Exercise Intolerance in Adult Congenital Heart Disease
Comparative Severity, Correlates, and Prognostic Implication
Background— Although some patients with adult congenital heart disease (ACHD) report limitations in exercise capacity, we hypothesized that depressed exercise capacity may be more widespread than superficially evident during clinical consultation and could be a means of assessing risk.
Methods and Results— Cardiopulmonary exercise testing was performed in 335 consecutive ACHD patients (age, 33±13 years), 40 non–congenital heart failure patients (age, 58±15 years), and 11 young (age, 29±5 years) and 12 older (age, 59±9 years) healthy subjects. Peak oxygen consumption (peak V̇o2) was reduced in ACHD patients compared with healthy subjects of similar age (21.7±8.5 versus 45.1±8.6; P<0.001). No significant difference in peak V̇o2 was found between ACHD and heart failure patients of corresponding NYHA class (P=NS for each NYHA class). Within ACHD subgroups, peak V̇o2 gradually declined from aortic coarctation (28.7±10.4) to Eisenmenger (11.5±3.6) patients (P<0.001). Multivariable correlates of peak V̇o2 were peak heart rate (r=0.33), forced expiratory volume (r=0.33), pulmonary hypertension (r=−0.26), gender (r=−0.23), and body mass index (r=−0.19). After a median follow-up of 10 months, 62 patients (18.5%) were hospitalized or had died. On multivariable Cox analysis, peak V̇o2 predicted hospitalization or death (hazard ratio, 0.937; P=0.01) and was related to the frequency and duration of hospitalization (P=0.01 for each).
Conclusions— Exercise capacity is depressed in ACHD patients (even in allegedly asymptomatic patients) on a par with chronic heart failure subjects. Lack of heart rate response to exercise, pulmonary arterial hypertension, and impaired pulmonary function are important correlates of exercise capacity, as is underlying cardiac anatomy. Poor exercise capacity identifies ACHD patients at risk for hospitalization or death.
Received December 15, 2004; revision received April 20, 2005; accepted April 28, 2005.
Many patients with adult congenital heart disease (ACHD) report unlimited exercise capacity, but self-reported status may be unreliable in ACHD,1 as in chronic heart failure.2 In those ACHD patients who do perceive limitation, the underlying mechanisms are not well understood. Furthermore, it is not clear whether measurement of exercise capacity is useful in assessing risk.
Quantifying exercise capacity by measurement of peak oxygen consumption (peak V̇o2) is an established investigation in the management of patients with chronic heart failure,3,4 but corresponding studies in ACHD either have been restricted to isolated patient groups or have reported on data acquired over >1 decade.5–10 As a result, it is unclear whether different cardiac lesions cause different degrees of exercise intolerance and whether there is any prognostic value in formal measurement of peak V̇o2 in the ACHD population. Identifying the factors associated with impaired peak V̇o2 may help to improve our understanding of the physiological derangement and ultimately may assist in targeting clinical care for ACHD patients.
The aims of our study were (1) to evaluate objective exercise capacity in a large cohort of ACHD patients, (2) to clarify correlates of exercise capacity across the anatomic spectrum of ACHD, and (3) to investigate whether the degree of objective exercise intolerance has prognostic implications in these patients.
We analyzed retrospectively all cardiopulmonary exercise tests performed in ACHD patients between January 2003 and May 2004 at the Royal Brompton Hospital (London). Patients were referred for exercise testing as part of routine clinical follow-up protocols used for patients with ACHD at our institution. Informed consent was obtained from all patients undergoing exercise testing. Almost all patients underwent only 1 test during the period; we addressed only the last test for those who underwent 2 tests. A main diagnosis was determined for every patient from hospital records. Patients with multiple, complex cardiac lesions substantially affecting hemodynamics were classified as complex anatomy (comprising mainly patients with single-ventricle physiology). The NYHA class was determined by physician assessment of patients’ self-reported symptoms before the date of the exercise test. In addition, data on resting saturations in air and residual left-to-right shunts in patients with biventricular heart circulation were recorded. Cyanosis was defined as resting saturations ≤90% in air after 2 minutes of complete rest. The presence of pulmonary arterial hypertension was also recorded on the basis of Doppler echocardiography (estimated systolic right ventricular pressure >35 mm Hg in the absence of pulmonary stenosis) and/or cardiac catheterization data according to current guidelines.11
Cardiopulmonary exercise testing was performed on a treadmill according to a modified Bruce protocol12,13 with the addition of a “stage 0” in which the patient walks at a velocity of 1 mile/h and a gradient of 5% for 3 minutes. All subjects were encouraged to exercise to exhaustion regardless of the maximal heart rate achieved. Ventilation, oxygen uptake, and carbon dioxide production were measured continuously with a respiratory mass spectrometer (Amis 2000, Innovision). The peak respiratory exchange ratio (RER) was calculated as the ratio of carbon dioxide production to oxygen consumption at peak exercise. Heart rate was measured by continuous electrocardiography, and arterial blood pressure was recorded manually by sphygmomanometry.
