Importance of the First Link
Description and Recognition of an Out-of-Hospital Cardiac Arrest in an Emergency Call
Background— The content of emergency calls for suspected cardiac arrest is rarely analyzed. This study investigated the recognition of a cardiac arrest by dispatchers and its influence on survival rates.
Methods and Results— During 8 months, voice recordings of 14 800 consecutive emergency calls were collected to audit content and cardiac arrest recognition. The presence of cardiac arrest during the call was assessed from the ambulance crew report. Included calls were placed by laypersons on site and did not involve trauma. Prevalence of cardiac arrest was 3.0%. Of the 285 cardiac arrests, 82 (29%) were not recognized during the call, and 64 of 267 suspected calls (24%) were not cardiac arrest. We analyzed a random sample (n=506) of 9230 control calls. Three-month survival was 5% when a cardiac arrest was not recognized versus 14% when it was recognized (P=0.04). If the dispatcher did not recognize the cardiac arrest, the ambulance was dispatched a mean of 0.94 minute later (P<0.001) and arrived 1.40 minutes later on scene (P=0.01) compared with recognized calls. The main reason for not recognizing the cardiac arrest was not asking if the patient was breathing (42 of 82) and not asking to describe the type of breathing (16 of 82). Normal breathing was never mentioned in true cardiac arrest calls. A logistic regression model identified spontaneous trigger words like facial color that could contribute to cardiac arrest recognition (odds ratio, 7.8 to 9.7).
Conclusions— Not recognizing a cardiac arrest during emergency calls decreases survival. Spontaneous words that the caller uses to describe the patient may aid in faster and better recognition of a cardiac arrest.
Received April 24, 2008; accepted January 30, 2009.
The first link in the chain of survival,1 early call, includes the dispatchers’ recognition of a cardiac arrest. It takes a relatively long time for the emergency medical dispatcher to recognize a cardiac arrest. In Seattle, the average recognition time from the start of the call to the start of cardiopulmonary resuscitation (CPR) instructions is 75 seconds. Factors contributing to no or delayed recognition of cardiac arrest include vague description of breathing and agonal breathing, unnecessary questions, and little experience in call taking.2–5 As a result, dispatching may be unduly delayed, the wrong level of ambulance is sent, and no telephone CPR instructions are offered.6 Current guidelines call for CPR in unconscious patients when breathing is not normal.7 A similar rule is applied in dispatcher protocols to recognize a cardiac arrest for ambulance dispatch and telephone-instructed CPR. In 1986, Eisenberg et al8 suggested almost 100% sensitivity when those questions were asked to identify a cardiac arrest. However, this was never confirmed by other studies.
Editorial p 2023
Clinical Perspective p 2102
The purposes of our study were to investigate whether recognition of a cardiac arrest by the dispatcher influences long-term survival and to analyze which words used by the caller to describe the emergency indicate the presence of a cardiac arrest, with an emphasis on the description of the patient’s breathing.
At the time of the study, callers contacted a regional dispatch center in case of an emergency. The Dutch emergency medical system was 1 tiered. If dispatchers suspected a cardiac arrest, they sent 2 ambulances; otherwise, 1 ambulance was sent. If the dispatcher did not suspect a cardiac arrest at the time of the call but ambulance personnel on site found the patient in cardiac arrest, they requested a second ambulance. Ambulance dispatchers were mostly nurses or other healthcare workers who completed additional training in dispatching. The dispatchers followed a standard protocol when questioning the caller about the event. They first established the exact location of the incident and phone number of the caller because caller identification was not automatically displayed. Next, the protocol required dispatchers to ask the caller to describe the emergency, to indicate whether the patient was conscious, and to indicate whether the patient was breathing. Dispatchers were taught that agonal breathing was a sign of cardiac arrest. They suspected cardiac arrest by the caller’s description of the patient. Time of call and emergency medical system dispatch, departure, and arrival on scene were automatically generated in the dispatch computer.
Study Design and Data Collection
This prospective, observational study was carried out in the Greater Amsterdam Dispatch Unit, which serves the city of Amsterdam and a surrounding area covering a population of 1.3 million inhabitants. Between January 1, 2004, and September 1, 2004, all digitalized voice recordings of consecutive high-priority emergency calls were collected and audited. We included all calls placed by laypersons who were at the site of the emergency. We excluded calls placed by police officers, firefighters, or general practitioners on duty, as well as calls by the patients themselves. We further excluded calls related to trauma and calls about unequivocally conscious patients. Second calls were left out of the analysis. Patients who were declared dead by ambulance paramedics and those for whom advanced cardiac life support was not initiated were excluded. The Medical Ethics Committee of the Academic Medical Center in Amsterdam approved this study.
