In this study, exertional chest pain, a pathologic ST-T segment on resting ECG, and angina diagnosis according to the patient’s opinion were independent predictors of exercise-test outcome. The observed OR for having any cardiovascular event within six months was 54.6 among patients with a positive exercise test and 11.3 among those with a non-conclusive test, compared to negative exercise tests. Patients with a negative exercise test had a very low risk of cardiovascular events. No patients predicted to have a negative exercise test by the multivariable model (>90% probability) were subject to hospitalization or death from coronary disease within six months.
Forty two percent of all patients with negative exercise tests reported chest pain on exertion (Table
2). Prevalent conditions among these patients were chronic pulmonary diseases, a clinical diagnosis of angina, anxiety disorder, non-specific chest pain, chest-wall pain, and upper gastrointestinal disorders. The diagnoses were not validated, but they offered an impression of the possible conditions associated with exertional chest pain that might result in a negative exercise test.
Comparisons with previous research
The low incidence of cardiovascular events found in patients with a negative exercise test in the present study is comparable to findings by Sumanen et al.
[35, 36]; they found that the 2-year incidence of IHD in patients with negative exercise tests was 2% among middle-aged patients and 3% in older patients.
The review by Mant and colleagues stated that the exercise test is a relatively weak diagnostic test (LR + of 2.79 and likelihood ratio for negative test of 0.44 for 1 mm ST depression), and that results should not be interpreted isolated from the patient’s clinical history
. Our approach, which used exercise-test outcome as a proxy for coronary disease, is supported by the marked differences in the cardiovascular event rates experienced by our patients. A bivariate approach that compares positive and non-conclusive tests in one group with negative tests has similarities in clinical decision-making. Non-conclusive tests carry positive information in one of two possible respects, chest pain or ST-segment depression during exercise, which indicate coronary disease. Both positive and non-conclusive exercise tests need to be considered for clinical management.
The diagnostic importance of exertional angina is well established previously
[5, 25, 38–42]. Exertional chest pain as an independent predictor in this study is therefore not surprising. The current investigation also considered the patient’s self-assessment of angina. A similar question was previously used in an investigation of the Marburg Heart Score
, which expressed this factor as “patient assumes pain is of cardiac origin.” Neither age nor previous cardiovascular events were independent predictors in our study; this finding contrasted with the investigation based on the Marburg Heart Score. This might be explained by the different approaches used in these two studies. First, we recruited patients that had been referred to exercise testing due to a suspected IHD, not due to explicit chest pain symptoms. Thus, even patients with atypical symptoms were included in our cohort. Second, in the Marburg score analysis, age was dichotomized; in contrast, in our analysis, age was a continuous variable. Third, the resting ECG outcome was included in our model, but not in the Marburg score. The Marburg score has three modalities of chest pain: “Pain worse during exercise”, “pain not reproduced by palpation” and “patient assumes pain is of cardiac origin.” We did not include “pain upon palpation.” Our approach lacked an external expert panel as a reference standard; this approach simulated the primary care scenario. These methodological differences may have contributed to the differences between our results and those of the Marburg score analysis.
Our questionnaire required the patient to describe angina symptoms as “yes” or “no.” In the total adjusted model, the patient’s opinion was a significant predictor. In the regression analysis of men and women separately, the patient’s opinion was not a significant predictor in men (OR 1.48, 95% CI 0.83-2.63), but it was significant in women (OR 1.86, 95% CI 1.04-3.34). This observation requires explanation, but the impact of a smaller sample size is highlighted by widened confidence intervals.
The predictive importance of the resting ECG was addressed by Pryor et al. in 1993
 in a study of patients referred for a treadmill test. Q waves and ST-T segment changes predicted significant coronary artery disease on angiography and survival within three years in a multivariable regression model. A systematic review by Mant et al. in 2004
 identified 13 relevant studies of the use of resting ECGs in the diagnosis of IHD in primary-care patients with suspected exertional angina. The review concluded that a resting ECG was only of limited value for evaluating patients with suspected angina. ST- and T-wave changes were not useful, and neither was a normal resting ECG. Q waves were the most frequently reported ECG change, within a wide confidence interval (LR+, OR 2.56, 95% CI 0.89-7.30)
. In our study, a pathologic ST-T segment on resting ECG was a significant predictor of exercise-test outcome in the total model and for women and men separately. Q waves on resting ECG were not helpful predictors in our study. Comparing our findings with previous reports on resting ECG as a diagnostic tool is not straightforward; in our study, we predict the outcome of exercise tests, not IHD proven by angiography. We believe that the potential utility of a resting ECG as a predictor of coronary disease is dependent on which type of reference standard is chosen.
Cardiovascular events and predictive model
A delayed-type reference standard has been used in several previous studies of chest pain and IHD in primary care
[8, 10, 24, 26, 41, 44] because validation by coronary angiography is not justified in a setting with a low disease prevalence. Use of the clinical course as a reference standard is achievable in primary-care practice and is not subject to bias, as long as follow-up is complete for all patients. We found that patients with a negative exercise test had a low cardiovascular risk within 180 days. Of 35 patients with revascularization, 31 were recruited from the 197 patients with a positive or non-conclusive test and only four patients were derived from the population of 653 patients with a negative exercise test.
Among patients classified as negative by the multivariable regression model (104/865), no cases with hospitalization for coronary disease or cardiovascular death occurred within six months. The probability cut off of >10% for an exercise test to be classified as positive/non-conclusive was chosen from a clinical standpoint. A prediction model designed to rule out disease must be evaluated against the number of false-negative cases to be clinically useful.
Strengths and limitations
Enrolment occurred in a clinical setting representing the normal care of patients who consulted GPs for suspected angina symptoms. Registration of cardiovascular events was completed without losses to follow up. Background variables contained few missing data and all exercise tests were performed by the same laboratory. The cardiovascular events during the follow-up were recorded from medical records, scrutinized by GN (author) and one assistant. A blinding to the outcome of exercise tests was not possible, because the records were complete, with no pre-selection of data.
We did not use coronary angiography for verification of disease, which is often standard. In primary care with predominantly low-risk patients, coronary angiography would not be justified in all cases, and we sought to study all patients referred to exercise tests for a suspicion of IHD. Registration of previous cardiovascular morbidity from a questionnaire is not faultless, but neither are medical record data. The use of self-reported history of IHD was supported by previous research
. Medication was registered from questionnaires, since medication lists sometimes are not up to date and patient compliance toward a medication list is not always complete. Physicians conducting the ECG readings were not blinded to the patient; blinded reading would be preferable in a study that specifically aims to explore the potential of resting ECG to predict exercise-test outcome.
Nearly half of the population was living in rural municipalities, and the average educational level was low. The access to primary health care was equivalent among the rural municipalities, compared to the central municipality of Östersund. Long distances to the central hospital might have been a problem for patients living in remote areas. Our results should be evaluated in relation to the availability of health care and baseline demographics in other populations.
For internal validation, we applied a bootstrapping procedure. An alternative approach would have been to divide the cohort in two to create a derivation cohort and a validation cohort. This was considered, but not applied, because it would have resulted in a loss of statistical power compared to the bootstrapping procedure. Due to this limitation, the clinical characteristics and the multivariate model described in this study should be validated in an external primary-care cohort.
We developed a clinical prediction model to characterize patients before exercise testing. We consider our results to be representative for patients of similar backgrounds living in countries with similar access to health care.