Standard spirometry was available in 160 patients. Forced expiratory volume in 1 second (FEV1) and forced vital capacity (FVC) were measured before exercise testing and are expressed as percentages of predicted values.
Systemic and subpulmonary ventricular function (systemic) was assessed semiquantitatively by an experienced echocardiographer (n=273) blinded to the exercise data and quantified as previously described.14 The following grades were used: 1=normal, 2=mildly impaired, 3=moderately impaired, and 4=severely impaired systolic function.
Patients were followed up after exercise testing, and the time taken to reach an end point of death or hospitalization at our institution was recorded.
Healthy volunteer data were obtained from cardiopulmonary exercise tests undergone by healthy subjects for research projects at our institution using the same exercise protocol and measurement equipment as ACHD patients.
All values are presented as mean±SD. Comparisons between groups were made with Student t test, Mann-Whitney U test, or the χ2 test as appropriate. For statistical analysis, NYHA class was considered a continuous variable, and Spearman rank correlation was used to estimate correlation between NYHA class and peak V̇o2. Relationships between variables were initially studied by single-variable regression, with significant variables subsequently incorporated into stepwise forward multiple regression models. Single-variable Cox proportional-hazards analysis was used to assess the association between variables and hospitalization-free survival. Significant single-variable prognosticators were subsequently included in a multivariable Cox proportional-hazards analysis (stepwise forward). For all analyses, a value of P<0.05 was considered significant. Receiver-operating characteristic curve analysis was used to assess area under the curve. The relationship between variables and the frequency or duration of hospitalization was assessed with a zero-inflated negative binomial regression model, which accounts for the number or length of hospitalization and the time at risk and deals with the excess number of patients with 0 hospitalizations. SPSS (SPSS Inc) was used for all the analyses except for the zero-inflated negative binomial regression, which was analyzed with Stata 7.0 for Windows (Stata Corp).
Three hundred thirty-five consecutive ACHD patients were included (Table 1). The diagnoses of 314 patients are shown in Table 2 (the remaining 21 had miscellaneous lesions such as AV septal defects, transposition of the great arteries (TGA) after arterial switch operation, and lesions after Rastelli-type procedures). Overall, 49% of ACHD patients were asymptomatic, whereas 41% of patients were in NYHA class II and only 10% were in NYHA class III. The vast majority of patients (n=296) were in sinus rhythm at the time of exercise testing. Only 13 patients had underlying supraventricular arrhythmias (9 atrial fibrillation, 3 atrial flutter, and 1 complete AV block), whereas 26 had a pacemaker (dual chamber, 65%; rate responsive, 85%; upper sensor rate, 142±10 minutes−1). In addition, from the exercise database from our institution on ambulatory patients with acquired chronic heart failure in that period of time, we selected the first 40 whose only stated diagnosis was chronic heart failure (ie, without significant independent lung or musculoskeletal disease). In this reference group (see Table 1), 45% had ischemic etiology, and mean left ventricular ejection fraction was 35±14%. Finally, we recruited from the same database an additional reference group of healthy subjects of age similar to that of the ACHD and chronic heart failure patients comprising 11 younger subjects with a mean age 29.0±4.9 years (6 men) and 12 older subjects with a mean age of 59.2±9.3 years (6 men), respectively.