Recognition of a Cardiac Arrest by the Dispatcher
The dispatch protocol required the dispatcher to send out 2 ambulances if he or she suspected a cardiac arrest and 1 if not. Calls were classified as suspected cardiac arrest if 2 ambulances had been sent out during the first call. This information was retrieved from the automated records of the dispatch unit.
An ambulance dispatcher (F.B.) and a researcher (J.B.) reviewed all calls classified as suspected cardiac arrest to mark the time when the dispatcher first suspected the presence of cardiac arrest. The time of first suspicion was determined by mutual agreement of the assessors, who were blinded to all subsequent information about the emergency. In case of disagreement, a third assessor made the final decision.
Presence of Cardiac Arrest (Gold Standard)
The actual presence of cardiac arrest was assessed from the ambulance crew report. Assessment and treatment were done according to the 2000 guidelines.9
If the patient’s initial rhythm was asystolic and the delay between collapse and ambulance arrival was ≥15 minutes, the paramedics did not start advanced cardiac life support and the patient was excluded. If bystanders initiated basic life support but paramedics felt a pulse during their first check, the patient was considered not to be in cardiac arrest during the call. If a patient was found to be in cardiac arrest on arrival but the content of the call clearly involved other complaints (eg, chest pain), cardiac arrest was not considered to be present during the call.
Trigger Words Used by the Caller to Describe the Emergency
Calls were audited by an ambulance dispatcher (F.B.) and a researcher (J.B.) to denote the words that the caller used to describe the emergency. A list of words derived from an earlier pilot study was used in which the 67 most common trigger words and statements associated with cardiac arrest were identified (Table I of the online-only Data Supplement).
The auditor classified the patient’s breathing in each call into one of the following categories. First, the call was listed as “normal breathing” if the caller affirmed that the patient was breathing normally. Second, if the dispatcher did not ask the caller about the patient’s breathing and the caller gave no spontaneous description, the call was listed as “breathing not mentioned.” Third, if the dispatcher received a positive answer to the question “Is the patient breathing?” but the type of breathing was not given, the call was categorized as “breathing undefined.” Fourth, breathing was defined as “abnormal breathing” when the caller used phrases such as “occasional breathing,” “barely/hardly breathing,” “heavy breathing,” “labored or noisy breathing,” “sighing,” and “strange breathing.” Sometimes, callers mimicked the sound or put the phone to the mouth of the patient so that the dispatcher could evaluate the type of breathing. Fifth, if the caller stated that the patient was not breathing, the call was listed as “not breathing.” This scoring was applied to all calls in which a cardiac arrest was suspected and/or present. Furthermore, the scoring was applied to a 5.5% random sample of calls in which cardiac arrest was neither suspected nor present.
The survival status of the patients at 3 months was verified in the civic registry.
The time interval between call initiation and suspicion of the cardiac arrest was expressed in medians and 25% to 75% percentiles. The time interval between call initiation and dispatch was expressed in means and SD.
Baseline comparisons between cardiac arrest patients and non–cardiac arrest patients were analyzed by Student t test, the χ2 test, or the Mann–Whitney U test when appropriate. Time from call to cardiac arrest suspicion as a result of the breathing description was compared by the Kruskal-Wallis test, and the Mann–Whitney test was used for posthoc pairwise comparisons. Values of P<0.01 were considered significant. Survival was estimated with the Kaplan-Meier method and compared by use of the log-rank test.
The predictive value for cardiac arrest of the trigger words used in >10 calls was quantified by calculating the odds ratios. Because we analyzed only a random sample of the calls in which cardiac arrest was neither suspected nor present but analyzed all calls in which cardiac arrest was either suspected or present, we performed a 2-stage sampling analysis as described by Breslow and Chatterjee.10 The analyzed 5.5% of the calls in which cardiac arrest was neither suspected nor present were given a weight of 18.2 (100/5.5). We used logistic regression analysis implemented in the complex sampling module of SPSS version 15 (SPSS Inc, Chicago, Ill) for this analysis, and we performed backward stepwise selection to identify the most important trigger words.
