Methods

The ESCAPE trial was a cluster-randomised trial in which 14 hospitals were randomly allocated to either establish CPU care or continue with routine care.8 The seven hospitals randomised to intervention all successfully established CPU care and collected data from patients managed by the CPU over 1 year. The structures, processes and the outcomes of the CPUs have been described previously.9 Essentially, they were all based on a common protocol in which low-risk patients with acute chest pain were investigated by a combination of creatine kinase MB(mass) (CK-MB(mass)) and troponin testing, followed by treadmill exercise testing for selected patients. Those with positive findings for ACS were admitted to hospital, whereas those with negative tests were discharged home.

Patients attending the emergency department with acute chest pain of possible cardiac ischaemic origin were eligible for CPU care unless they met the following exclusion criteria: (1) ECG changes diagnostic for ACS (>1 mm ST segment deviation or >3 mm T wave inversion), (2) known coronary heart disease with recurrent or prolonged episodes of their typical anginal pain, (3) suspected or proven life-threatening non-cardiac pathology (eg, pulmonary embolus or aortic dissection and (4) comorbidities requiring inpatient care (eg, arrhythmia or heart failure).

The CPU protocol consisted of CK-MB(mass) measurement at baseline and at least 2 h and troponin measurement at least 6 h after the last significant episode of pain. Patients presenting 12 h after their worst pain were tested with a single sample for troponin. Following this, patients underwent treadmill exercise testing, either immediately or on the next working day, unless they were unable to perform a treadmill test or had recently had provocative cardiac testing or cardiac imaging. Previous studies have shown that these tests identify ACS reliably in isolation and in combination.10 11 12 13 14 15

Doctors or specialist chest pain nurses collected data on a standardised form from all patients managed according to the CPU protocol. The data collection form recorded presenting clinical characteristics, risk factors for coronary heart disease and the results of CPU diagnostic testing. A clinical researcher (JA) working with the ESCAPE trial reviewed all data collection forms, checked data accuracy and obtained missing data, where possible, by reference to the inpatient notes. The primary purpose of data collection was to audit each CPU. Anonymised data from the form were entered on to an Excel spreadsheet by JA, and summaries of these data were reported to the Trial Data Monitoring Committee.

We used the anonymised data to undertake this secondary analysis. Cases were excluded if they did not have a troponin measurement during their CPU assessment. This occurred most commonly because they were admitted to the hospital with further symptoms or complications, although some patients self-discharged before completing assessment. We used the following criteria to define the reference standard diagnosis of ACS: (1) troponin concentration >0.03 ng/ml, (2) CK-MB(mass) gradient of >3.0 ng/ml or (3) early positive treadmill exercise test. The treadmill test was performed by either cardiac technicians or chest pain nurses and interpreted by a single doctor interpreter who was not blind to patients’ presenting characteristics. The test was considered positive if there was >1 mm ST elevation or depression in any two contiguous leads. Clinical characteristics were recorded prior to diagnostic testing and thus blind to the criteria for ACS diagnosis.

Data were analysed using SPSS for Windows V.15. Logistic regression was used to test univariate associations between each categorical variable and ACS. t Tests were used to test univariate associations between each continuous variable and ACS. We then entered each variable that showed a univariate association with ACS (p<0.1) into a multivariate logistic regression model. The diagnostic value of each variable was then assessed by calculating likelihood ratios for categorical variables, using Confidence Interval Analysis software and constructing a receiver operator characteristic curve for continuous variables, using SPSS for Windows V.15.

Results

The ESCAPE intervention hospitals collected data from 1644 patients. We excluded 68 cases that did not receive a troponin measurement during their CPU assessment and therefore analysed data from 1576 patients (56.8% men, mean age 53 years (SD 14, range 17–98 years)). Overall, 132 (8.4%) of 1576 were diagnosed as having ACS: 18 had both elevated troponin and positive CK- MB(mass) gradient, 32 had elevated troponin but negative CK-MB(mass) gradient, two had a positive CK-MB gradient but no troponin elevation and 27 patients had elevated troponin but insufficient CK-MB(mass) measurements to calculate a gradient. Thus, 79 of 1576 patients were diagnosed as having ACS on the basis of biochemical markers (5.0%).

