Maximal Blood Lactate Level Acts as a Major Discriminant Variable in Exercise Testing for Coronary Artery Disease Detection in Men
Background The interpretation of exercise stress testing for coronary artery disease detection is affected by the many differences in chosen variables and mathematical methods. We conducted a prospective trial to evaluate a global muscle fatigue parameter—the blood lactate level achieved at maximal exercise—as a method of distinguishing between diseased and nondiseased coronary status.
Methods and Results We evaluated 236 consecutive male patients without previous myocardial infarction who had been referred for the diagnosis of coronary artery disease. None of the patients had cardiomyopathy, severe cardiac heart failure, or valvular heart disease. Blood lactate concentration at maximal exercise was measured as well as other classic variables. Correlations between variables and coronary status as assessed by coronary arteriography were described using receiver operating characteristic (ROC) curves and logistic regression analysis. The first four most powerful variables (lactate level, maximal power output, exercise duration, and percentage of maximal predicted heart rate), which are directly representative of the global functional capacity, showed values of 0.777, 0.775, 0.760, and 0.740, respectively, by ROC curve analysis. Mean±SD blood lactate level at peak exercise reached 7.68±2.70 mmol/L in the 153 diseased and 10.56±2.75 mmol/L in the 83 nondiseased patients (P<.0001). After adjustment for other variables, blood lactate level remained a significant predictor of coronary artery disease by logistic regression analysis (adjusted odds ratio, 1.2; confidence interval, 1.04 to 1.4).
Conclusions Global muscle fatigue as assessed by lactate levels in the blood at maximal exercise appears to be a powerful distinguisher of diseased and nondiseased coronary status.
The present attitude regarding coronary artery disease (CAD) detection too often is based on coronary arteriography. Despite American Heart Association recommendations, recent publications reveal the use of stress testing before coronary arteriography to be as low as 29%.1 This attitude is reinforced because patients are anxious to be reassured, because of the large availability of balloon angioplasty, and because of the uncertainty of the diagnosis of CAD assessed by exercise testing. Exercise stress testing, which was first based on the discovery of ST-segment depression as an indicator of myocardial ischemia during exercise,2 was rapidly considered to be a tool for the diagnosis of CAD. In 1940, Riseman et al3 discussed the low ability of exercise to distinguish healthy from diseased subjects. The isolated analysis of ST segments was further discussed by Weiner,4 who described the use of this variable for the diagnosis of CAD equivalent to tossing a coin.
These controversies can be explained by the simultaneous analysis of several factors. Bayesian analysis has been used to determine the probability of disease by exercise stress testing. However, according to the exercise protocol used, the most determining variables are not always the same, the prevalence of the disease varies among centers and modifies the decision weight attached to a given parameter, and the equations are not easy to use and very often are not implemented in the diagnosis processing.
The most determining variable isolated by Ellestad,5 exercise duration, has been considered to be a good indicator of the overall exercise capacity. Thus, we investigated the discriminant capacity of another global muscular exhaustion parameter, the blood lactate level reached at maximal exercise capacity.
Most often, mathematical analysis is oriented toward building decision trees based on multivariate analysis. As areas under receiver operating characteristic (ROC) curves represent the magnitude of the information given by the most significant variables,6 we tried to investigate further, extracting from the ROC curve analysis information allowing better characterization of subgroups of patients and looking for combined variables that demonstrated very high sensitivity or specificity.
Five different cardiology centers were involved in the study. We prospectively evaluated 236 male patients aged 31 to 76 years (mean age, 55.2±8.9 years). Patients with an abnormal ECG at rest, ST-segment depression, or changes in T-wave polarity induced by hyperventilation and with prior myocardial infarction, known cardiomyopathy, severe congestive heart failure, or with valvular heart disease were excluded from the study.
Patients were referred for suspicion of CAD based on suggestive symptoms. In the absence of contraindication, a stress test was conducted systematically before any invasive procedure. Cardiac catheterization was then performed in all patients with a positive test result and in the cases specifically referred for cardiac catheterization, regardless of the exercise test result. The two procedures were performed within an interval of less than 3 months for each patient. Thus, 146 cardiac catheterizations (62%) were performed in patients with a positive exercise test and 90 were performed (38%) in patients with a negative exercise test.
