Comparison of Accuracy of Estimation of Cardiac Output by Thermodilution Versus the Fick Method Using Measured Oxygen Uptake

Abstract

The thermodilution (TD) method is routinely used for the estimation of cardiac output $({\stackrel{.}{\mathrm{Q}}}_{\mathrm{C}})$. However, its accuracy, compared with the gold-standard Fick method, where systemic oxygen uptake $(\stackrel{.}{\mathrm{V}}{\mathrm{O}}_2)$ is directly measured, and ${\stackrel{.}{\mathrm{Q}}}_{\mathrm{C}}$ calculated from $\stackrel{.}{\mathrm{V}}{\mathrm{O}}_2$ and the arterio-venous oxygen difference (“direct” Fick), has not been well validated. The present study determined the agreement between TD and Fick methods in consecutive patients who underwent pulmonary artery catheterization for a broad range of clinical conditions. This is a subanalysis of a previous study comparing the indirect versus Fick method based on a prospective, consecutive patient registry of 253 patients who underwent pulmonary artery catheterization for clinical indications at a single center between 1999 and 2005. We included patients that had an estimation of ${\stackrel{.}{\mathrm{Q}}}_{\mathrm{C}}$ both by the Fick method using measured $\stackrel{.}{\mathrm{V}}{\mathrm{O}}_2$ by exhaled gas analyses from timed Douglas bag collections and by TD. Cardiac index was classified as low when ≤2.2 L/min/m² or normal when >2.2 L/min/m². The median (25th, 75th percentile) age of the cohort was 59 (50,67) years, and 50% were female. A total of 43.5% had normal left ventricular function by ventriculography, and 25.7% had ischemic heart disease. Median overall Fick and TD ${\stackrel{.}{\mathrm{Q}}}_{\mathrm{C}}$ were 4.4 (3.5, 5.5) and 4.3 (3.7, 5.2) L/min, respectively (p = 0.04). The median absolute percent error between Fick and TD ${\stackrel{.}{\mathrm{Q}}}_{\mathrm{C}}$ was 17.5 (7.7, 28.4)%, with a typical error of 0.88 L/min (95% confidence interval [CI] 0.82 to 0.95). Median absolute percent error was comparable in the low (n = 118) and normal ${\stackrel{.}{\mathrm{Q}}}_{\mathrm{C}I}$ (n = 135) groups (16.9% vs 18.9%, respectively, p = 0.88). typical error was 0.3 (95% CI 0.27 to 0.33) and 0.49 (95% CI 0.45 to 0.55) L/min/m² in that comparison. Percent error >25% between Fick and TD ${\stackrel{.}{\mathrm{Q}}}_{\mathrm{C}}$ was observed in over 30% of patients. Overall, Fick and TD ${\stackrel{.}{\mathrm{Q}}}_{\mathrm{C}}$ modestly correlated (R_s = 0.64, p <0.001), with a nondirectional error introduced by TD ${\stackrel{.}{\mathrm{Q}}}_{\mathrm{C}}$ [mean bias of 0.21 (−2.2, 2.7) L/min].

There was poor agreement between TD and the gold-standard Fick method, highlighting the limitations of making clinical decisions based on TD.

Introduction

Accurate estimation of cardiac output $\left({\dot{\mathrm{Q}}}_{\text{C}}\right)$ critical in cardiovascular care. This includes assessment of hemodynamics during inpatient management of acute decompensated heart failure, response to and titration of therapeutics in the intensive care setting, prioritizing listing status for patients with end-stage heart failure with an indication for heart transplantation, estimation of pulmonary and systemic vascular resistance, and estimation of stenotic valve area to inform decision making for mechanical intervention. The Fick method ${\dot{\mathrm{Q}}}_{\text{C}}$.= directly measured oxygen uptake $\left[\dot{\mathrm{V}}{\mathrm{O}}_2\right]$/systemic arterio-venous oxygen difference)¹ is the gold standard for estimating ${\dot{\mathrm{Q}}}_{\text{C}}$ and has been used to validate other techniques to estimate other methods of ${\dot{\mathrm{Q}}}_{\text{C}}$ estimation such as indicator dilution and inert gas rebreathing.^2,3 The thermodilution (TD) method that is commonly used in contemporary clinical scenarios requires the use of a pulmonary artery (PA) catheter, where a sensor on the catheter in the PA measures the change in blood temperature caused by an injection of a known volume of fluid at a known temperature into a proximal port (i.e., right atrium), with unidirectional mixing in the right ventricle, and temperature measured in the PA.⁴ The present study was performed to determine the agreement of ${\dot{\mathrm{Q}}}_{\text{C}}$ estimation between TD and Fick methods in consecutive adult patients who underwent a clinically indicated PA catheterization for a broad range of conditions.

