Abstract
BACKGROUND: We hypothesize that oxygen consumption (V̇o2) estimation in patients with respiratory symptoms is inaccurate and can be improved by considering arterial blood gases or spirometric variables.
METHODS: For this retrospective study, we included consecutive subjects who underwent cardiopulmonary exercise testing. Resting V̇o2 was determined using breath-by-breath testing methodology. Using a training cohort (n = 336), we developed 3 models to predict V̇o2. In a validation group (n = 114), we compared our models with 7 available formulae.
RESULTS: Our first model (V̇o2 = −184.99 + 189.64 × body surface area [BSA, m2] + 1.49 × heart rate [beats/min] + 51.51 × FIO2 [21% = 0; 30% = 1] + 30.62 × gender [male = 1; female = 0]) showed an R2 of 0.5. Our second model (V̇o2 = −208.06 + 188.67 × BSA + 1.38 × heart rate + 35.6 × gender + 2.06 × breathing frequency [breaths/min]) showed an R2 of 0.49. The best R2 (0.68) was obtained with our last model, which included minute ventilation (V̇o2 = −142.92 + 0.52 × heart rate + 126.84 × BSA + 14.68 × minute ventilation [L]). In the validation cohort, these 3 models performed better than other available equations, but had wide limits of agreement, particularly in older individuals with shorter stature, higher heart rate, and lower maximum voluntary ventilation.
CONCLUSIONS: We developed more accurate formulae to predict resting V̇o2 in subjects with respiratory symptoms; however, equations had wide limits of agreement, particularly in certain groups of subjects. Arterial blood gases and spirometric variables did not significantly improve the predictive equations.
Introduction
The available equations to estimate oxygen consumption (V̇o2) were obtained from selected populations (eg, children with congenital heart diseases) that commonly differ from the patients encountered in clinical practice (Table 1). More importantly, the estimates of resting V̇o2 by the available formulae are inaccurate when compared with the V̇o2 directly measured either by the breath-to-breath methodology or Douglas bag technique combined with mass spectrometry.1 Due to these limitations, investigators have questioned the value of estimating V̇o2 and supported its direct measurement.2
Formulae to Estimate V̇o2
Equations to estimate V̇o2 have not been systematically studied in patients with lung diseases or respiratory symptoms. It is possible that arterial blood gases and spirometric parameters may be of value to better predict the resting V̇o2 in patients with respiratory symptoms or diseases. We hypothesize that the estimation of V̇o2 in patients with respiratory symptoms or diseases is inaccurate and that the predictive equations can be improved by adding parameters obtained from arterial blood gases and/or pulmonary function tests.
QUICK LOOK
Current knowledge
Equations to estimate oxygen consumption (V̇o2) have not been systematically studied in patients with lung diseases or respiratory symptoms. When knowledge of V̇o2 is important, direct measurement is typically recommended.
What this paper contributes to our knowledge
New formulae to predict V̇o2 in subjects undergoing cardiopulmonary exercise testing for respiratory symptoms were developed using FIO2, minute volume, and breathing frequency. Arterial blood gases and spirometric variables did not significantly improve predictive equations.
Methods
Subject Selection and Study Design
The research protocol was approved by the Cleveland Clinic institutional review board (study 13-1068). Written informed consent was waived. We included consecutive unique subjects who underwent cardiopulmonary exercise testing at the Pulmonary Function Test Laboratory of the Respiratory Institute, Cleveland Clinic, between October 2010 and September 2013. Cardiopulmonary exercise testing was ordered at the discretion of the treating physicians and was performed according to the American Thoracic Society/American College of Chest Physician recommendations.3 Common reasons for ordering the test included unexplained dyspnea and preoperative evaluation to better estimate the risk of surgery in subjects with lung diseases. We have not included patients with cardiovascular diseases because these individuals are tested by the Heart and Vascular Institute and are not captured in our database. Similarly, we did not include patients with known pulmonary hypertension, given that patients with this condition rarely undergo cardiopulmonary exercise testing at our laboratory.