Impact of Underlying Cardiac Lesion on Exercise Capacity
A gradual decline in peak V̇o2 was found across the spectrum of congenital heart disease (Figure 1). Patients after repair of aortic coarctation demonstrated the highest peak V̇o2 values (28.7±10.4 mL · kg−1 · min−1). Patients with congenitally corrected TGA (18.6±6.9 mL · kg−1 · min−1), complex anatomy (14.6±4.7 mL · kg−1 · min−1), and Eisenmenger physiology (11.5±3.6 mL · kg−1 · min−1) demonstrated the worst exercise capacity (P<0.0001, ANOVA). This result persisted after inclusion of age and gender as possible confounders in an ANCOVA (P<0.001).
Depressed Peak V̇o2, Even in Asymptomatic Patients
Peak V̇o2 was reduced in ACHD patients compared with healthy subjects of similar age (21.7±8.5 versus 45.1±8.6 mL · kg−1 · min−1; P<0.0001). Even supposedly asymptomatic ACHD patients (NYHA I) showed dramatically lower peak V̇o2 values (Figure 2) than healthy subjects of similar age (26.1±8.2 versus 45.1±8.6 mL · kg−1 · min−1; P<0.0001). In fact, ACHD patients had peak V̇o2 values equivalent to those of heart failure patients of corresponding NYHA class (Figure 3). This equivalence of peak V̇o2 values between conditions occurred despite the great disparity in peak V̇o2 between the corresponding age reference groups (45.1±8.6 versus 28.0±3.5 mL · kg−1 · min−1; P<0.0001). Exercise capacity declined gradually between functional classes in ACHD (Figure 3).
Correlates of Peak V̇o2
On single-variable regression analysis, several characteristics were related to exercise capacity: age, age at surgery, gender, body mass index, %FEV1, %FVC, systemic and pulmonary ventricular function, presence of a residual left-to-right shunt, pulmonary arterial hypertension, cyanosis, and NYHA class (Table 3). During exercise testing, peak heart rate and peak systolic blood pressure were also associated with peak V̇o2 in single-variable analysis. No significant correlation was found between resting heart rate, resting blood pressure, QRS duration, creatinine or sodium, and peak V̇o2 values.
Including the significant predictors of peak V̇o2 on single-variable analysis (except age at surgery and residual left-to-right shunts that, by definition, would exclude unoperated patients and single-ventricular physiology) in multivariable analysis revealed that peak heart rate, %FEV1, pulmonary arterial hypertension, gender, and body mass index remained correlated with peak V̇o2 (R2=0.53; Table 4). Systemic and pulmonary ventricular function was not retained in this model. When systemic and pulmonary ventricular function was subsequently excluded from the multivariable analysis (thus increasing the number of observations used for analysis), the significant correlates of peak V̇o2 were peak heart rate, gender, body mass index, %FEV1, pulmonary hypertension, NYHA class, and cyanosis (R2=0.54).
Impact of Cardiac Rhythm, β-Blockade, and Pacemakers on Peak V̇o2
Patients in sinus rhythm at the time of exercise testing had a higher peak heart rate (156.1±29.8 versus 140.4±39.5 bpm; P=0.003) and a higher peak V̇o2 (22.1±8.6 versus 18.5±7.1 mL · kg−1 · min−1; P=0.015). There were too few patients in atrial fibrillation or atrial flutter for subanalysis for a potential effect of different types of arrhythmia on peak V̇o2. Patients on β-blockade (n=36) had lower peak heart rate (130.8±39.8 versus 157.3±29.3 bpm; P<0.0001) and peak V̇o2 (17.9±8.0 versus 22.1±8.3 mL · kg−1 · min−1; P<0.0001). Similarly, patients with a permanent pacemaker also had lower values of peak heart rate (133.6±25.9 versus 156.0±31.3 bpm; P<0.001). However, there was only a trend toward lower peak V̇o2 in this latter group (19.5±7.1 versus 21.9±8.6 mL · kg−1 · min−1; P=0.17).
Pulmonary Function, Reasons for Test Termination, and Impact of Respiratory Exchange Ratio
Pulmonary function was depressed in ACHD patients (Table 1) but no different between cardiac lesions. The vast majority of patients terminated exercise because of either shortness of breath (n=136) or fatigue (n=142). In 57 patients, other reasons such as presyncope, syncope, arrhythmias, or a substantial drop in oxygen saturations caused test termination. There was no statistically significant difference between reason for test termination and specific diagnostic groups.