Treadmill exercise tests were performed on 1034/1576 (65.6%). The reasons for not performing a treadmill test were: inability to perform (108), contraindication to treadmill testing (56), recent provocative testing or cardiac imaging (70), other reasons (122) and no reason recorded (186). Among those tested, 55/1034 (5.3%) had an early positive treadmill exercise test. Two of these had undergone treadmill testing despite having a small troponin elevation recorded. Thus a total of 132/1576 patients were diagnosed as having ACS (8.4%).

The proportion of the cohort with each categorical feature and the association between each categorical feature and ACS are shown in table 2. The associations between continuous variables (age and pain duration) and ACS are shown in table 3. Patients with ACS tended to be older and have a longer duration of pain; were more likely to be a man, hypertensive or an ex-smoker and more likely to have pain radiating to their right arm. All other features were not significantly associated with ACS.

The six variables with univariate association with ACS were entered into a multivariate model. The results are shown in table 4. Older age, male sex, longer duration of pain and radiation of pain to the right arm were independent predictors of ACS. Hypertension and previous smoking were not associated with ACS on multivariate analysis.

The area under the receiver operator characteristic curve for age was 0.629 (95% CI 0.573 to 0.686) and for duration was 0.546 (95% CI 0.481 to 0.610). Sensitivity, specificity and likelihood ratios for all variables are reported in table 5.

Discussion

This study reports one of the largest cohorts evaluating the diagnostic value of clinical features for ACS. The findings show that in patients with a normal or non-diagnostic ECG, no serious comorbidity and no obvious alternative cause for their symptoms, there are few diagnostically useful clinical features. Age, sex, duration of symptoms and radiation of pain to the right arm were the only independent predictors of ACS. Of these, arguably only radiation of pain to the right arm, with a likelihood ratio of 2.9, provides diagnostic information that might influence clinical decision making.

Systematic reviews of previous studies have identified a number of diagnostically useful clinical features, as outlined in table 1.2 5 This study confirms that radiation of pain to the right arm appears to be the most powerful indicator of ACS. This contrasts with traditional teaching that radiation of pain to the left arm is typical of cardiac pain. One possible explanation is that patients are aware that radiation of pain to the left arm suggests cardiac pain and so attend the emergency department with any pain radiating to the left arm but attend with pain radiating to the right arm only if it is exceptionally severe or unexplained by other factors.

Other features previously shown to be diagnostically useful were not supported by our study. Nausea, vomiting, sweating, radiation of pain to the left arm5 and burning-type7 pain have previously been shown to predict ACS, while stabbing pain and pain reproduced by palpation have been shown to reduce the likelihood of ACS.5 The odds ratios for these features in our analysis showed weak trends in the same direction as those in previous studies, but these were not significant on univariate analysis. This suggests that our study may not have had sufficient power to detect an association for these variables, and/or the associations between these variables and ACS are weaker in a cohort of patients with normal or non-diagnostic ECG.

This study has a number of limitations that should be taken into account when interpreting the findings. These limitations principally relate to the fact that this study was a secondary analysis of data collected for another purpose. There was no sample size calculation, and the study was not powered to detect potentially important associations between clinical features and ACS. Data collection had to be feasible for the clinical staff responsible for undertaking this role, so we did not examine all potentially useful clinical features comprehensively. For the same reason, we were unable to examine interobserver error. Furthermore, although the reference standard was pragmatic and reasonably robust, it is possible that some cases with and without ACS may have been misclassified. In particular, the choice of threshold for CK-MB(mass) gradient was somewhat arbitrary, reflecting a compromise between the value of this measure in early detection of ACS and the risk of a false-positive CK- MB(mass) increase, the use of a 6-hour protocol could have missed some patients with a small, late troponin increase, and patients with a positive treadmill test may have had significant coronary artery disease but not necessarily ACS.

Finally, it should be recognised that although the study population was reasonably well defined and relevant to clinical practice, it was selected from the wider population with chest pain on the basis of diagnostic uncertainty. Patients with clearly non-cardiac chest pain, such as musculoskeletal chest pain or anxiety-related chest pain, may not have been included in the study cohort, so our findings may not apply to patients with a clear non-cardiac diagnosis. Conversely, our findings should not be generalised to higher risk patients, such as those with known coronary heart disease.

The implication of this study is that the individual clinical features that we examined have very limited value in diagnosing ACS in patients with a normal or non-diagnostic ECG and cannot be used reliably to select patients for diagnostic testing or hospital admission. The decision regarding whether to use diagnostic testing or hospital admission for these patients depends on weighing the costs, risks and benefits of the various alternatives. If testing or admission is considered worthwhile, then it should be applied to all patients with suspected ACS.