All patients who were enrolled in the study underwent stress testing according to a continuous maximal graded symptom-limited protocol with the use of a bicycle ergometer, beginning at 30 to 60 W with 20- to 30-W increments each 3 minutes to a positive test or exhaustion. A 12-lead ECG was recorded on paper every 3 minutes during the test and the recovery period; symptoms were noted, and blood pressure was recorded at the same time intervals. End points for the test were reaching a positive stress test criteria or the development of exertion-limiting symptoms. Positive stress test criteria were horizontal or downsloping ST-segment depression of ≥1.0 mm or upsloping ST-segment depression of ≥1.5 mm at 60 milliseconds after the J point. Predictive maximal heart rate was calculated using the formula [210−(0.65×age)].7 The tests were always conducted after an interruption of medical treatment in accordance with specific drug half-lives. Stress test duration was planned to last 10 minutes; this was controlled by the cardiologist in charge of the test, using both the watts programmed at the beginning of the test and the increment.
Blood Lactate Level
Capillary blood samples were taken using a finger stick within 3 to 5 minutes after the end of the test. The samples were diluted in an hemolizing solution and stored at room temperature. The measurements were carried out within 10 days with a lactate analyzer (LA 640 Kontron, Roche Biomedics, F. Hoffman, La Roche et Cie SA). The methods of sampling, storage, and analysis have been previously validated.8
Selective coronary arteriography was performed in multiple projections using cranial and caudal angulations. A significant coronary stenosis was defined as ≥70% luminal diameter narrowing of at least one major coronary artery. All stenosis measurements were made twice with the use of calipers by the same operator who was blinded to the results of his first measurements. The coefficient of variability was <5%.
The reference was the result of the coronary arteriography: a patient demonstrating a significant coronary artery stenosis was considered to be diseased; in the opposite situation, he was declared to be nondiseased.
Statistical analysis was performed to evaluate the ability of each variable level to discriminate nondiseased from diseased subjects. Thus, the dependent variable was the disease status. The independent variables analyzed were age, weight, maximal lactate level at the end of the exercise, test duration, maximal power output (W), maximal heart rate, percentage of maximal predicted heart rate, maximal systolic and diastolic blood pressures, maximal heart rate multiplied by maximal systolic blood pressure, angina, maximal ST depression, and work capacity index (maximal power output divided by the weight).
ROC curve analysis was used,6 9 with calculations of the areas under the ROC curves represented by the letter W. This quantity represents the probability for a randomly chosen subject free of disease to demonstrate a variable level higher than the level observed among randomly chosen diseased subjects. A W value of 0.5 means the variable level distributions are similar in both populations; conversely, a W value of 1.0 means that the two populations’ variable distributions do not overlap. The ROC curve also allows determination of variable thresholds according to the target sensitivity or specificity. A different test, a χ2 analysis, was used to test the dependence between angina during the test, a nominal variable, and disease.
Logistic regression was also used to analyze the data via a simple logistic regression of the disease status versus each of the covariates to verify the ROC curve analysis previously performed. Multiple logistic regression analysis was then used to evaluate the strength of the association of each variable with diseased status, after adjustment for the other variables.
A first decision tree was built using the discriminant variables as indicated by the ROC curves and the logistic regression analysis. The first variable used was the one determining better separation between diseased and nondiseased subjects. The other variables were then introduced according to a descending order of discriminative capacity. For each continuous variable, the cutoff value acting as a separator to make the decision was chosen as that offering the smallest number of classification errors, ie, that which minimizes the sum of false-positive and false-negative results. A second decision tree, called operational, was built with the intention to select the combinations of the value thresholds of the variables leading to the best possible identification of nondiseased and diseased populations. This allowed isolation of homogeneous nondiseased or diseased populations. It was made by identifying the common characteristics of groups of nondiseased or diseased populations. This exploration was made by progressively increasing or decreasing by increments the thresholds of the variables until a given population was isolated. Only the variables identified as discriminant were included in this particular screening process. Every variable combination was tested to characterize populations of nondiseased and diseased subjects.
Data were analyzed with statview and jmp (SAS Institute). Differences were considered significant at P<.05. Values are expressed as mean with a 95% standard deviation (mean±SD).