The predictive performance of the logistic model was quantified with the area under the curve (AUC), plotting the predicted probability of a cardiac arrest against the observed probability. To obtain an unbiased estimate of the AUC, we randomly divided the data into a training set and a test set. The training set consisted of 75% of the observations; the test set consisted of 25%. We fitted the predictive model using the data in the training set and calculated AUC in the test set. We repeated this procedure 10 times. The mean AUC was calculated. Finally, the selected model was fitted on the total data set to assess its goodness of fit; the predicted risk of cardiac arrest was tested against the observed rate of cardiac arrest using the Hosmer-Lemeshow test and plotted in a graph.
All statistics were performed in SPSS (SPSS version 15.0 for Windows). Significance was accepted at 2-sided values of P<0.05.
The authors had full access to and take full responsibility for the integrity of the data. All authors have read and agree to the manuscript as written.
Figure 1 shows the steps in data handling and analysis. In 8 months, 14 800 high-priority emergency calls were received. Trauma was involved in 3384 calls; therefore, they were excluded. In 436 calls, 2 ambulances were eventually sent. All 436 voice recordings were audited for exclusion criteria and content, and 87 were excluded because the caller was a professional or the patient was conscious. In 203 of the remaining 349 calls, the dispatcher sent 2 ambulances and thus suspected a cardiac arrest, which the ambulance paramedics on site confirmed. In 64 calls, the paramedics did not confirm the suspected cardiac arrest. In the remaining 82 calls, the dispatcher sent only 1 ambulance and thus did not suspect a cardiac arrest. The paramedic on site requested a second ambulance because he or she diagnosed a cardiac arrest. In 10 980 calls, only 1 ambulance was sent, indicating that a cardiac arrest was neither suspected by the dispatcher nor diagnosed by the ambulance personnel on site. We audited a random 5.5% sample (n=602) for exclusion criteria. The remaining 506 calls were used for analysis.
The baseline characteristics of the patients are shown in Table 1. Patients in calls that concerned true cardiac arrest were older than patients in calls for other emergency conditions, with a significant trend over decades of age (P<0.0001). They were also more likely to be male, and their emergency occurred more often at home. On the other hand, the non–cardiac arrest emergencies were more often witnessed.
Cardiac Arrest Survival and Recognition
Prevalence of cardiac arrest was 3.0% in all emergency calls with a high-priority response (Table 2). A cardiac arrest was present in 285 calls, of which 203 were recognized during the call, corresponding to a sensitivity of 71%. In 64 calls, a cardiac arrest was incorrectly suspected, resulting in a specificity of 99.3%.
The 82 patients with a cardiac arrest not recognized by the dispatcher showed lower survival rates compared with the 203 patients with recognized cardiac arrests, as shown in an unadjusted Kaplan-Meier model (log-rank P=0.04; Figure 2). No difference in mean age (66 versus 65 years, respectively; P=0.9) or gender (P=0.7) was found between patients with a recognized and unrecognized cardiac arrest.
When a cardiac arrest was suspected, the mean time interval between call and dispatching was 1.88 minutes (SD, 1.10 minutes) versus 2.82 minutes (SD, 1.60 minutes) when cardiac arrest was not suspected (P<0.001). Mean time interval from call to arrival also differed significantly: 8.55 minutes (SD, 4.93 minutes) for calls with cardiac arrest suspicion versus 9.95 minutes (SD, 3.73 minutes) for calls without cardiac arrest suspicion (P=0.01). No CPR instructions were given in unrecognized cardiac arrest calls. Of the correctly recognized cardiac arrest calls, 24% of the callers indicated that they knew how to give CPR, and 37% received instructions from the dispatcher.
When not recognizing the cardiac arrest, the dispatcher did not ask about the patient’s breathing in 42 calls (51%); the caller gave a positive answer when asked about presence of breathing in 16 calls (20%); and the patient was reported to breathe abnormally in 20 calls (24%). For the calls in which cardiac arrests were recognized, these numbers were 51 (25%), 10 (5%), and 41 (20%), respectively. Asking for breathing differed significantly in these groups (P<0.001).
Trigger Words and the Presence of a Cardiac Arrest
The description of breathing in relation to the time it took to recognize that the call was about a patient in cardiac arrest is shown in Table 3. Normal breathing was never described in patients with a cardiac arrest. When a patient was reported to breathe but the dispatcher did not ask for a description of the breathing, the dispatcher’s recognition was significantly delayed by up to half a minute compared with all other descriptions of breathing. Abnormal breathing was described in ≈40% of the calls in which the dispatcher asked if the patient was breathing. In those patients, prevalence of cardiac arrest was 32%.