Overall, the mean RER in ACHD patients was 1.04±0.13. Cyanotic patients were less likely to reach an RER of 1.0 compared with noncyanotic ACHD patients (P<0.001), with only 50% of cyanotic patients reaching this value in our study. In healthy subjects, chronic heart failure patients, and acyanotic ACHD patients, RER and peak V̇o2 were related. However, a gradual decline in correlation coefficients between RER and peak V̇o2 was found from healthy individuals over heart failure patients to acyanotic ACHD patients (r=0.70 to r=0.50 and r=0.39, respectively; P<0.01 for each). In contrast there was no such correlation in cyanotic ACHD patients (r=0.09, P=0.50). In addition, patients with an RER <1.0 had no significantly different prognosis compared with patients with an RER >1.0 (log-rank P=0.08; indeed, the trend was toward a worse prognosis).
Predictors of Hospitalization or Death
Follow-up data were available for 328 patients. During a median follow-up of 304 days from index cardiopulmonary exercise testing (range, 17 to 580 days), 60 patients (17.9%) were hospitalized for a total of 454 days. The median hospitalization frequency was 1 (range, 1 to 6), with a median hospitalization of 4 days (range, 1 to 54 days). Nine patients (2.7%) died; hospital admission or death occurred in 62 patients (18.5%).
Peak V̇o2 as a continuous variable significantly predicted the combined end point of hospitalization or death (P=0.02) and death alone (P=0.04). The area under the receiver-operating characteristics curve was 0.75 (95% CI, 0.63 to 0.85) for death alone and 0.68 (95% CI, 0.63 to 0.73) for hospitalization or death. Peak V̇o2 <15.5 mL · kg−1 · min−1 (corresponding to the 25th percentile for peak V̇o2 in our sample) conferred a higher risk of hospitalization or death (hazard ratio, 2.9; 95% CI, 2.2 to 7.4; P<0.0001) and death alone (hazard ratio, 5.6; 95% CI, 1.4 to 31.2; P=0.02) than higher peak V̇o2 values (Figure 4). Other predictors of hospitalization or death on single-variable analysis were diagnosis, NYHA class, age at surgery, and peak heart rate. Age, gender, body mass index, cyanosis, %FEV1, %FVC, resting heart rate, resting blood pressure, creatinine, sodium, and peak blood pressure were not related to hospitalization or death on single-variable analysis.
On multivariable Cox analysis, only peak V̇o2 (hazard ratio, 0.94 per mL · kg−1 · min−1; P=0.01) and NYHA class (hazard ratio, 2.15 with each increase in NYHA class; P=0.002) predicted hospitalization or death during follow-up (Table 5). Furthermore, peak V̇o2 was related to the frequency [exp(β)=0.91; 95% CI, 0.85 to 0.98; P=0.01] and duration [exp(β)=0.90; 95% CI, 0.83 to 0.98; P=0.01] of hospitalization adjusted for follow-up time, even after accounting for NYHA class, age, age at surgery, gender, and laboratory parameters.
Our study shows that ACHD patients have markedly depressed exercise capacity (even if they consider themselves asymptomatic) to an extent similar to that seen in non–congenital heart failure patients. The presence and severity of symptoms signify a worse objective exercise capacity in these patients. Chronotropic response to exercise is an important correlate of exercise capacity, as are pulmonary arterial hypertension, cyanosis, pulmonary function, and the underlying cardiac anatomy. Impaired exercise capacity predicts hospitalization and death over the following year.
Exercise Intolerance in Different ACHD Subgroups
Peak V̇o2 is an established and reliable measure of exercise intolerance that is widely used to assess patients with heart failure.3,4 We found a significantly reduced peak V̇o2 across the spectrum of ACHD compared with healthy subjects of similar age. This finding is consistent with prior studies that have reported exercise limitation in ACHD patients.5–10 Our study extends this previous work, showing that the degree of exercise intolerance is related to underlying anatomical features. Patients with simple, noncyanotic lesions with biventricular hearts without pulmonary arterial hypertension had a far better exercise capacity than patients with complex or cyanotic lesions with pulmonary arterial hypertension.
Patients with Eisenmenger physiology and complex anatomies (comprising mainly those with single-ventricle physiology) exhibited a markedly impaired exercise capacity in the present study. Despite being congenitally or surgically “corrected,” patients with TGA also had significantly reduced peak V̇o2 values compared with healthy subjects. Most congenitally corrected TGA patients had associated lesions, potentially explaining their worse peak V̇o2 values compared with Mustard patients. It should also be emphasized that atrial septal defect closure was performed late in most of the atrial septal defect patients included in our study, who themselves represent the oldest patient subgroup. In contrast, higher peak V̇o2 values were recorded in patients with a ventricular septal defect who underwent early surgical repair and were significantly younger than atrial septal defect patients.