Immediately before each cardiopulmonary exercise test, we measured height and weight, which were then used to calculate body surface area (BSA) by the equation of Du Bois et al4:
Furthermore, we recorded information on smoking status, type and time of last meal, medications, reason for the test, and need for supplemental oxygen therapy. We measured blood pressure and heart rate. Immediately before the V̇o2 determination and per protocol, we performed spirometry following American Thoracic Society/European Respiratory Society recommendations5 and obtained arterial blood gases from the radial artery. We later collected data on the type and severity of lung disease, serum thyroid-stimulating hormone concentration, and echocardiographic measurements. We graded severity of air flow limitation following the recommendations of the Global Initiative for Chronic Obstructive Lung disease.6 Restrictive lung disease was considered present when spirometry showed FEV1 over forced vital capacity (FEV1/FVC) ≥ 0.7 and the total lung capacity was less than 80% of predicted.7 In case of unspecific spirometric findings or overlap (obstructive and restrictive disease), we reported the predominant ventilatory defect, based on a comprehensive evaluation by the authors (ART and AA).
V̇o2 Determination
Subjects were told to continue taking all medications and only eat a light meal on the day of the test (> 2 h before the test). All examinations were performed by experienced respiratory therapists. No sedatives were given. Room temperature was maintained at 70°F. After an acclimatization phase, we measured V̇o2 over a period of 5 min at rest in sitting position. V̇o2 was determined using the Ultima CardiO2 gas exchange analysis system (MGC Diagnostics, Saint Paul, Minnesota), which is commercially available and uses breath-by-breath testing methodology. All subjects breathed spontaneously using an adequately sized 7450 series V2 face mask (Hans Rudolph, Inc, Shawnee, Kansas) held in place by head gear. We used a polycarbonate plastic adapter (Hans Rudolph, Inc) to connect the preVent flow sensor (MGC Diagnostics) to the mask. We performed flow and gas calibration before each test. All pertinent information was uploaded to an Excel database.
In the event the subject was on continuous O2 therapy, we provided supplementary O2 with a FIO2 of 30%, using a 30% O2 tank and a 30 L nondiffusing gas bag made of multilayer polyethylene thermoplastic laminate with a layer of aluminum foil at the center (Hans Rudolph, Inc). Corrugated tubing connected the 3-way stopcock attached to the bag and the mouthpiece.
Formulae for Estimating Resting V̇o2
For the estimation of resting V̇o2, we used the formulae by Bergstra et al8 (V̇o2 [mL/min]) = 157.3 × BSA [m2] + gender [men = 10 or women = 0] − [10.5 × loge age] + 4.8), Dehmer et al9 (V̇o2 [mL/min] = 125 [mL/min/m2] × BSA [m2 ]), Krovetz et al10 (V̇o2 [mL/min] = −35.6 + 1.39 × height [cm] + 0.84 × weight [kg]), LaFarge et al11 (V̇o2 [mL/min] = 138.1 − [gender [men = 11.49 or women = 17.04] × loge age] + [0.378 × heart rate] × BSA [m2]), Lindahl12 (V̇o2 [mL/min] = 4 × weight [kg] + 35.8), Lundell et al13 (V̇o2 [mL/min] = 157.9 × BSA [m2] + 0.79 × heart rate − 61.8) for males or (V̇o2 [mL/min] = 159 × BSA [m2] + 0.77 × heart rate − 61.6) for females, and Wessel et al14 (V̇o2 [mL/min] = [144.8 × BSA [m2]] + 5.6).
Statistical Analysis
Continuous data are presented as mean ± SD or median (interquartile range) where appropriate. Categorical data are summarized as discrete values and percentages (n [%]). The data were randomly separated into training and testing cohorts at a ratio of 3:1. Although we randomly divided these 2 groups, it was expected that some variables would be statistically different between them, a fact that tests the V̇o2 predictive equations more rigorously. The training data were used to build the linear regression model. Univariate linear regression analyses were performed on all predetermined independent variables and the outcome of interest, that is, resting V̇o2 (mL/min). We performed correlation analysis of the predictors to avoid multicollinearity in multivariate regression models. The algorithm by Kuhn15 was applied to find the minimal set of predictors, and we excluded variables that were highly correlated (correlation value > 0.8). The remaining predictors whose P values were < .10 in the univariate linear regression analyses were used for building the multivariate linear regression models. Stepwise variable selection procedure using Akaike information criterion16 was applied to identify the final multivariate models. Nonlinearity and interactions were explored in the model building process.