After coronary arteriography, CAD was diagnosed in 153 patients; among them, 91, 39, and 23 presented with a significant stenosis of one, two, and three vessels, respectively. CAD was absent in 83 patients.
Mean±SD blood lactate level at peak exercise (Fig 1⇓) reached 7.68±2.70 mmol/L in the 153 patients with CAD and 10.56±2.75 mmol/L in the 83 patients without CAD (P<.0001).
An ROC curve (continuous data) was built for continuous variables. They were classified according to their ability to distinguish diseased from nondiseased patients. In Table 1⇓, the obtained W values for each continuous variable are shown and listed by descending order. The variables with significant W values are lactate level, maximal power output, exercise duration, percentage of maximal predicted heart rate, maximal ST depression, age, and double product. The statistical significance of lactate level is not different from that of maximal power output, exercise duration, or percentage of maximal predicted heart rate by ROC curve analysis; it is significantly different from that for maximum ST depression. For each continuous variable, the thresholds giving a chosen sensitivity or specificity were calculated and listed in Table 2⇓.
Angina during exercise (nominal data) and disease status are dependent (χ2, P<.0001). The sensitivity of the variable angina during exercise is 56% and specificity is 87%.
The five most discriminant continuous variables in Table 1⇑ did not significantly correlate with an increased severity of CAD in terms of the number of diseased vessels; however, their values were significantly different in diseased subjects (P<.0001) than the corresponding values observed in nondiseased subjects (Table 3⇓).
A separate simple logistic regression analysis of the disease status versus each of the covariates was performed to verify the ROC curve analysis previously performed and to select a reasonable subset of explanatory variables for further investigation without too much power being lost in the statistical procedures. Maximum likelihood estimates were obtained for each of the important explanatory covariates (Table 4⇓). These results confirmed the ROC analysis. The most significant variables retained for further analysis by regression trees and classification are illustrated in Tables 1⇑ and 4⇓.
With the use of a multiple logistic regression analysis, only two variables appeared to be independently and significantly associated with disease status: the level of lactate at peak exercise and the presence of angina during the stress test (Table 4⇑), indicating that all other variables are expressed through these two and thus reinforcing the variable lactate level as a major discriminant. After adjustment for all other variables, such as maximal power output, exercise duration, maximal predicted heart rate, maximal systolic and diastolic blood pressures, maximal ST depression, occurrence of angina, age, and weight, lactate level remained a significant predictor of CAD status (adjusted odds ratio, 1.2; confidence interval, 1.04 to 1.4). The other significant independent predictor of CAD was angina during the test (adjusted odds ratio, 2.5; confidence interval, 1.6 to 3.9).
The most significant variables from the ROC curves allowed us to build a decision tree with the first eight variables of Table 1⇑. Tree-specific thresholds were calculated for each of these variables (Fig 2⇓). The first variable was lactate, with a threshold found at 11.0 mmol/L. The tree’s expansion was limited by the number of patients and left 46 patients undetected (30.06% of diseased patients). The positive and negative predictive values were 90.7% and 80.4%, respectively.
Thus, we used a combination of several variables from the ROC curves with specific threshold values presenting a strong negative predictive value and a high sensitivity to build the operational tree; they are illustrated in Fig 3⇓. The combinations described are always headed by the threshold of lactate value because this variable was the only one always present as discriminant. For a high negative predictive value (96%), a lactate level as high as 15 mmol is sufficient on its own; when this value is lower, other variable thresholds have to be progressively introduced to keep the negative predictive value sufficiently high. It is not surprising that angina is the next variable, followed by maximal power output and percentage of maximal predicted heart rate. For a high positive predictive value (97%), lactate level alone is sufficient when below 5.8 mmol. For lactate values above 5.8 mmol, angina and percentage of maximal predicted heart rate have to be added. The overall negative predictive value reaches 96%, and the sensitivity is 99% (82% to 99%). Only ≈40% of the patients were identified by these combinations. This tree illustrates a simple but fundamental fact—the variation of the predictive weight of a particular factor in a given set of predictive factors with exercise intensity. However, calculation of the predictive weight of each predictive variable, for a given lactate level, was not possible due to the limited number of patients involved.
