Background

Acute coronary syndrome (ACS) is a frequent pathology worldwide. Management consists in coronary reperfusion, by percutaneous intervention or fibrinolysis. The worst evolution consists in ventricular fibrillation and cardiac arrest. Survival depends on the total ischemic time: delay between chest pain onset and reperfusion [1]. The main symptom is chest pain or discomfort [2]. The first link for out-of-hospital chain of survival is patient education to call the emergency medical communication centre (EMCC) in case of chest pain. The second is the ability of the dispatcher, medical or not, to identify patients with ACS and to send an ambulance (with an emergency physician or not, depending on the country).

In a Copenhagen EMCC, chest pain was the second identified reason after minor trauma for calls, with 11% of the requests [3]. The sex ratio was 1:1 and over half of the patients were over 65 years of age [4]. When a contact occurred with an EMCC, the prevalence at 30 days of ACS ranged from 12 to 16% [5, 6]. For patients with ACS, the use of the EMCC was in the range of 23 to 43% [7, 8]. Patients with ST elevation myocardial infarction (STEMI) were given the highest priority in 82% of cases [9].

ACS diagnosis depends on electrocardiography (ECG) findings and biomarkers. At an EMCC, these data are not available. Only patient history and characteristics of chest pain can be investigated. A persistent pain or a typical location with radiation without associated symptoms influences the dispatcher to send a Mobile Intensive Care Unit (MICU) in France [10]. In Sweden, the intensity, the localisation of the pain and a history of ischemic heart disease were associated with the final diagnosis of ACS [11]. The accuracy of a computer-based decision support was compared with dispatchers’ decisions to predict ACS [5]. Sensitivity was greater with the computer- than dispatcher-based decision (90% vs. 83%), and that under triage (false negative) was 10 and 17%, respectively. The factors correlated to under triage were the time of the call (lunch time) and the level of medical knowledge of the dispatcher (assistant nurse versus dispatcher with no medical training) [12]. These findings confirm the need for a decision support tool to help dispatchers identify patients at risk of ACS.

Atypical clinical presentations are difficult to diagnose [13]. In older and diabetic patients, chest discomfort can be absent [14, 15]. Males and females also differ in clinical presentation of ACS [2]. Women, particularly those < 55 years old, most often describe atypical chest pain, such as discomfort, pinching, or burning, [16,17,18,19,20]. These differences in initial presentation lead to an increase in mortality among women (10% versus 5%) [21]. Total ischemic time is longer in women than men [8, 9, 22]. One explanation is a lower prioritization for women when calling call centres (79% for women and 89% for men) [9]. These results suggest that attention should be paid to recognize these patients as soon as possible. In creating a “by-phone” predictive score of ACS, items should differ according to sex.

We aimed to establish and validate a model to predict ACS for men and women calling an EMCC from information that can be recorded by phone.

Methods

Study design and setting

The DOREMI 2 prospective cohort study was conducted in three French university hospitals. This study is a follow-up to DOREMI 1, which was a pilot and feasibility study. The three participating EMCCs were located in Toulouse, Bordeaux and Saint-Denis de la Réunion. In 2017, the EMCCs served 1.318, 1.506 and 0.843 million inhabitants, respectively. After an evaluation by a dispatcher assistant, a physician manages every call for medical reasons. Depending on the dispatcher prioritization, calls are handled by a general practitioner or an emergency physician. In France, calls from patients with chest pain or discomfort are generally transferred to the emergency physician dispatcher. In response to a call, a medical dispatcher can give medical advice, recommend going into a medical care structure or send an ambulance, fire brigade, physician or MICU (ambulance with an emergency physician on board).

Selection of participants

From May 2010 to November 2011, we consecutively included adults at 18 years of age or older who called the EMCC for non-traumatic chest pain. The “non-traumatic” characteristic was verified directly by asking the patient. The exclusion criterion was any difficulty in communicating: uncommunicative patient, language barrier, or inability to speak with the patient.

Measurements

At the first call to the EMCC, the emergency physician recorded patient characteristics, cardiovascular risk factors, medical history and clinical presentation on a standardized form (Additional file 1). Follow-up data were collected 30 days after the call (D30) by telephone interview. A research assistant contacted the patient’s general practitioner and/or the patient directly. The patient was contacted in case of non-response from the general practitioner or in case of incomplete information.

At the D30 follow-up, the research assistant retrieved reports from the emergency department, hospitalization and additional examinations. They collected data on major adverse cardiovascular events (rehospitalisation, myocardial infarction, urgent revascularization or death), admission or a consultation in a cardiology unit, non-invasive imaging (transthoracic echocardiography, stress imaging with exercise or drug, cardiac MRI), coronary angiography and the final diagnosis during hospitalization. The final diagnosis of ACS was based on these data.

Outcome

The outcome was a diagnosis of ACS by experts according to current guidelines [23].

STEMI was defined by the onset of a persistent ST-elevation on ECG, considered suggestive in the following cases: 1) at least two continuous leads with ST-segment elevation > 0.2 mV in leads V1-V3 or > 0.1 mV in leads V4-V9, V3R and V4R or 2) left bundle branch block with the presence of concordant ST-segment elevation.