Exercise Intolerance and NYHA Class
The NYHA classification, originally established for patients with chronic heart failure, is now widely used in congenital heart disease. It represents a simple classification of exercise intolerance based on subjective symptoms. To the best of our knowledge, there are no data assessing the relationship between peak V̇o2 and functional class in a large cohort of ACHD patients. In our study, NYHA class does stratify patients, distinguishing patients with mild impairment from those with moderate or severe impairment of objective exercise capacity. However, our results also suggest that NYHA class underestimates the true degree of exercise limitation in ACHD patients. Even asymptomatic ACHD patients exhibit markedly impaired peak oxygen consumption, 42% below healthy subjects. It is likely that ACHD patients have made lifelong adaptations to their cardiovascular disease and its slow progression, so they are not aware of the true extent of their exercise intolerance. In our study, ACHD patients had exercise capacities as poor as those of patients with acquired chronic heart failure, even though the latter were much older.
Other Correlates of Exercise Intolerance in ACHD
So far, only limited attempts have been made to assess systematically the impact of different physiological variables on exercise capacity across the spectrum of ACHD. For example, it is unknown to what extent cardiac function correlates with exercise capacity in ACHD. Our results suggest that exercise intolerance in these patients is multifactorial. Numerous variables were associated with peak V̇o2. Systemic and pulmonary ventricular function determined peak V̇o2 on single-variable regression. However, after accounting for other variables related to peak V̇o2, ventricular function had no additional predictive value. In contrast, peak heart rate and pulmonary function were the most powerful predictors of exercise capacity across the spectrum of ACHD even after accounting for NYHA class. Not surprisingly, therefore, factors that impair chronotropic response to exercise such as β-blockade or abnormal cardiac rhythm were associated with a lower peak V̇o2. Previous studies have shown that some ACHD patients have impaired pulmonary function15 and depressed peak heart rate.5 The present study extends these findings by demonstrating that both heart rate response and pulmonary function correlate positively with peak V̇o2 even after accounting for other physiological parameters. In addition, our study demonstrates that pulmonary arterial hypertension and cyanosis are strong correlates of poor exercise capacity and that the combination of both, as in patients with Eisenmenger physiology, severely affects exercise capacity.
Despite including all correlates of peak oxygen consumption in a best model, 46% of the variance in peak V̇o2 remained unexplained. This finding is consistent with reports from non–congenital heart failure patients.16
Exercise Intolerance Identifies Patients at Risk of Hospitalization or Death
Hospitalization is a major hazard for ACHD patients, associated with substantial costs and an amplified risk of subsequent mortality.17,18 Risk assessment has the potential to assist therapeutic targeting and appropriate resource allocation for ACHD services.
The present study demonstrates that peak V̇o2 is a predictor of hospitalization or death even after accounting for age, gender, NYHA class, laboratory parameters, and underlying cardiac lesion in contemporary ACHD patients. Patients with a worse exercise capacity were more likely to be admitted to hospital and spent more days hospitalized than patients with higher peak V̇o2 values. Patients with a peak V̇o2 <15.5 mL · kg−1 · min−1 (lowest quartile) demonstrated a 2.9-fold increased risk of hospital admission or death compared with patients with a peak V̇o2 ≥15.5 mL · kg−1 · min−1. The results of the present study indicate that in addition to peak V̇o2, higher NYHA class is associated with an increased risk of hospitalization or death. Although NYHA class is used widely, the underlying criteria are subjective and the reproducibility is low.2 In contrast to NYHA, peak V̇o2 is a continuous variable that represents a more objective and reproducible method19 for assessing the severity of exercise limitation.20
Peak V̇o2 predicted all-cause mortality in ACHD patients. However, because of the small number of deaths in our study, multivariable survival analysis was not possible.