Effect of the Trigger Words
A complex samples stepwise backward logistic regression analysis identified all trigger words that significantly predicted a cardiac arrest (Table 4). These trigger words were used more often in a call that concerned a cardiac arrest than in other high-priority emergency calls. Besides the 2 terms indicating the absence of normal breathing, description of the facial color as blue, gray, or pale contributed substantially to the probability of a cardiac arrest. The description of the patient as “is dead” had the highest odds (67) for the presence of cardiac arrest in the multivariable analysis. Only the trigger word “fainted” lowered the probability of a cardiac arrest. The receiver-operating characteristics curve for the predictive accuracy of the multivariable model of the trigger words for the presence of cardiac arrest is shown in Figure 3. The average AUC was 0.861 (95% CI, 0.834 to 0.888), indicating good accuracy for the logistic model. The Hosmer-Lemeshow graph in Figure 4 shows the predicted risk plotted against the observed rate of cardiac arrest and demonstrates excellent correspondence between observed and predicted rates over a wide range of probabilities. The assumption of good model fit was not rejected (Hosmer-Lemeshow test, P=0.38).
In this study, we found a prevalence of 3.0% calls involving cardiac arrest in all emergency calls, of which the dispatchers correctly identified 71%. The most important reason for missing cardiac arrest diagnosis was insufficient questioning; in half of those calls, dispatchers did not ask about respiration. In 20%, they did ask if the patient was breathing but misinterpreted the confirming answer as a sign of circulation because no type of breathing was described. A quarter of the callers reported abnormal breathing, but the dispatchers did not consider the possibility of a cardiac arrest. If dispatchers treated an unconscious patient who was not breathing normally as a possible cardiac arrest patient, all cardiac arrests would have been recognized. This confirms the recommendations of Eisenberg et al.8 We then could expect more people to survive the resuscitation effort because we found a significant association between survival and cardiac arrest recognition. No telephone CPR instruction was offered, and the first ambulance was dispatched significantly later than if a cardiac arrest was correctly suspected.
The Description of Breathing
The present study is one of the first in which not only recordings of the cardiac arrest calls but also the calls not related to cardiac arrest were analyzed. In this way, we were able to estimate the value of breathing description in the broader perspective of all emergency calls. We could determine the predictive value of the trigger words, not just their sensitivity, in patients with a cardiac arrest. In our study, abnormal breathing contributed little to differentiate cardiac arrest calls from all emergency calls; the incidence of abnormal breathing was >10% in all emergency calls (967 of 9579), whereas prevalence of cardiac arrest in abnormal breathing victims was only 7% (64 of 967). Only the absence of breathing or normal breathing was likely to contribute to dispatchers’ cardiac arrest recognition. Description of normal breathing completely excluded the possibility of a cardiac arrest; absence of breathing augments the probability of a cardiac arrest to almost 50%. Mentioning the presence of breathing without further description delayed recognition for half a minute; dispatchers probably found this answer confusing and were looking for other signs that could indicate a cardiac arrest. Agonal breathing is known to mislead dispatchers into thinking the patient still breathes normally; dispatchers rarely give telephone CPR when the caller describes gasping.11,12
Model of Cardiac Arrest Prediction
We created an explanatory model of the probability of a cardiac arrest based on the dialogue with the caller. The AUC of the receiver-operating characteristics curve is between 0.8 and 0.9, indicating that the model provides an accurate diagnostic test. Furthermore, the Hosmer-Lemeshow test showed good fit of the model. No protocol or dispatch system takes into account the spontaneous words the caller says to describe the situation. We could identify several trigger words that pointed toward the presence of a cardiac arrest. Blue, gray, or pale facial color significantly predicts a cardiac arrest, as do terms that describe the condition of the patient as “is dead” or “is dying.” This implies that the lay caller is able to describe the victim’s condition quite accurately. Besides the absence of breathing, mentioning “abnormal breathing” also contributes to an increased probability of a cardiac arrest being present. If the caller thinks the victim has fainted, the presence of a cardiac arrest is less likely. Some trigger words that significantly contributed to this model would cease to become valuable if actively elicited by the dispatcher. For example, “send ambulance quickly” when said spontaneously implies that the caller is afraid the ambulance may come too late. If the dispatcher were to ask all emergency callers if they would like the ambulance to be sent quickly, a high positive response is expected, and thus the question loses any value for identifying a cardiac arrest.