Non–STEMI was diagnosed with compatible clinical presentation and ECG abnormalities in two continuous leads such as ST-segment depression or T-wave changes and elevated cardiac troponin level higher than the 99th percentile.

Unstable angina was considered when the patient had a compatible clinical presentation and ECG abnormalities without elevated cardiac troponin level and at least one of the following abnormalities: 1) dynamic changes of the ST-segment within 30 days or during the stress test; 2) a positive test result from stress echocardiography, cardiac MRI, coronary CT angiography; 3) coronary angiography with > 70% occlusion, 4) death within 30 days and, 5) rehospitalisation within 30 days with a diagnosis of ACS.

All patient files were retrospectively analysed by two experts to determine the final diagnosis of ACS or not. Three pairs of experts were recruited from the three centres. They did not belong to the team that included or cared for the patient. Files were randomly assigned. In case of discordance between the two experts, a third one was consulted. The diagnosis was based on pre-hospital data, reports of emergency departments, hospitalization and/or additional examinations, and follow-up on D30.

Analysis

Sample size calculation

At least 10 events per independent variable are recommended to ensure satisfactory statistical power in multivariate regression models [24, 25]. Because we planned to include a maximum of 15 independent variables in the final predictive model for each sex, we needed 150 calls with ACS for men and 150 for women. Given that approximately 16% of calls for non-traumatic chest pain have a definite diagnosis of ACS in the French EMCC, we needed 938 calls for each sex [6]. Given that the percentage of lost to follow-up is estimated to reach 20% (undetermined diagnosis), we needed to include 1123 calls for non-traumatic chest pain for each sex. This sample size estimation concerns the training (derivation) dataset, which corresponds to two-thirds of the overall included patients. Thus, we needed to include a total of 1705 men and 1705 women for the derivation and validation datasets. Owing to an expected sex ratio of women/men of 40%/60%, we needed to include 4263 consecutive calls for non-traumatic chest pain.

Analysis

Data are expressed as numbers with percentages for categorial variables or means with standard deviation or medians with interquartile range [IQR] for continuous variables. Categorial data were compared by chi-square or Fisher exact test when appropriate and continuous data by Student t or Mann and Whitney test as appropriate. Inter-expert agreement for the final diagnosis was calculated with the Kappa coefficient and its 95% confidence interval.

Model development

For each sex, we randomly selected a training (derivation) dataset from two-thirds of the data. Potential predictors were identified as variables associated with an ACS diagnosis significant at p < 0.2 on bivariate analysis or already known to be associated in the literature. We used a backward stepwise logistic regression to retain the final independent predictive variables, based on both p < 0.05 and the log-likelihood test. Then, we built a receiver operating characteristic (ROC) curve for each sex and defined the area under the ROC curve (AUC).

Model validation

The main characteristics of the derivation and validation models were compared for each sex by using appropriate bivariate statistical tests. The internal validity of the predictive model for each sex was tested in the validation dataset. First the discriminative performance of the score was evaluated with the validation dataset. Then, mean predictive probabilities were plotted against observed proportions of ACS in each quintile of predictive probabilities. Differences between observed and predicted probabilities were tested with the Hosmer-Lemeshow test.

All tests were two-sided, with statistical significance set at p < 0.05. All analyses were performed with STATA v11.2 (StataCorpLP, College Station, TX, USA) and CART software (Salford System, CA 92126 USA).

Results

Characteristics of study subjects

Over the 18 months of the study, 4205 patients were enrolled. A final diagnosis was established for 3727 (89%) (1630 [44%] women and 2097 [56%] men). Flowcharts of the final diagnosis by sex are presented in Fig. 1. Sensitivity analyses are presented in Appendix. Overall, 647 (17%) participants had an ACS diagnosis (508 [24%] men and 139 [9%] women), including 260 (7%) with STEMI. The inter-expert agreement was excellent (Kappa = 0.91 [0.89–0.93]) for diagnosing ACS.

Fig. 1
figure 1

Flowchart based on final diagnosis for men (a) and women (b). ACS: Acute Coronary Syndrome; NSTEMI: Non ST Elevation Myocardial Infarction; STEMI: ST Elevation Myocardial Infarction

Full size image

Main results

Males with ACS

The derivation dataset consisted of 1398 men. In total, 324 (23%) men had an ACS diagnosis, 135/324 (42%) with STEMI. The flowchart of the derivation dataset is in the Additional file 2. The general characteristics and type of pain for males are in Table 1. Men with an ACS diagnosis were older, more frequently had cardiovascular risk factors (except family history of coronary diseases and tobacco use) and had more typical pain typography characteristics than those without a diagnosis (Table 1).