In a separate analysis of the predictive value of the variable ST-segment depression alone, stress tests were found to be positive because of ST-segment depression in 150 patients and negative in 86. Among these 150 patients, 112 had the disease. Thus, the sensitivity of ST-segment depression alone was 73.8% and the specificity was 55%.
In several investigations, the level of global performance during exercise clearly appeared to be a major criterion for determining the presence or absence of CAD, with the lowest levels being associated with a disease status and the highest being associated with a nondiseased status.5 10 A biological marker of whole body maximal exercise intensity—blood lactate level—was thus chosen to test its relation with CAD. We hypothesized that the lower the lactate value is at peak exercise, the more severe would be the CAD, and vice versa.
The intent to use maximal lactate level as a marker for fitness was reinforced by the observation of its interest when evaluating sportsmen. The ability of maximal lactate level to predict the performance of athletes has been clearly outlined.11 12 13 Also, exercise limitation of diseased patients, eg, as in chronic obstructive pulmonary disease, is often correlated with a limitation in cardiac output adaptation.14 Although aware of its variability, we chose to measure lactate because its measurement is very easy to perform with automatic devices, it can be performed within seconds, and the procedure is inexpensive.
In the population we investigated, lactate level at peak exercise clearly appeared to be a potent discriminant factor by ROC curve analysis. The importance of blood lactate level compared with other variables was underlined by the result of the multiple logistic regression analysis, indicating that this variable presents a major discriminant power and that it is independently associated with disease status (adjusted odds ratio, 1.2; confidence interval, 1.04 to 1.4). Two facts explain why the increase in blood lactate was limited in the presence of CAD, in the absence of any functional impairment. First, exercise was stopped when there was any evidence of a positive stress test, limiting the metabolic synthesis of lactate. Second, patients with severe ischemic symptoms stopped the test on their own with the same consequences on blood lactate level. However, the variable lactate does not allow separation of diseased patients into subgroups with different disease severity, ie, one-, two-, or three-vessel disease. Also, the large standard deviations of lactate values in the diseased and nondiseased populations are responsible for an important overlap. This overlap reinforces the need for a multivariate approach; this is illustrated in Fig 3⇑, in which it can be seen that maximal exercise intensity expressed by lactate level modifies the relative weight of each predictor in a given set of predictive variables according to the level of the lactate value. At both extremes, ie, maximal and minimal levels, lactate value carries by itself the entire weight of the predictive power, whereas for intermediate values, this weight is shared with other variables.
Angina was analyzed separately from continuous variables. However, the logistic regression analysis shows the importance of this symptom during the test as an independent predictor of CAD (adjusted odds ratio, 2.5; confidence interval, 1.6 to 3.9). This further underlines the need to use decisional trees to interpret stress testing for CAD detection.
All patients were free of cardiomyopathy, severe cardiac heart failure, and valvular heart disease. The exercise limitations introduced by these diseases would have altered the interpretation of lactate values at maximal exercise for CAD detection, introducing low values without clear association with CAD. The patients were also free of previous myocardial infarction, which would have required a more specific test to assess the presence of residual myocardial ischemia, such as a thallium scintigraphy.
With these limitations, these results underline the importance of global performance as an indicator of the presence or absence of CAD. They are in accordance with the results of Fortini,10 who found that maximal oxygen consumption was a very good determinant of CAD; however, these authors did not have reference coronary angiograms. Our results are also in accordance with Ellestad’s findings.5 The two variables that we added, compared with their list, were lactate level at peak exercise and maximal power output. These two variables are ranked as first and second, respectively, by ROC curve analysis. Among other variables that have commonly been examined, we found nearly the same hierachy as Ellestad for the most determinant ones: percentage of maximal predicted heart rate, ST depression, and age. Regarding ST depression, it must be underlined that stress tests were interrupted when a patient demonstrated a significant ST depression; thus, the statistical significance of ST depression is lower than if the patient had been asked to exercise 3 or 4 minutes longer; this strategy would probably have decreased the predictive value of the lactate level. For Ellestad, the ST-segment variable was less discriminant than duration of exercise and percentage of maximal predicted heart rate. We thus confirm these data and add that it is also less precise than lactate and maximal power output.4
One of the variables most frequently used to assess exercise intensity is heart rate. We chose a target heart rate calculated by the formula [210−(age×0.65)] as it has been considered to be more appropriate for people older than age 65 compared with the classic formula (220−age), without modifying the calculated target heart rate for younger people.7 When compared with lactate level, heart rate did not have an equivalent predictive power. This difference in predictive power can be explained by the fact that at maximal exercise intensity, heart rate reaches an asymptote and is not modified further by one or more exercise stages, which gives to the values beyond 85% of maximal predicted heart rate too limited a scale to accurately reflect the possible large variations in exercise intensity; on the contrary, lactate level does not vary much at low-intensity exercise but increases the most at the last stages of maximal exercise when one is close to maximal power output. Even when considering that lactate level is highly variable, its large increase at the end of a maximal exercise makes it able to separate different exercise intensities. Futhermore, the predicted maximal heart rate is far from being perfectly fitted to a curve, and in our population, the maximum reached by the nondiseased subjects differed from the above formula by 10.7%. It could be inferred that heart rate should be a better separator at the lowest exercise intensities, where lactate level does not show large variations, whereas lactate level would become a better discriminant factor at higher exercise intensities, when subjects are close to their maximal power output.