Methods

This is a subanalysis of a study that tested the accuracy of the indirect Fick (${\dot{\mathrm{Q}}}_{\text{C}}$ derived from the calculation of $\dot{\mathrm{V}}{\mathrm{O}}_2$ through estimating formulas) versus the Fick method (also termed the Direct Fick method when oxygen uptake is measured).⁵ From a prospectively collected cohort of consecutive patients at Parkland Memorial Hospital who underwent a clinically indicated PA catheterization between 1999 and 2005, we included those who had a direct measurement of $\dot{\mathrm{V}}{\mathrm{O}}_2$ at rest and had paired ${\dot{\mathrm{Q}}}_{\text{C}}$ estimates by the Fick and TD methods. Common indications for PA catheterization included hemodynamic assessments of heart failure, pulmonary hypertension, and valvular disease. This study was approved by the University of Texas Southwestern Medical Center Institutional Review Board, with a waiver of participant informed consent. Demographics, baseline clinical characteristics, comprehensive hemodynamics, and data from ventriculography (qualitative grading) were included.

$\dot{\mathrm{V}}{\mathrm{O}}_2$ at rest was measured in all patients in the catheterization laboratory using the gold standard technique of Douglas,¹ with analysis of a 3-minute collection of exhaled air. Exhaled volume was measured with a Tissot spirometer, and concentrations of oxygen, carbon dioxide, and nitrogen were determined by mass spectrometry (Marquette MGA 1100, MA Tech Services, St. Louis, Missouri), calibrated before every measurement. All testing was completed while patients remained in the supine position. Hemodynamics were collected at a steady state with the patient in the supine position using a 7 French balloontipped PA (Edwards Lifesciences) with a tTD port and PA temperature sensor. Baseline vital signs were collected and reported before the administration of conscious sedation when appropriate. Hemodynamics, including right atrial (RA) pressure, right ventricular pressure, PA pressure, and pulmonary capillary wedge pressure, were recorded at end-expiration. Left ventricular end-diastolic pressure (LVEDP) was recorded during left heart catheterization. Systemic and pulmonary vascular resistances were calculated based on ${\dot{\mathrm{Q}}}_{\text{C}}$ measurements and mean systemic and arterial pressures. Fick ${\dot{\mathrm{Q}}}_{\text{C}}$ was then calculated by the following equation: [measured $\dot{\mathrm{V}}{\mathrm{O}}_2$]/ [(13.6 × hemoglobin (g/L)*(arterial oxygen saturation − mixed venous saturation)]. Cardiac index $\left({\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}\right)$ calculations were derived using body surface area according to the formula of Dubois.⁶ In addition, 3 to 5 TD measurements were performed, with the average of the measurements reported. Iced saline 10-ml injections were administered at 1-minute intervals, with typical inter-measurement variability within the laboratory measuring ±4%.^{7,8 ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$} was stratified into normal (>2.2 L/min/m²) and low (≤ 2.2 L/min/m²) subgroups to determine if the TD or Fick method appeared superior when the ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ was low.⁹