We built 3 multivariate models from the training dataset, as different sets of variables may be available to healthcare providers. The 3 models as well as 7 previous models in the literature were assessed in the testing (validation) dataset. The concordance correlation coefficients (CCC) between estimated and measured V̇o2 determinations17 were calculated to evaluate the performance among all models. CCC is a standardized coefficient that contains a measurement of precision (Pearson correlation coefficient) and accuracy (bias correction factor). It ranges between −1 and 1 (1 represents perfect agreement). We calculated the coefficient of variation (100 × SD/mean) as well as the median and interquartile range of the absolute and percentage (absolute difference × 100/measured V̇o2) difference between the measured and estimated resting V̇o2. Bland-Altman methodology18 was used to plot the percentage difference between measured and estimated V̇o2 against the measured V̇o2 (accepted standard).19 The mean difference and 95% limit of agreement are reported when appropriate.18 All P values are 2-tailed, and a value of < .05 was considered significant. SPSS 20 (SPSS, Chicago, Illinois), MedCalc (Ostend, Belgium), and R studio software were employed for the analyses (R Project, Vienna, Austria).
Results
Overall Characteristics of the Study Population
We included 450 subjects, of whom 336 subjects formed part of the training cohort (used to build the lineal regression models) and 114 of the validation group (used to test the proposed and available formulae to estimate V̇o2). Characteristics of the training and validation cohort are presented in Table 2.
Characteristics of the Testing and Validation Cohorts
Building a Model to Predict V̇o2
We determined the measured V̇o2 and V̇o2/kg for a large number of variables of interest (Table 3). On occasion, only V̇o2 or V̇o2/kg were significant between groups; however, we only focused in predicting the measured absolute V̇o2 rather than V̇o2/kg, because performing regression analysis on indexed values that are then transformed to absolute values can lead to errors due to regression to the mean.8
V̇o2 and V̇o2/kg Measured in Different Subgroups of Interest
We identified several predictors that had a P value < .10 on univariate linear regression analysis (gender, height, weight, BSA, smoking status, heart rate, breathing frequency, FIO2, hemoglobin concentration, O2 content, lactic acid concentration, FVC, FEV1, FEV1/FVC, maximum voluntary ventilation [MVV], severity of obstructive or restrictive lung disease, and minute ventilation).
After correlation analyses, we removed arterial oxygen content, FEV1 % of predicted, FVC percentage of predicted, and measured MVV. We kept BSA instead of height and/or weight, because BSA was preferentially selected by our models.
Our initial model (model 1) to estimate resting V̇o2 used all the variables with the exception of minute ventilation, as this determination requires especial equipment. After stepwise selection, we obtained the following equation (V̇o2 = −184.99 + 189.64 × BSA [m2] + 1.49 × heart rate [beats/min] + 51.51 × FIO2 [21% = 0; 30% = 1] + 30.62 × gender [male = 1; female = 0]). The R2 for this model was 0.5 (adjusted R2: 0.49); therefore, this combination of predictive variables explained half of the variance of V̇o2. Multivariate adaptive regression splines made the model more complex by creating splines with 2 knots for heart rate and BSA without improving the R2.
Given that the criteria for using an FIO2 of 30% may vary among institutions, we developed a model (model 2) without this variable. In this case, the equation that best estimated resting V̇o2 (model 2) was: −208.06 + 188.67 × BSA (m2), 1.38 × heart rate (beats/min) + 35.6 × gender (male = 1; female = 0] + 2.06 × breathing frequency (breaths/min). The R2 for this equation was 0.49 (adjusted R2: 0.48).
An alternative model was constructed using minute ventilation, a value obtained from the breath-to-breath analysis system (model 3). In this case, resting V̇o2 = −142.92 + 0.52 × heart rate (beats/min) + 126.84 × BSA (m2) + 14.68 × minute ventilation (L). The R2 for this model was 0.68 (adjusted R2: 0.68).
