Dyspnea, a high-risk symptom in patients suspected of myocardial infarction in the ambulance? A population-based follow-up study

Design and setting

This is a population-based follow-up study of 23,184 telemedical contacts registered in the Danish Tele-database from 1 June 2008 to 1 January 2013. The setting is the Central Denmark Region which covers an area of 13,053 km² of mixed urban, suburban and rural land with 1.3 million inhabitants corresponding to 23 % of the Danish population [13]. The EMS response in this region is two-tiered with basic-level care delivered by ambulances sent to all urgent cases. Anesthesiologist or nurse anesthetist staffed prehospital critical care teams deliver supplementary advanced prehospital care in suspected life-threatening cases. Acutely ill patients are brought to one of five public hospitals within the region. Revascularization therapy (percutaneous coronary intervention or surgery) is centralized at Aarhus University Hospital. Health care is public and provided free of charge by the Danish National Health Service.

Ambulance personnel establish an ECG-based telemedical contact to an on-call cardiologist at a local department of cardiology in patients suspected of an MI according to predefined criteria: 1) patients with on-going or recent chest pain within the past 12 h and/or clinical suspicion of MI (e.g. pallor and diaphoresis) 2) patients with new onset of dyspnea within the past 12 h and no known lung disease 3) patients with other symptoms raising suspicion of heart disease (e.g. syncope/somnolence, palpitations, abdominal/back pain, intoxication etc.) and 4) patients resuscitated from OHCA. In case of STEMI, OHCA with no obvious non-cardiac cause, or cardiogenic shock, the ambulance is re-routed directly to the pre-alerted catheterization laboratory at Aarhus University Hospital for coronary angiography and primary percutaneous coronary intervention. Patients not re-routed directly to the catheterization laboratory are generally admitted to the nearest hospital receiving acute patients. On-call cardiologists receiving the telemedical calls perform all registrations in the Danish Tele-database, including which of the above-mentioned symptoms/conditions is the predominant cause of MI suspicion in the individual patient.

Study population

The criterion for inclusion in the study was a registration of an ECG-based telemedical contact in the Tele-database at a department of cardiology in the Central Denmark Region with a valid personal identification number. Only the first contact during the study period in each patient was included. Participants were followed until 9 October 2013. Contacts with invalid personal identification number or foreigners were considered lost to follow up and were excluded from the study.

Data sources

Vital status was retrieved from the Danish Civil Registration System; and data on previous diseases and present diagnoses according to the 10^th version of the International Classification of Disease (ICD-10) were retrieved from the Danish National Registry of Patients. Full medical files were retrieved from a total of 200 patients with dyspnea for development of an algorithm for determining the cause of dyspnea in the ambulance and for validation of this algorithm.

The present study was exempt from ethical approval according to Danish law. The Danish Data Protection agency (Journal nr. 1-16-02-158-12) and the Danish Health and Medicines Authority (Journal nr. 3-3013-247/1/) approved the study.

Exposures and outcomes

The predominant symptoms/conditions leading to suspicion of MI as registered by the on-call cardiologists were regarded as exposures. The primary outcome was 30-day mortality. Secondary outcomes were long-term mortality (4 years), the number of patients in each group diagnosed with an MI and the associated mortality rates, and causes of dyspnea in patients triaged because of dyspnea. The following potential confounders were considered to be relevant: comorbidities, age, gender, smoking status, alcohol consumption, heart rate and systolic blood pressure. Because of an expected mixture of different underlying diseases causing the different symptoms, we opted on using a comorbidity index rather than adjusting for specific diseases. We chose the Charlson Comorbidity Index based on the ICD-10 diagnoses as described by Sundararajan et al. as this is extensively validated - also in a Danish population [14, 15]. Information on age, gender, heart rate and systolic blood pressure was obtained from the Tele-database. Vital signs were measured at first encounter – in patients resuscitated from cardiac arrest immediately following ROSC. We had no information on smoking status or alcohol consumption.