Descriptive statistics are reported as median (25% to 75% interquartile range [IQR]) when non-normally distributed and mean (SD) when normally distributed. Comparisons between groups of continuous variables were made using unpaired t test or Mann-Whitney U test as appropriate. Categorical data were compared between groups by chi-square and Fischer’s exact test when appropriate. One-way analysis of variance testing was used when comparing the means of 3 or more independent groups. The degree of disagreement between Fick and TD ${\dot{\mathrm{Q}}}_{\text{C}}$ was calculated as a percent error, dividing the absolute difference by the corresponding Fick ${\dot{\mathrm{Q}}}_{\text{C}}$ and multiplying by 100.¹⁰ Percent error was also reported for Fick and TD ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ for both low and normal ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ subgroups. For assessment of reliability (precision), typical error (TE) and coefficients of variation (TE expressed as a percentage) were calculated.¹¹ Spearman rank correlation coefficients with scatter plots were calculated and generated to assess the correlation between Fick and TD ${\dot{\mathrm{Q}}}_{\text{C}}$ and ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ measurements. The Bland-Altman method was used to assess the agreement of Fick and TD ${\dot{\mathrm{Q}}}_{\text{C}}$ and ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$, where the difference between Fick and TD calculations for each patient is plotted against the corresponding average of Fick plus TD calculation.¹² Diagnostic performance analysis using sensitivity and specificity was performed to assess the accuracy of the TD method in patients with low ${\dot{\mathrm{Q}}}_{\text{C}}$ (defined as a ${\dot{\mathrm{Q}}}_{\text{CI}}\text{£}2.2\text{L}/\text{min}/{\text{m}}^2$).⁹ Linear regression models were used to assess the association between covariates of interest, including mean PA pressure, LVEDP, body mass index (BMI), and systolic blood pressure with a degree of TD ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ error (raw numeric difference between Fick and TD ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$). The Breusch-Pagan test was used to test for heteroscedasticity in the linear regression models. Two-sided p values <0.05 were considered statistically significant. Statistical analysis was performed using Prism (Waltham, Massachusetts) version 9 (GraphPad Software, San Diego, California) and R version 3.6.1 (The R Foundation for Statistical Computing, Vienna, Austria).

Results

Cohort characteristics are listed in Table 1. The median age was 59 (50, 67) years, 50% male. A total of 43.8% had normal left ventricular function by ventriculography, and 25.7% had ischemic heart disease. Median ${\dot{\mathrm{Q}}}_{\text{C}}$ was 4.4 (3.5, 5.5) L/min by Fick and 4.3 (3.7, 5.2) L/min by TD; median ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ was 2.3 (1.9, 3) L/min/m² by Fick, 2.3 (1.7, 2.7) L/min/m² by TD (p = 0.04, for both ${\dot{\mathrm{Q}}}_{\text{C}}$ and ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$). Overall percent error between Fick and TD ${\dot{\mathrm{Q}}}_{\text{C}}$ was 17.5 (7.7, 28.4)%, with a TE of 0.88 L/min (95% confidence interval [CI] 0.82 to 0.95) and coefficient of variation (CV) of 37.8% (Table 2). When comparing low (≤2.2 L/min/m²) and normal (>2.2 L/min/m²) Fick ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ groups, the median percent error did not significantly differ (16.9% vs 18.9%, respectively, p = 0.88). TE was 0.3 (95% CI 0.27 to 0.33) and 0.49 (95% CI 0.45 to 0.55) in low and normal ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ groups, respectively. CV for low and normal Fick ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ groups was 22.5% and 17%, respectively. Patients in the low ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ group had a lower mean arterial pressure than the normal ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ group (90 vs 99 mm Hg, respectively; p = 0.04). In addition, mean PA pressure, PA diastolic pressure, pulmonary capillary wedge pressure, and systemic and pulmonary vascular resistance were all significantly higher in the low ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ group compared with the normal ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ group (Table 1, all p <0.05).

Table 1.