The variable “duration of exercise” on a treadmill was the first indicator of CAD for Ellestad, whereas it was not the case in our analysis. Duration of exercise, as measured according to our protocol, cannot be directly compared with the value of duration of exercise measured when using a treadmill. In our standard exercise protocol, the operator arbitrarily chooses the initial work load, as well as the additional work load added at each stage, trying to reach exhaustion in ≈10 minutes. In that case, the percentage of the target duration of exercise should be used and not the duration per se. However, the use of percentages precludes direct use of exercise duration as a predictive variable and does not allow a comparison of intensities of exercise. Equations have been described15 that allow conversion of performance on a treadmill based on weight, speed, and slope to performance on a bicycle directly expressed in watts. However, this approach loses its reliability at high exercise intensities5 and would be cumbersome to use on a regular basis. Peak exercise lactate level appears to be a better common denominator between these two types of exercise tests and is easier to use and to measure in the same way the intensity of exercise in both conditions, particularly at high exercise intensities. In other words, lactate level at peak exercise somewhat standardizes exercise intensities obtained when a treadmill or bicycle is used.
The data of the operational tree were compared with those of Morise and Duval,16 who analyzed the same population by using the Cadenza, tabular, and a composite dual Bayes analysis. Dual Bayes is a mixture of Cadenza and tabular, with the Cadenza being used first and then the tabular adjustment being used when the Cadenza gives a posttest probability of >50%. Compared with the dual Bayes analysis, we found better negative (96% versus 85%) and positive (97% versus 77%) predictive values, with the same percentage (59% versus 60%) of undeterminate diagnoses. However, this improvement was not statistically significant.
In conclusion, blood lactate concentration at peak exercise appears to be a strong independent predictor of CAD in men; the decision tree that we built using the discriminant power of lactate level should now be tested in an independent population.
This work was supported in part by a grant from the Région Rhône-Alpes.
Reprint requests to Dr J.C. Barthélémy, Laboratoire de Physiologie-GIP Exercice, CHU Nord-Niveau 6, F-42055 Saint-Etienne, Cedex 2, France.
- Received June 27, 1995.
- Revision received October 25, 1995.
- Accepted November 3, 1995.
- Copyright © 1996 by American Heart Association
Topol EJ, Ellis SG, Cosgrove DM, Bates ER, Muller DWM, Schork NJ, Schork MA, Loop FD. Analysis of coronary angioplasty practice in the United States with an insurance-claims data base. Circulation. 1993;87:1489-1497.
Blousfield G. Changes in electrocardiogram during paroxysm. Lancet. 1918;2:457-463.
Ellestad MH. Stress Testing. Philadelphia, Pa: FA Davis; 1986:505.
Jones NL. Clinical Exercise Testing. Philadelphia, Pa: WB Saunders; 1988:173.
Fortini A. Anaerobic threshold in patients with exercise-induced myocardial ischemia. Circulation. 1991;83(suppl III):III-50-III-53.
Wagner P, Gale G. Ventilation-Perfusion Relationships. New York, NY: Marcel Dekker; 1991:121-142.
Givoni B, Goldman RF. Predicting metabolic energy cost. J Appl Physiol. 1971;30:429-433.