Ventricular function was assessed semiquantitatively at rest with transthoracic echocardiography by a single experienced investigator. Additional imaging with cardiac magnetic resonance and possibly assessment of cardiac function during exercise may show stronger correlations with peak V̇o2. The population of ACHD patients used in this study reflects the clinical and research workload of a tertiary ACHD center. Hence, we cannot exclude the possibility that the patients in the study could be a biased sample, favoring those with more symptoms and lower perceived functional capacity. Further studies with secondary and community-based ACHD patients with longer follow-up may provide additional insights into the correlates of exercise intolerance and should examine the potential benefits of physical training on exercise capacity, hospitalization, and survival in this patient group. Our study did not specifically address coronary artery disease and its potential contribution to exercise intolerance. However, the population was young (mean age, 33 years), and only 4 patients had a history of coronary disease or coronary intervention. Finally, a larger study may allow sufficient power to identify the prognostic value of peak V̇o2 within individual anatomic subgroups of ACHD.
Exercise intolerance is prevalent in ACHD, even among asymptomatic patients, and is as severe as it is in patients with non–congenital heart failure. Chronotropic response to exercise, pulmonary arterial hypertension, and impaired pulmonary function are important correlates of exercise intolerance, as is the underlying cardiac anatomy. Impaired exercise capacity is associated with an increased risk of hospitalization or death (independent of symptom status), suggesting that measurement of peak oxygen consumption should be incorporated in the routine, periodic assessment of ACHD patients because it may serve as a useful prognostic tool.
Darlington Okonko and Sonya Babu-Narayan are supported by a British Heart Foundation fellowship; Dr Broberg, by the Waring Trust; Dr Dimopoulos, by the European Society of Cardiology; and Dr Johansson, by the Swedish Heart and Lung Foundation. The Royal Brompton Adult Congenital Heart Program received support from the Brompton Clinical Research Committee and the British Heart Foundation. We thank the exercise laboratory staff for their ongoing support.
Brochu MC, Baril JF, Dore A, Juneau M, De Guise P, Mercier LA. Improvement in exercise capacity in asymptomatic and mildly symptomatic adults after atrial septal defect percutaneous closure. Circulation. 2002; 106: 1821–1826.
Cohn JN, Johnson GR, Shabetai R, Loeb H, Tristani F, Rector T, Smith R, Fletcher R. Ejection fraction, peak exercise oxygen consumption, cardiothoracic ratio, ventricular arrhythmias, and plasma norepinephrine as determinants of prognosis in heart failure: the V-HeFT VA Cooperative Studies Group. Circulation. 1993; 87 (suppl V): V-I5–V-16.
Fredriksen PM, Chen A, Veldtman G, Hechter S, Therrien J, Webb G. Exercise capacity in adult patients with congenitally corrected transposition of the great arteries. Heart. 2001; 85: 191–195.
Task Force on Diagnosis and Treatment of Pulmonary Arterial Hypertension of the European Society of Cardiology. Guidelines on diagnosis and treatment of pulmonary arterial hypertension. Eur Heart J. 2004; 24: 2243–2278.
Francis DP, Shamim W, Davies LC, Piepoli MF, Ponikowski P, Anker SD, Coats AJ. Cardiopulmonary exercise testing for prognosis in chronic heart failure. Eur Heart J. 2000; 21: 154–161.
Gatzoulis MA, Clark AL, Cullen S, Claus GH, Redington AN. Right ventricular diastolic function 15 to 35 years after repair of tetralogy of Fallot. Circulation. 1995; 91: 1775–1781.
Bolger AP, Sharma R, Li W, Leenarts M, Kalra PR, Kemp M, Coats AJ, Anker SD. Gatzoulis MA. Neurohormonal activation and the chronic heart failure syndrome in adults with congenital heart disease. Circulation. 2002; 106: 92–99.
Fredriksen PM, Therrien J, Veldtman G, Warsi MA, Liu P, Siu S, Williams W, Granton J, Webb G. Lung function and aerobic capacity in adult patients following modified Fontan procedure. Heart. 2001; 85: 295–299.
Kaemmerer H, Fratz S, Bauer U, Oechslin E, Brodherr-Heberlein S, Zrenner B, Turina J, Jenni R, Lange PE, Hess J. Emergency hospital admissions and three-year survival of adults with and without cardiovascular surgery for congenital cardiac disease. J Thorac Cardiovasc Surg. 2003; 126: 1048–1052.
Lipkin DP. The role of exercise testing in chronic heart failure. Br Heart J. 1987; 58: 559–566.