Baseline cohort characteristics

	All (n = 253)	${\dot{\mathrm{Q}}}_{\text{CI}}\le 2.2\text{L}/\text{min}/{\text{m}}^2$ by Fick (n= 118)	${\dot{\mathrm{Q}}}_{\text{CI}}>2.2\text{L}/\text{min}/{\text{m}}^2$ by Fick (n = 135)	p Value
Age, median (IQR) (years)	59 (50, 67)	60 (51.8, 66.5)	58 (49, 67)	0.47
Male, no. (%)	127 (50)	64 (54.2)	63 (46.7)	0.25
BMI, median (IQR) (kg/m2)	27.8 (23.8, 31.9)	27.9 (23.2, 32.4)	27.6 (23.9, 30.9)	0.96
Ischemic Heart Disease	65 (25.7%)	32 (27.1%)	33 (24.4%)	0.66
Race/ethnicity,				0.23
Black	87 (34.4%)	45 (38.1%)	42 (31.1%)
Hispanic	46 (18.2%)	25 (21.2%)	21 (15.6%)
White	102 (40.3%)	40 (33.9%)	62 (45.9%)
Other	18 (7.1%)	8 (6.8%)	10 (7.4%)
Direct Fick cardiac output, median (IQR) (L/min)	4.4 (3.5, 5.5)	—	—	—
Thermodilution cardiac output, median (IQR) (L/min)	4.3 (3.7, 5.2)	—	—	—
Direct Fick cardiac index, median (IQR) (L/min/m2)	2.3 (1.9, 3)	1.9 (1.7, 2.1)	2.9 (2.5, 3.4)	—
Thermodilution cardiac index, median (IQR) (L/min/m2)	2.3 (1.7, 2.7)	2 (1.8, 2.3)	2.5 (2.3, 2.9)	<0.0001
Cardiac index percent error, median (IQR) (%)	17.5 (7.7, 28.4)	16.9 (7.9, 28)	18.9 (6.9, 28.9)	0.88
Primary indication for heart catheterization				0.71
Ischemic evaluation	26 (10.3%)	11 (9.3%)	15 (11.1%)
Hemodynamic evaluation	73 (28.9%)	32 (27.1%)	41 (30.4%)
Valvular assessment	154 (60.8%)	75 (63.6%)	79 (58.5%)
LVEF by ventriculogram				0.09
Normal	110 (43.8%)	47 (40.2%)	63 (47%)
Mildly reduced	53 (21.2%)	24 (20.5%)	29 (21.8%)
Moderately reduced	46 (18.3%)	19 (16.2%)	27 (20%)
Severely reduced	42 (16.7%)	27 (23.1%)	15 (11.2%)
Mitral regurgitation by ventriculogram				0.29
None	148 (58.9%)	69 (58.5%)	79 (59.4%)
Mild	36 (14.3%)	17 (14.4%)	19 (14.2%)
Moderate	46 (18.3%)	21 (17.8%)	25 (18.8%)
Severe	21 (8.4%)	11 (9.2%)	10 (7.6%)
Invasive assessment of mitral stenosis				0.07
None	224 (88.5%)	101 (85.6%)	123 (91%)
Mild	9 (3.6%)	6 (5.1%)	3 (2.2%)
Moderate	8 (3.2%)	2 (1.7%)	6 (4.4%)
Severe	12 (4.7%)	9 (7.6%)	3 (2.2%)
Invasive assessment of aortic stenosis				0.77
None	147 (58.2%)	69 (58.5%)	78 (57.8%)
Mild	22 (8.7%)	8 (6.8%)	14 (10.4%)
Moderate	39 (15.4%)	19 (16.1%)	20 (14.8%)
Severe	45 (17.7%)	22 (18.6%)	23 (17%)
Aortic regurgitation by ventriculogram				0.43
None	216 (85.4%)	97 (82.2%)	119 (88.8%)
Mild	10 (4%)	5 (4.2%)	5 (3.8%)
Moderate	17 (6.7%)	10 (8.5%)	7 (5.2%)
Severe	9 (3.5%)	6 (5.1%)	3 (2.2%)
Systolic blood pressure, median (IQR) (mm Hg)	130 (115, 146)	127 (110.8, 146.3)	132 (116, 144)	0.20
Diastolic blood pressure, median (IQR) (mm Hg)	72 (64, 80)	70 (63, 82)	73 (64, 80)	0.42
Mean arterial pressure, median (IQR) (mm Hg)	95 (85, 108)	90 (82, 105)	99 (86, 110)	0.04
Right atrial pressure, median (IQR) (mm Hg)	7 (5, 10)	8 (5, 11)	6 (4, 9)	0.10
Systolic pulmonary artery pressure, median (IQR) (mm Hg)	44 (34, 56)	49 (37, 60)	42 (33, 51)	0.003
Diastolic pulmonary artery pressure, median (IQR) (mm Hg)	20 (16, 25)	22 (16, 28)	18 (14, 23)	0.0002
Mean pulmonary artery pressure, median (IQR) (mm Hg)	30 (22, 38)	32 (24, 40)	27 (21, 35.5)	0.002
Pulmonary capillary wedge pressure, median (IQR) (mm Hg)	18 (12, 23)	19 (12, 25.6)	16.4 (11, 22)	0.01
Left ventricular end-diastolic pressure, median (IQR) (mm Hg)	17 (11, 23)	19 (12, 26)	16 (11, 21)	0.01
Systemic vascular resistance, median (IQR) (dynes*sec/cm5)	1,555 (1247, 2012)	1,934 (1,159, 2,393)	1,328 (1,069, 1569)	<0.0001
Pulmonary vascular resistance, median (IQR) (Wood Units)	2.5 (1.6, 3.7)	3.3 (2.4, 4.94)	1.8 (1.2, 2.7)	<0.0001
PAPi, median (IQR)	3.8 (2.3, 5.3)	3.6 (2.3, 5.5)	3.9 (2.3, 5.3)	0.76