Table 1 Characteristics and type of pain for male patients in the derivation set with and without a diagnosis of acute coronary syndrome (ACS)
Full size table

Eight factors mostly contributed to the final model for predicting ACS in males: age, tobacco use, severe and permanent pain; retrosternal, breathing non-related and radiating pain; and additional symptoms (Table 2). The AUC value for the final male model was 0.76 (95% CI 0.73–0.79), with no differences between observed and predicted probabilities (Hosmer-Lemeshow test: p = 0.78).

Table 2 Final model for predicting ACS in males after multivariate analysis
Full size table

General characteristics were well balanced between the derivation dataset (n = 1398) and the validation dataset (n = 699). The accuracy of the male model to predict ACS was validated, with no differences between observed and predicted probabilities (Hosmer-Lemeshow test: p = 0.554, Fig. 2a). The AUC value for the male prediction score was 0.76 (0.73–0.80).

Fig. 2
figure 2

Proportions of acute coronary syndrome cases observed and predicted for males (a) and females (b)

Full size image

Females with ACS

The derivation dataset consisted of 1087 women. Overall, 92 (8%) women had an ACS diagnosis, 32/92 (35%) with STEMI. The flowchart of the derivation dataset is in the Additional file 2. The general characteristics and type of pain for females are in Table 3. Women with an ACS diagnosis were older, with more cardiovascular risk factors (except family history of coronary diseases and tobacco use) and had more typical pain typography than those without a diagnosis (Table 3).

Table 3 Characteristics and type of pain for female patients in the derivation set with and without a diagnosis of ACS
Full size table

Four factors mostly contributed to the final model for predicting ACS in females: age ≥ 60 years, personal history of coronary artery disease, and breathing non-related and radiating pain (Table 4). The AUC value for the final female model was 0.79 (95% CI: 0.75–0.83), with no differences between observed and predicted probabilities (Hosmer-Lemeshow test: p = 0.70).

Table 4 Final model for predicting ACS in males after multivariate analysis
Full size table

General characteristics were well balanced between the derivation dataset (n = 1087) and the validation dataset (n = 543). The female model’s accuracy to predict ACS was not validated: predicted probabilities significantly differed from observed values (Hosmer-Lemeshow test: p = 0.035, Fig. 2b). The AUC value for the female prediction score was 0.67 (0.60–0.74).

Discussion

Main results

For adults calling an EMCC with chest pain or discomfort, predictors of a final ACS diagnosis differed by sex. The discriminative performance of the model was poor for women and good for men.

Explanation of the findings

In our study, a predictive variable for ACS in males agreed with traditional typical angina. The pain characteristics were so typical that even the coexisting conditions, such previous coronary artery disease, did not significantly add to the prediction in the multivariate model. Therefore, decision-making in men is based on the characteristics of pain. For females, except for age and personal history of ACS, factors were not related to typical angina. Thus, decision-making in women is mainly based on criteria other than the pain characteristics. The initial presentation of ACS is well known to differ by sex. In contrast to men, women complain of discomfort or pain due to pinching or burning [16,17,18,19,20]. These discrepancies could be due to disparities in pathophysiology and aetiologies [26, 27]. Women with an ACS diagnosis were more likely to have a normal or mild angiographic coronary heart disease [19, 21]. One-tenth of ACS cases involve spontaneous coronary artery dissection, mainly in young women [28]. Myocardial infarction with a non-obstructive coronary artery (MINOCA) occurs mostly in women and includes coronary endothelial dysfunction, myocarditis or Takostubo syndrome [1]. In these pathologies, traditional cardiovascular risk factors have a low implication.

The identification of a woman presenting an ACS remains a challenge. The mortality rate is higher in women than men because of their atypical initial presentation, older age, and less recourse to coronary angiography [21, 29, 30]. When adjusted on the same level of care, mortality is similar between the sexes, which led the authors of one study to advocate for a diagnosis of ACS in women [31, 32]. In our study, the discriminative performance of our model to predict ACS in women was not reproducible. Determinants were limited in number, not typical and above all not reproducible. An explanation is a potential lack of knowledge regarding variables to investigate to detect ACS in women. Currently, women are assessed with the variables used for men. Specific factors in women need to be investigated.

Strengths

Our study is the first prospective multicentric study to focus on predicting an ACS diagnosis in patients calling an EMCC for chest pain or discomfort. Other studies analysed patients with an established ACS diagnosis and cared for in an emergency department or cardiology department. Furthermore, this large study improves on the small number of studies evaluating the effectiveness of EMCC [33].

Limitations

We excluded uncommunicative patients because they could experience impending death, an at-risk sign. This state has already been highlighted in the evaluation of imminent delivery at an EMCC [34]. Thus, we investigated ACS as the only outcome without considering other life threatening causes of chest pain. The outcome was established retrospectively by experts, witch could be consider as subjective and potentially biased. Nevertheless, decision was based on medical records that were collected prospectively, limiting this bias. Lastly, in women we failed to propose a model that accurately predicted an ACS diagnosis in the validation sample.

Conclusions

A sex disparity exists in screening for ACS in people calling an EMCC because of chest pain. A score could be proposed for men. For women, a better understanding of pathophysiology and symptomatology are needed to increase the detection of ACS.