Patients diagnosed with MI were identified by ICD-10 codes ranging from DI21.0 to DI23.8A. By studying information from the full medical files from the first 100 of the 1,464 patients with dyspnea, we developed an algorithm to predict the cause of dyspnea (heart disease/lung disease/other cause) based on the ICD-10 codes assigned at discharge (Additional file 1 and 2). A new random sample of 100 patients from the 1464 patients with dyspnea was then drawn to test the algorithm. For these patients, full medical files including imaging and blood tests were retrieved and reviewed by an end-point adjudication committee consisting of two investigators blinded to the result of the ICD-10 algorithm. The members of the committee adjudicated the cause of dyspnea, dividing it into one of three main categories: 1) heart disease 2) lung disease and 3) other cause. In case of incongruence, a third investigator reviewed the material and chose one of the two previously adjudicated causes. Thus, the majority vote was accepted as consensus. It was decided a priori that the performance of the ICD-10 algorithm was satisfactory if in agreement with the adjudication committee in at least 75 % of cases. This decision was based on previously reported variations in sensitivity and specificity in which ICD-10 codes were used to evaluate whether a diagnosis of heart failure and chronic obstructive pulmonary disease was established [16, 17].

Analyses

Mortality was displayed as Kaplan Meier curves which were truncated when less than 10 % of the original cohort was left. To adjust for potential confounders, we planned to perform a Cox proportional hazards regression analysis. The survival curves did not fulfill the proportional hazards assumption, so this plan was waived. Instead we calculated adjusted risk differences by a generalized linear regression using pseudo-observations. The pseudo-observations approach is a transformation of data like log-transformation and others – although it is mathematically more complex. In contrast to Cox regression it allows for non-proportional survival curves and analysis of risk differences at specific time points. It has been well described by Klein et al. [18]. Analyses were conducted in STATA12 (StataCorp LP, Texas, USA) according to the method described by Parner and Andersen [19]. P-values <0.05 was considered to be significant and 95 % confidence intervals (CIs) were used. Participants were categorized into Charlson Comorbidity Index Groups according to the original classification by Charlson et al. [20]. Participants were categorized according to systolic blood pressure and heart rate into four groups approximated to quartiles. Missing data were expected to be missing completely at random and no imputation was made.

Cause of dyspnea

Two endpoint committee adjudicators agreed on the primary cause of dyspnea in all but four of 100 patients sampled for validation. In these patients, the third adjudicator was involved and the majority vote was accepted. The ICD-10 diagnosis algorithm was in agreement with the endpoint adjudication committee in 78 % of the 100 patients sampled for validation. The predominant underlying cause of dyspnea in this cohort of patients suspected of MI was heart disease, but only 8,3 % actually had an MI (Table 4). Among patients in whom heart disease was the primary cause of dyspnea, 20 % were also diagnosed with a lung disease. When lung disease was the primary cause for dyspnea, 41 % of patients were also diagnosed with a heart disease.

Sensitivity analyses

We performed three sensitivity analyses, all related to different ways of adjusting for age and comorbidity (included in Additional file 3). We accounted for a possible higher risk with increasing age in older age categories than in younger by including both age and age squared and for a possible polynomial development by including age as qubic splines with seven knots. This reduced the effect of dyspnea slightly but did not change the overall patterns. Including specific diseases related to chest pain and dyspnea rather than the Charlson Comorbidity Index in addition to using qubic splines to adjust for age did not change the results.