BMI = body mass index; cm = centimeter; IQR = interquartile range; kg = kilograms; L = liters; LVEF = left ventricular ejection fraction; m = meters; min = minutes; mm Hg = milliliters of mercury; no = number; PAPi = pulmonary artery pulsatility index; ${\dot{\mathrm{Q}}}_{\text{CI}}$ = cardiac index; SD = standard deviation; sec = seconds; TD = thermodilution; y = years.

Table 2.

Assessment of accuracy between Fick and TD methods

	n	Mean (SD) ${\dot{\mathrm{Q}}}_{\text{C}}/{\dot{\mathrm{Q}}}_{\text{CI}}$	Bland-Altman Mean Bias (+/− 2 SD)	Median Percent Error (IQR)	Rs*	TE*	CV
Overall Fick vs TD ${\dot{\mathrm{Q}}}_{\text{C}}$	253	fick 4.7 (1.7) thermodilution 4.5 (1.3)	0.21 (—2.2–2.7)	17.5 (7.7, 28.4)	0.64 (0.56, 0.71)	0.88 (0.82–0.95)	37.8%
Fick vs TD ${\dot{\mathrm{Q}}}_{\text{CI}}$ (>2.2 L/min/m2)	135	fick 3 (0.7) thermodilution 2.6 (0.6)	−0.42 (−1.76–0.9)	18.9 (6.9, 28.9)	0.34 (0.17, 0.49)	0.49 (0.45, 0.55)	22.5%
Fick vs TD ${\dot{\mathrm{Q}}}_{\text{CI}}$ (<2.2 L/min/m2)	118	fick 1.8 (0.3) thermodilution 2 (0.4)	0.23 (—0.59–1.06)	16.9 (7.9, 28)	0.31 (0.13, 0.47)	0.3 (0.27, 0.33)	17%

^*95% confidence intervals expressed for both R_s and TE.

CV = coefficient of variation; IQR = interquartile range; L = liters; m = meters; min = minute; ${\dot{\mathrm{Q}}}_{\text{C}}$ = cardiac output; ${\dot{\mathrm{Q}}}_{\text{CI}}$ = cardiac index; R_s = Spearman’s rank correlation coefficient; SD = standard deviation; TD = thermodilution; TE = typical error.

There was no systematic overestimation or underestimation of Fick ${\dot{\mathrm{Q}}}_{\text{C}}$ by the TD method for lower or higher values of ${\dot{\mathrm{Q}}}_{\text{C}}$ in the overall cohort; the mean bias was 0.21 L/min with limits of agreement (mean bias +/− 2 SDs) ranging from −2.2 to 2.7 L/min (Figure 1). Similarly, for the normal ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ subgroup, there was no systematic directionality of error observed, with a mean bias of −0.42 L/ min/m² and limits of agreement ranging from −1.76 to 0.9 L/min/m² (Figure 1). For the low ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ subgroup, no directionality of error was observed, with a mean bias of 0.23 L/min/m² and limits of agreement ranging from −0.59 to 1.06 L/min/m² (Figure 1).

Figure 1.

Agreement between Fick and thermodilution measurements using the Bland-Altman method stratified by the (A) Overall cohort ${\dot{\mathrm{Q}}}_{\text{C}}$ (B) Normal ${\dot{\mathrm{Q}}}_{\text{CI}}$ and (C) Low ${\dot{\mathrm{Q}}}_{\text{CI}}$ groups. L = liters; m, = meters; minutes, = minutes.

