In this prospective study of patients with non-traumatic chest pain, we analyzed the diagnostic values of the overall clinical assessment of ACS likelihood, and the values of the main diagnostic modalities underlying this assessment, namely the chest pain history, the ECG and the initial troponin result. Our main findings were three: First, age < 40 years, chest pain history and overall gestalt not suspicious of ACS all practically ruled out ACS. Second, a positive initial TnT and an ischemic ECG were strong predictors of ACS and seemed superior to pain history for ruling in ACS. Third, in patients with a normal initial TnT and non-ischemic ECG, chest pain history typical of AMI was not a significant predictor of AMI while chest pain history typical of UA was a moderate predictor of UA.
The present study shows, not surprisingly, that the overall clinical gestalt was better than its components both at ruling in (“Obvious ACS”, LR 29) and at ruling out (“No Suspicion of ACS”, LR 0.01) ACS. None of the 441 patients with a “No Suspicion of ACS” gestalt had ACS within 30 days. In accordance with the results by Kline et al. 2014 (Kline and Stubblefield), a “No Suspicion” gestalt thus seems to rule out ACS in the ED, and to obviate the need for admission, serial troponins, and stress-testing for the exclusion of ACS. Miller et al. found that 2.8% of ED patients assessed as “noncardiac chest pain” had adverse cardiac events within 30 days (Miller et al. 2004). However, in that study TnT was not measured in all patients and the gestalt impression was recorded before biomarkers were drawn. About half of their patients with adverse cardiac events turned out to have elevated troponins meaning they would not have been classified in the “No Suspicion of ACS” group in our study. A “No Suspicion of ACS” gestalt and a non-suspicious chest pain history both almost excluded ACS, and one might speculate that the ED physicians based their no suspicion gestalt primarily on the chest pain history. Our results thus suggest that a non-suspicious pain history may in many cases be enough to rule out ACS in the ED, at least if the pretest probability is low. Only 0.2% of the patients with a non-suspicious pain history had ACS within 30 days in our population. Further, the study confirms previous findings that age < 40 years argues strongly against ACS (Collin et al. 2011; Marsan et al. 2005), while older age has limited predictive value (Chun and McGee 2004).
We have found four previous publications with data on the diagnostic or prognostic value of the overall clinical gestalt in patients with possible ACS. However, two of the studies were from before the modern biomarker era and focused on AMI only (Karlson et al. 1991; Tierney et al. 1986), one was much smaller than the present study and included patients without chest pain and/or TnT tests (Ekelund et al. 2002), and one was primarily prognostic and included only low risk patients (Chandra et al. 2009). In the present study, the gestalt had largest predictive ability when cases were assessed as “Obvious ACS” or “No suspicion of ACS”. In the strong suspicion group, it appeared that the physicians’ gestalt overestimated the likelihood of ACS, since 60% of these patients did not have ACS. The value of a high grade of suspicion of ACS may thus be less than generally believed. In this context, Kline et al. 2014 reported that emergency physicians tend to overestimate the likelihood of ACS also in low risk patients (Kline and Stubblefield). In the low suspicion gestalt group 6% of the patients had ACS, which indicates that these patients should in general undergo further evaluation.
In accordance with previous findings, a negative initial TnT and a non-ischemic ECG did not reliably rule-out ACS (Chun and McGee 2004; Ebell et al. 2000; Fesmire et al. 1989). On the other hand, TnT and ECG seemed superior to chest pain history for ruling in ACS. Our findings thus support the practice of admitting all chest pain patients with ischemic ECGs and/or elevated TnT for additional evaluation. In patients with a non-ischemic ECG and a negative initial TnT, a pain history typical of AMI was poorly predictive of AMI (LR 1.6). In contrast, pain typical of UA was still a moderate predictor of UA (LR 4.7). The results thus indicate that patients with a pain history typical of UA should undergo further evaluation, regardless of the ECG and TnT results, which is probably true even if highly sensitive troponins are used (Borna et al. 2014).
Limitations of the study
This study was performed at only one university hospital and the results are not necessarily generalizable to other hospitals. However, the prevalence of ACS among chest pain patients was 12.7%, which is comparable to that in other studies of unselected ED chest pain patients (Han et al. 2007; Scheuermeyer et al. 2012).
The discharge diagnoses were those used in routine clinical care. Since we aimed to study diagnostic value in routine care, we did not assess the diagnoses for accuracy, and we have no data on what proportion were based on objective testing, e.g. stress tests or coronary angiography. However, at our institution which is the academic cardiac center for the entire region, most patients are evaluated with stress testing and virtually all patients with ACS undergo coronary angiography. All discharge diagnoses were reviewed for quality and accuracy by the attending specialist physician (most often cardiologist, in a few cases internal medicine specialist). In addition, the patients were followed for 30 days after the ED visit. The discharge diagnoses reflected real life practice, and we believe that very few were inaccurate.
In our review of the patient records at 30 days to determine whether an ACS diagnosis was missed or if the patient died, we may have missed a small number of patients presenting to other hospitals. However, such misclassifications were probably few and unlikely to significantly affect the results of this study.
In the analysis, patients with or without ongoing chest pain were not separated. We have no data as to whether they were evaluated or treated differently.
Suggested definitions of the different levels of ACS suspicion were present on the study forms, and although they left considerable room for judgment, other (or no) definitions may have led to somewhat different results. The definition of typical symptoms of MI might have been suboptimal as it is somewhat non-specific, but it is a definition commonly used in guidelines (Amsterdam et al. 2010). Although the physicians were instructed to disregard ECG and TnT when evaluating the symptoms, we cannot exclude that ECG and TnT results influenced the symptom assessment in some cases.
As TnT was used in the gestalt assessment as well as in deciding the final diagnosis, incorporation bias could have been present. This was however probably limited by the fact that the emergency physicians only had access to the initial TnT, whereas the discharge diagnoses were most often based on repeated TnT analyses to assess for significant rise or fall.
Finally we did not have data regarding physician level of experience. However, at least in the assessment of pulmonary embolism, differences in the diagnostic accuracy of the gestalt depending on experience are small (Kabrhel et al. 2005).
Suggestions for further studies
Many of our results have broad confidence intervals suggesting that a larger study with a similar aim would be preferable in order to confirm the findings.
Several clinical decision support tools and risk prediction scores for patients with suspected ACS have been published, e.g. the HEART score (Six et al. 2008). For any such tool or score to be clinically useful, they have to be at least as good as the gestalt. We suggest that future studies compare new decision support tools and scores with the physician’s gestalt assessment. Interestingly, it has been shown that the gestalt performs better than the Wells score in the assessment of the probability of pulmonary embolism (Penaloza et al. 2013).