a special TTD endotracheal tube, in which the tip was embedded with an ultrasonic probe; it can only be used in patients with mechanical ventilation. In this study, seven patients with 36 simultaneous measurements were compared. We speculated that the TTD measurements accounted for most of the between-technique variability. Obtaining and maintaining good Doppler signals were difficult and time-consuming in TTD and were considered possible causes of error. Hausen et al. [37] also compared TTD and TD. They suggested that the TTD system does not provide accurate CO determinations (bias = 1.70 L/min, LOA = ±3.29 L/min, R = 0.248) and that several reasons can affect its accuracy and restrict its wider use, such as cuff deflation for >10 min, which cannot be tolerated by ICU doctors, sensitivity to patient movement, and that an optimal signal cannot often be attained if the probe is not in the appropriate place.
The bias in three studies [2,21,25] was > 2.0 L/min. One of these studies[21] used the ACM to monitor the CO in patients with high cardiac output (pregnant and pre-eclamptic women) and found that it was inaccurate compared with TD. Another study [2] found that the mean difference was larger in the spontaneous ventilation state than the intermittent positive pressure ventilation and apnea state. One possible reason was that the patients were not sedated during spontaneous ventilation; thus, the CO was increased. Cariou et al.
[25] compared the descending aortic blood flow using a pulse Doppler velocimeter with CO. Although the authors thought that the descending aortic blood flow determination had good correlation and consistency with TD in CO and that descending aortic blood flow provided a reliable noninvasive tool for estimating CO, the mean difference was obviously due to the descending aortic blood flow as a fraction of the CO.
Critchley and Critchley [79] thought that it can be acceptable clinically when the PE is <30%. They suggest that if the PE between the two methods is �±30%, then the two methods are interchangeable. In our studies, although the median of PE was 23.8%, the PE of the six studies [2,8,18,24,43,51] was >30%. Missant et al. [51] used Doppler echocardiography during off-pump coronary artery bypass grafting and believed that Doppler echocardiography was not always feasible when the heart was displaced from the esophagus and had lower accuracy; The accuracy in CO measurement may have been affected in three studies that included special patients or scenarios (Knirsch et al. [43] had used USCOM in children with congenital heart disease; Botero et al. [24] used NICO during cardiopulmonary bypass; and Castor et al. [2] used NICO in patients with low sedation levels during spontaneous ventilation). Moller-Sorensen et al. [18] thought that the possible explanation was that the SV is calculated from two variables (LVOT, CSA, and the VTi); measurements were made irrespective of the ventilatory cycle, arrhythmias, and the patients with different scenarios. Therefore, we should pay more attention to the evaluation of ultrasound CO results, when the cardiac function or physiological structural change occurs in some patients with heart disease or in special situations.
Moreover, imprecision in echocardiography CO measurements may be induced by technical or operator factors. By improving the operation level, repeated measurement may reduce the measurement error. In our meta-analysis, self-control was used in all studies, and most of the studies used repeated measurements and blinding methods. Therefore, the quality of literature was not evaluated.
Other limitations of this study include the following: (1) no further subgroup analysis was conducted on the research subjects and disease types due to the limitation of data integrity and the diversity of diseases; (2) determination of the best CO test method was difficult, as both have advantages and limitations; (3) the linear equations were not overfitted because finding a general linear equation to express the relationship between US and TD in CO measurement for the inconsistency of the research subjects, ultrasonic type, and sites is difficult.

Conclusions
This systematic review and meta-analysis showed that the overall effect of the CO measurements by echocardiography or TD has no significant difference. TEE can be the preferred method with accurate results and USCOM can be a good choice for its noninvasiveness in CO measurement; the AOV and LVOT can be the recommended sampling location. However, in some special scenarios, such as high CO, low sedation, or with physiological structural changes, the accuracy of CO measurement by echocardiography is questionable.
Supporting information S1 Table. Tailored QUADAS-2 to assess risk of bias and applicability judgments.