Scatter plots with an assessment of correlation between Fick and TD are shown in Figure 2. In the overall cohort $\left({\dot{\mathrm{Q}}}_{\text{C}}\right)$, there was a moderate positive correlation between the 2 methods (R_s = 0.61; 95% CI 0.56 to 0.71, p <0.001, Figure 2) In the normal ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ subgroup, a weakly positive correlation was observed between Fick and TD methods (R_s = 0.34; 95% CI 0.17 to 0.49, p <0.001, Figure 2). Similar results were observed with the low ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ subgroup, with a significant but weak positive correlation between Fick and TD methods (R_s = 0.31; 95% CI 0.13 to 0.47, p <0.001, Figure 2)

Figure 2.

Scatter plot of Direct Fick versus thermodilution method for the (A) Overall cohort ${\dot{\mathrm{Q}}}_{\text{C}}$ (B) Normal ${\dot{\mathrm{Q}}}_{\text{CI}}$_, and (C) Low ${\dot{\mathrm{Q}}}_{\text{CI}}$ groups. L = liters; m = meters; min = minute.

The magnitude of error between Fick and TD methods expressed as a percent error for the overall cohort by ${\dot{\mathrm{Q}}}_{\text{C}}$ and normal and low ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ subgroups is shown in Figure 3. An error >25% between Fick and TD methods was observed in approximately one-third of patients in all 3 groups. No significant differences in relevant baseline characteristics or hemodynamics were observed in groups stratified by range of percent error (Supplementary Table). A theorized analysis of the diagnostic performance of the TD method for patients with low ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ (defined as a ${\dot{\mathrm{Q}}}_{\text{CI}}\le 2.2\text{L}/\text{min}/{\text{m}}^2$) is shown in Figure 4. TD ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ was poorly sensitive (0.65) with a negative predictive value of 0.57 for diagnosing low ${\dot{\mathrm{Q}}}_{\text{C}}$ by the criteria of ${\dot{\mathrm{Q}}}_{\text{CI}}\le 2.2\text{L}/\text{min}/{\text{m}}^2$ by the Fick method. A false-positive rate of 14.6% was observed when TD ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ was calculated at ≤2.2 L/min/m² in those with a measured Fick ${\dot{\mathrm{Q}}}_{\text{CI}}>2.2\text{L}/\text{min}/{\text{m}}^2$. Linear regression models were performed to assess the association of TD ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ error and covariates of interest, including mean PA pressure, LVEDP, BMI, and systolic blood pressure (Figure 5). Higher values of mean PA pressure correlated with an overestimation of Fick ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ by the TD method (Figure 5). For every 10 unit increase in the mean PA pressure, there was a 0.13 (95% CI 0.06 to 0.2) L/min/m² change in the raw TD ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ error (p = 0.001). Higher values of LVEDP also correlated with an overestimation of Fick ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ by the TD method (Figure 5). For every 10 unit increase in the LVEDP, there was 0.14 (95% CI 0.03 to 0.24) L/min/m² change in the raw TD ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ error (p = 0.011). The Breusch-Pagan test demonstrated significant heteroscedasticity of the error (p = 0.007) in the linear regression model of TD ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ error and LVEDP. There was no significant change in the TD ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$ error across values of BMI and systolic blood pressure (Figure 5, p >0.05 for both). No significant heteroscedasticity of the error was observed in regression models for mean PA pressure, BMI, or systolic blood pressure (p >0.05 for all).

Figure 3.

Distribution of error between Direct Fick and thermodilution ${\dot{\mathrm{Q}}}_{\text{C}}$ in the overall cohort, normal, and low ${\dot{\mathrm{Q}}}_{\text{CI}}$ groups. L = liters; m = meters; min = minute.

Figure 4.

Performance of thermodilution ${\dot{\mathrm{Q}}}_{\text{CI}}$ for those with low cardiac output (defined as Fick ${\dot{\mathrm{Q}}}_{\text{CI}}\text{£}2.2\text{L}/\text{min}/{\text{m}}^2$) in patients with Fick ${\dot{\mathrm{Q}}}_{\text{CI}}>3\text{L}/\text{min}/{\text{m}}^2$. L = liters; m = meters; min = minutes; NPV = negative predictive value; PPV = positive predictive value.