Dyspnea vs. other symptoms

Studies comparing mortality rates in patients with dyspnea and patients with other symptoms in a prehospital setting are scarce. However, a large study in emergency medical technician witnessed OHCA showed that patients suffering chest pain prior to cardiac arrest were five times more likely to survive than patients suffering dyspnea prior to cardiac arrest [21]. In a retrospective analysis of hospitalized patients with first-time MI, dyspnea was associated with higher long-term mortality than other presenting symptoms, including chest pain [22]. In patients undergoing coronary angiography for suspected acute coronary syndrome, a four-fold higher mortality in dyspneic compared to non-dyspneic patients was demonstrated [23]. Thus, the observation of higher mortality among dyspneic patients is consistent across the prehospital setting, the acute hospital setting, and the catheterization laboratory. Nevertheless, we still do not have broadly accepted guidelines for management of the dyspneic patient in the prehospital setting.

Diagnoses, comorbidities and age

Approximately 8 % of the patients suffering dyspnea had an MI. ECG acquisition is thus important also in prehospital diagnostics in dyspnea patients, but it seems inadequate to establish a prehospital diagnosis and identify high-risk patients in this patient group in general. These patients suffer from a variety of diseases not only including MI but also heart failure, pulmonary infections, and obstructive pulmonary disorders. Furthermore, our study also demonstrates that heart and lung disease frequently coexist. This is supported by other studies [11, 12], and this co-existence deteriorates the prognosis [10]. As it is seen from Table 1, age and comorbidities do play a major role in the high mortality in patients with dyspnea. At long term, the independent effect on mortality of being in CCI group ≥ 3 or being 10 years older is more pronounced than the effect of suffering dyspnea. At short term, the same applies to age, but not comorbidities. In addition, no matter how we tried to adjust for age and comorbidities, we were not able to eliminate an independent effect of dyspnea compared to chest pain. This indicates, that there is a potential for improving care of patients with dyspnea. Based on the very high mortality observed in the first short period of time after ambulance contact, there seems to be a window of opportunity in the acute phase (refer to the Kaplan-Meier curves in Fig. 2).

Towards a systematic approach

As of now, a stethoscope and an ECG is the most advanced diagnostics routinely used in dyspneic patients in the prehospital setting. Treatment is often broad rather than specific and the evidence behind different new interventions is scarce. We are currently witnessing several, promising technological developments that may aid prehospital decision-making and treatment in dyspneic patients. Point-of-care testing with biomarkers, like brain natriuretic peptides and focused ultrasonography, has been suggested [24–26]. Establishing an accurate diagnosis already in the ambulance to facilitate earlier correct therapy and triage high-risk patients to specialized departments or intensive care unit may improve outcome, but this has not been examined. New treatment regimes facilitating oxygenation and ventilation by continuous positive airway pressure or non-invasive ventilation has also been suggested, but has not been sufficiently examined in relation to patient outcome [27]. Validating the findings of this study in unselected patients with dyspnea in the prehospital setting and the systematic use of prehospital point-of-care diagnostics and novel treatments for this patient group are potential subjects for future research.

Strengths and limitations

A high number of consecutively registered patients and adjustment for major confounding factors including age and comorbidity are strengths of this study. Unlike most epidemiological studies, we were also able to adjust for prehospital values of systolic blood pressure and heart rate. An endpoint-adjudication committee validated the use of ICD-10 diagnoses to establish the cause of dyspnea. Owing to the predefined criteria for telemedical contact, only a fraction of all patients with dyspnea in the ambulance were telemedically triaged and registered. A selection bias that favors a higher proportion of patients with heart disease is evident. Dyspneic patients suspected of MI might be more severely affected than those not suspected of MI because some of them may have large anterior infarctions and non-cardiac acute conditions like pulmonary embolism and pneumothorax. This might have led to an overestimation of the mortality in patients with dyspnea compared with a general prehospital dyspnea population. As we did not have data on smoking status, alcohol consumption, left ventricular systolic function or pulmonary function, residual confounding might be present. However, adjusting for smoking status did not change the association between dyspnea and mortality in several other studies [7, 28]. Pulmonary disease and heart failure are included in the Charlson Comorbidity Index and adjustment was inherent for pulmonary function and left ventricular systolic function.