Comparison of the New Models to Predict V̇o2 with Established Ones
In the validation cohort, we compared our 3 models with 7 other available formulae that estimate resting V̇o2 (Table 4). We used the CCC, coefficient of variation, absolute and percentage difference, as well as Bland-Altman analysis. The CCCs of our 3 models (particularly models 1 and 3) were higher than others. The coefficients of variation as wells as the percentage differences between measured and estimated V̇o2 were lower for our models as well as the formulae by Bergstra et al,8 Lundell et al,13 and Wessel et al14 Bland-Altman analyses for model 1 and the formulae by Bergstra et al,8 Lundell et al,13 and Wessel et al14 are presented in Figure 1. All plots showed a proportional difference, because they tend to overestimate lower V̇o2 values and underestimate higher V̇o2 determinations.
Comparison of Different Models That Predict V̇o2 (Validation Cohort)
Bland-Altman analyses between measured V̇o2 and the percentage difference between measured and estimated V̇o2. A: Proposed model 1 (mean difference: −5.7%, limit of agreement: −40.4 +29.1%). B: Bergstra et al8 (mean difference: 9.7%, limit of agreement:−22 +41.3%). C: Lundell et al13 (mean difference:−3.3%, limit of agreement: −41.3 +34.7%). D: Wessel et al14 (mean difference: 4.4%, limit of agreement: −31.2 +40%). Solid horizontal lines indicate the mean; dashed horizontal lines indicate ± 1.96 SD.
Characteristics of the Subjects Who Had > 25% Difference Between Estimated (Model 1) and Measured V̇o2
Subjects who had ≥ 25% difference between the measured and estimated resting V̇o2 (model 1) were older (63 ± 15 vs 50 ± 18 y, P = .004) and had shorter stature (163 ± 10 vs 169 ± 10 cm, P = .02), lower measured MVV (82 ± 40 vs 110 ± 44 L, P = .02), and higher heart rate (91 ± 17 vs 81 ± 14 beats/min, P = .07) than those with a small difference. No arterial blood gas or spirometric differences were noted between these groups. The addition of age or loge(age), height, or MVV to our models did not improve the R2.
Discussion
In the present study, we noted a poor accuracy between the measured and estimated V̇o2 in subjects with respiratory symptoms. Hence, we assessed the impact of a variety of factors on the measured V̇o2 and developed 3 models that included a selection of respiratory parameters (FIO2, breathing frequency, or minute ventilation) that unexpectedly did not incorporate arterial blood gas or spirometric determinations. Even though our models performed better than other formulae in predicting V̇o2 in this particular population, the limits of agreement were wide, particularly in older individuals with shorter stature, higher heart rate, and lower MVV.
Oxygen consumption is an essential measurement for the calculation of cardiac output by the accepted standard, Fick methodology; however, in the cardiac catheterization laboratories, V̇o2 is commonly estimated instead of measured, given that it is time-consuming and requires especial equipment (ie, breath-by-breath metabolic cart analysis) and expertise. As a result, in clinical practice, V̇o2 is often estimated by a variety of equations that generally use a combination of variables such as BSA, age, gender, and heart rate.8,9,11
Several investigators have noted poor agreement between measured and estimated V̇o2 in different groups of adult patients.1,2,20–24 In large part, this is due to marked differences between the population in which the formulae were derived and the characteristics of the cohort studied. In fact, many factors can affect the measured V̇o2 including age,8,11 gender,8,11,13,23,25 BSA,8,11,13,23,25 fat-free mass, heart rate,11,13,25 degree of sedation,9,14,25 temperature (of the patient and environment),25 time of day, prior food intake, level of activity,25 and position.9
Although several parameters were associated with resting V̇o2 in multivariate models, only BSA, gender, heart rate, breathing frequency, need for supplementary O2, and minute ventilation were significant predictors. Because V̇o2 directly depends on the body size of a person, BSA is the most commonly used variable in available equations. In fact, BSA was used in all but two10,12 of the tested formulae for resting V̇o2 estimation. The formulae proposed by Krovetz et al10 uses height and weight and the equation by Lindahl12 only weight. In our study, BSA area was selected for our models instead of height and/or weight.