Figure 5.

Linear regression models of TD ${\dot{\mathrm{Q}}}_{\text{CI}}$ error by selected covariates. (A) Higher values of mean PA pressure correlated with an overestimation of Fick ${\dot{\mathrm{Q}}}_{\text{CI}}$ by the TD method; for every 10 unit increase in the mean PA pressure, there was a 0.13 (95% CI 0.06 to 0.2) L/min/m² change in the raw TD ${\dot{\mathrm{Q}}}_{\text{CI}}$ error (p = 0.001). (B) Higher values of LVEDP correlated with an overestimation of Fick ${\dot{\mathrm{Q}}}_{\text{CI}}$ by the TD method; for every 10 unit increase in the LVEDP, there was 0.14 (95% CI 0.03 to 0.24) L/min/m² change in the raw TD ${\dot{\mathrm{Q}}}_{\text{CI}}$ error (p = 0.011); The Breusch-Pagan test demonstrated significant heteroskedasticity of the TD ${\dot{\mathrm{Q}}}_{\text{CI}}$ error (p = 0.007) for LVEDP but no other covariates (p >0.05). There was no significant change in the TD Q_CI error across values of BMI (C) and systolic blood pressure (p >0.05 for both), (D). kilograms = kilogram; L = liters; m = meters; mm Hg = milliliters of mercury; min = minutes.

Discussion

The agreement between Fick and TD estimated ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{C}}$ was analyzed in a cohort of patients who underwent a clinically indicated PA catheterization. The main findings of the present study were: (1) the median overall percent error between the Fick and TD methods was 17.5%, with no significant difference observed when stratified by low or normal ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{CI}}$; (2) a percent error of >25% between Fick and TD methods was observed in approximately one-third of participants; (3) no systematic directionality of discordance between the Fick and TD methods was observed. Overall, these data demonstrate that agreement between TD and the gold-standard Fick method was poor, with a significant potential impact on clinical decision-making depending on which measurement is used.

Most cardiac catheterization laboratories generally calculate ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{C}}$ using TD and/or Fick estimates of ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{C}}$ using algorithm-based estimations of $\dot{\mathrm{V}}{\mathrm{O}}_2$¹³ (indirect Fick). ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{C}}$ assessed by measurement of $\dot{\mathrm{V}}{\mathrm{O}}_2$ through either mass spectrometry analysis of timed Douglas bag collections of exhaled air or by breath-by-breath analysis of exhaled air using metabolic cart analysis is more time-consuming given the need for calibration and requires inherent cost to acquire and maintain the necessary equipment to make such exhaled gas measurements. Formulas used to estimate $\dot{\mathrm{V}}{\mathrm{O}}_2$ to ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{C}}$ using the Fick equation are commonly used but are inaccurate when compared with Fick measurements using directly measured $\dot{\mathrm{V}}{\mathrm{O}}_2$, especially in obese patients.^5,10,14 The TD method is often performed in concert with indirect Fick ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{C}}$ measurements, and TD recently was shown in a large observational cohort to predict mortality more accurately as compared with the indirect Fick method.¹⁵ However, there are few previous studies comparing the accuracy of the TD method versus the gold standard Direct Fick method, where agreement between the 2 methods demonstrated acceptable correlation (R > 0.8),^16–18 including in a subset of patients with pulmonary hypertension, severe tricuspid regurgitation and low ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{C}}$.¹⁸ This larger study cohort contrasts with these findings indicating the potential error of TD measurements when compared with the Fick method. A variation in the range of error observed may be due in part to the type of cohort studied.

There was a high degree of error between Fick and TD methods in the overall cohort and across a spectrum of ${\dot{\mathrm{Q}}}_{\text{CI}}{}_{\text{C}}$ ranges, with greater than one-third of each study group having errors in excess of 25%. In addition, clinical characteristics and hemodynamics were similar in groups stratified over a broad range of percent error categories. This is a representative sample of patients who presented for ischemic, heart failure, or valvular evaluations, without one singular disease pathology. The low ${\dot{\mathrm{Q}}}_{\text{CI}}{}_{\text{C}}$ group was more likely to have left ventricular systolic dysfunction by ventriculography, higher mean pulmonary and left ventricular end-diastolic pressures, and higher pulmonary vascular resistance. No consistent directionality in error between the 2 methods was observed, with broad scatter of underestimated and overestimated TD measurements seen in all 3 Bland-Altman analyses.

Higher values of LVEDP correlated with an overestimation of ${\dot{\mathrm{Q}}}_{\text{CI}}{}_{\text{C}}$ by the TD method, possibly misclassifying patients who may have low-output heart failure. In addition to these observations, considerable variance or “noise” in the scatter plots was noticeably present across the range of covariates values plotted against TD ${\dot{\mathrm{Q}}}_{\text{CI}}{}_{\text{C}}$ error. Previous small-sample reports have suggested that significant tricuspid regurgitation may underestimate TD ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{C}}$_,^19,20 possibly because of less volume of the saline injectate reaching the distal thermistor. Other studies contradict this finding,¹⁸ although collectively, large-scale analyses are lacking to make robust assertions of directionality in error between the 2 methods in the setting of severe right-sided valvular regurgitation.

Estimates of ${\dot{\mathrm{Q}}}_{\text{C}}{}_{\text{C}}$/index are critical in several areas within cardiovascular care, including calculation of aortic valve area for patients with suspected severe aortic stenosis, assessment of pulmonary vascular resistance for those being considered for vasodilator therapies, and risk assessment in patients with advanced heart failure. For example, in patients listed for heart transplantation with a systolic blood pressure <90 mm Hg and a pulmonary capillary wedge pressure >15 mm Hg, strict cutoffs of ${\dot{\mathrm{Q}}}_{\text{CI}}{}_{\text{C}}$ (<1.8 L/min/m² in those off inotropes or <2 L/min/m² when on inotropes) are required for high-priority listing status.²¹ A difference in ${\dot{\mathrm{Q}}}_{\text{CI}}{}_{\text{C}}$ by >25% depending on which method is used can alter prioritization to receive a donor organ, a precious and scarce resource. This is problematic in the current allocation system, where a higher than expected number of patients are listed at higher priority statuses (1 or 2).²² Our theoretical diagnostic accuracy model demonstrated poor discrimination of TD ${\dot{\mathrm{Q}}}_{\text{CI}}{}_{\text{C}}$ in identifying those with a Fick ${\dot{\mathrm{Q}}}_{\text{CI}}\le 2.2\text{L}/\text{min}/{\text{m}}^2$ (low ${\dot{\mathrm{Q}}}_{\text{C}}$) because of a high false-negative rate resulting in a sensitivity of 0.65 and negative predictive value of 0.57. We recognize the present cohort should not be classified as having advanced heart failure or cardiogenic shock given a median systolic blood pressure >120 mm Hg, and none on inotropic therapy or mechanical circulatory assist devices. On the other hand, there is no reason to assume that the accuracy of TD ${\dot{\mathrm{Q}}}_{\text{C}}$ would increase in those with suspected cardiogenic shock. Previous studies of ${\dot{\mathrm{Q}}}_{\text{C}}$ measurement, such as inert gas rebreathing, have suggested greater accuracy than both TD and indirect Fick methods, and may serve as reliable, complements to the Fick method in a variety of clinical settings.²³

Slight differences in methods between operators regarding exhaled air collection and analysis could have influenced the accuracy of the $\dot{\mathrm{V}}{\mathrm{O}}_2$ at rest measurement using the method of Douglas. A specific protocol was followed by all operators, and the equipment was routinely calibrated to mitigate this risk. Relevant clinical information, including certain elements of the medical history, including detailed medications and clinical outcomes, was not available during data abstraction. We lacked echocardiographic data at the same time as PA catheterization to determine an association between TD method accuracy and specific disease phenotypes. The PA catheterizations were performed over 15 years ago. However, the TD method has changed little over that time, and the Fick method using measured $\dot{\mathrm{V}}{\mathrm{O}}_2$ by exhaled gas analyses from timed Douglas bag collections is arguably the gold-standard method.