Men have a higher V̇o2 than women, given the higher fat-free mass. Therefore, gender was selected in three8,11,13 out of the 7 formulae tested. Similarly, heart rate, an indicator of stress that changes in relation to the metabolic rate, was used in 2 out of the 7 equations. Breathing frequency, need for supplementary O2, and minute ventilation were not included in prior equations that estimate resting V̇o2, likely because these parameters are more applicable to our particular group of subjects with respiratory symptoms. It is important to consider that, immediately before cardiopulmonary exercise testing or cardiac catheterization, patients are not in a strictly basal state as they may be anxious, a condition that may lead to an increase in minute ventilation, heart rate, and V̇o2.2
Age, as loge(age), was used in two8,11 formulae, obtained in cohorts that included children and adults. In the present study, we also noted that age was indirectly associated with V̇o2/kg; however, age or loge(age) was not selected in any of our 3 multivariate models that predict V̇o2. Furthermore, specific determinations of general value in pulmonary patients, that is, arterial blood gases or spirometry, did not add prognostic value.
The 3 models presented in our study had wide limits of agreement and were not able to explain more than 68% of the V̇o2 variance. In addition, adding parameters obtained from arterial blood gases and/or pulmonary function tests did not improve the performance of the equations to predict V̇o2, a fact that negated our working hypothesis. This could reflect that FIO2 (model 1), breathing frequency (model 2), and minute ventilation (model 3) are better and probably more comprehensive estimates of V̇o2 than other more specific parameters obtained from spirometry or arterial blood gases.
In general, the formulae that estimate V̇o2 are associated with systematic error as they tend to underestimate the higher V̇o2 values and overestimate the lower determinations. This is relevant because differences between the measured and estimated V̇o2 of ≥ 25% may result in large errors in the determination of Q̇ and calculation of dependent variables. In our study, we observed that the estimated V̇o2 was mostly inaccurate in older individuals with shorter stature, higher heart rate, and lower MVV. The reasons for the greater discrepancies between measured and estimated V̇o2 in this group of subjects remain unclear. A potential explanation include a nonlinear relationship between V̇o2 and age, height, MVV, or heart rate. It is possible that, due to the limited number of subjects at the end of these spectra, our model was not sensitive enough to detect this nonlinear association. Future studies testing particular subgroups of subjects may be of value to refine our equations.
The V̇o2 variance not captured in our study might be explained by variables that we did not measure, such as fat-free mass, core temperature, and general level of activity. It is also likely that there are unknown factors that might be responsible for the V̇o2 variation. Given that V̇o2 varies greatly and estimations are still generally inaccurate, we support the direct measurement of V̇o2 instead of its estimation. It is also important to point out that V̇o2 varies in patients with sepsis,26,27 hypovolemia,28 pulmonary hypertension,29 congestive heart failure,30 or COPD,31 both at baseline and with treatment.32 Thus, current formulae are likely inaccurate in these groups of patients.
Our study has limitations: (1) the results are only applicable to nonsedated patients at rest in a sitting position with respiratory symptoms and/or conditions, (2) body temperature and fat-free mass were not measured at the time of V̇o2 determination, and (3) we cannot exclude potential bias due to the retrospective nature of our study; however, cardiopulmonary exercise studies were performed by a few experienced respiratory therapists under similar conditions and following a prespecified protocol. Notwithstanding these limitations, our study emphasizes the lack of accuracy of formulae to predict V̇o2 and is the first to generate different models to estimate resting V̇o2 in subjects with respiratory symptoms or conditions, and to compare them with available formulae.
Conclusions
We developed more accurate formulae to predict oxygen consumption in the group of subjects who underwent cardiopulmonary exercise testing for respiratory symptoms and/or conditions. Arterial blood gases and spirometric variables did not significantly improve the predictive equations.
Acknowledgments
This study would not have been possible without the hard work of Cleveland Clinic respiratory therapists who performed the cardiopulmonary exercise test and contributed to the database.
Footnotes
- Correspondence: Adriano Tonelli MD, 9500 Euclid Avenue A-90, Cleveland, OH 44195. E-mail: tonella{at}ccf.org.
This publication was supported by CTSA KL2 grant TR000440 (to Dr Tonelli) from the National Center for Research Resources, a component of the National Institutes of Health, and by National Institutes of Health Roadmap for Medical Research. The authors have disclosed no other conflicts of interest.
- Copyright © 2015 by Daedalus Enterprises