The study by Nurmi et al13 also attempted to model the dialogue between caller and dispatcher. Only questions about consciousness and breathing were associated with cardiac arrest when a single-predictor Bayesian model was used in witnessed cardiac arrest calls. Because non–cardiac arrest calls were not audited, this model shows when dispatchers recognize a cardiac arrest rather than how a caller describes a cardiac arrest.
Comparison With Other Studies
The percentage of recognized cardiac arrests is similar to those reported by other studies, varying between 51% and 83% recognition.3,6,13–16 All dispatch centers used similar questions to interrogate the caller. Working with a computerized system like the Medical Priority Dispatch System did not result in higher sensitivity and specificity than when such a system was not used.14–16
Important design differences are evident between all these studies and ours. We excluded calls placed outside the patient’s site, because the caller cannot describe the condition of the patient if he or she is not in the vicinity. Calls before collapse were defined as “no cardiac arrest present at the moment of call.” In contrast, Calle et al3 included 39% prearrest calls. In that study, a call also was diagnosed as not correctly evaluated if the caller was not at the patient’s side or if the second-tier ambulance with advanced life support care was not at the post or outside the region. We attempted to select those calls in which the dialogue was meaningful for the recognition process because the caller observed the victim.
Not recognizing a cardiac arrest obviously results in not offering telephone-assisted CPR whereas telephone-assisted CPR has been shown to increase survival rates.17 In addition, if the ambulance is dispatched later, CPR could be delayed even more, and the first defibrillation shock is delayed, with a negative impact on survival.1,18,19
The Consequence of 100% Sensitivity
Strict protocol compliance alone should minimize the group of calls in which a cardiac arrest was not suspected but was present (false-negative). In its present form, strict compliance to the dispatch protocol gives high false-positive rates.14 If all 1181 calls in which breathing was described as abnormal or absent would result in a dispatch for suspected cardiac arrest, no case of cardiac arrest would be missed (100% sensitivity), but the false-positive rate would be 76%. This rate has financial and organizational implications. If financial resources are of no concern, this approach is optimal. Otherwise, the protocol would benefit from further optimization.
This study analyzes the interaction between the dispatcher and the caller. Not all information was elicited in a structured dialogue; dispatchers did not consistently ask about the patient’s breathing or any other trigger word that could implicate a cardiac arrest. Therefore, if a trigger word was not scored, the trigger could be either absent or present but simply not asked for. Our model accurately describes the available information but cannot be copied in a protocol for future call handling. For this, a more extended protocol for the dialogue needs to be developed on the basis of these findings.
The time interval from collapse to call in this study was not known. The time it takes to recognize the emergency and to dial the emergency number may vary and could influence survival rates in an unpredictable manner. However, we have no indication that this unknown part of the total delay has created a bias in our analysis.
This study demonstrated that it makes a difference if a cardiac arrest is recognized or not by the dispatcher. If not recognized, it takes dispatchers significantly longer to send out the right response team, and they do not offer telephone CPR instructions. These 2 factors significantly decrease survival.
Dispatchers are less likely to recognize a cardiac arrest if they did not follow the protocol. Complying with a standardized protocol increases recognition.16,20 If this protocol includes only unconsciousness and the absence of normal breathing, the sensitivity will be 100%, but the false-positive rate also will be high. We should attempt to develop a more sophisticated protocol and training by using spontaneously uttered trigger words describing the victim such as a description of facial color and other information. The tradeoff in various decision policies is not simple and in most dispatch centers not explicitly formulated. This study may be of assistance in making the decisions better founded.
We thank the dispatchers of the dispatch centers Amsterdam en Omstreken and Noord-Holland Noord for their contribution to data collection.
Source of Funding
This study was supported by a grant 2001B153 from the Netherlands Heart Foundation, Den Haag, the Netherlands.
Cummings RO, Ornate JP, This WH, Pepe PE. Improving survival from sudden cardiac arrest: the “chain of survival” concept: a statement for health professionals from the Advanced Cardiac Life Support Committee, American Heart Association. Circulation. 1991; 83: 1832–1847.
Heward A, Damiani M, Hartley-Sharpe C. Does the use of the Advanced Medical Priority Dispatch System affect cardiac arrest detection? Emerg Med J. 2004; 21: 115–118.
Rea TD, Eisenberg MS, Culley LL, Becker L. Dispatcher assisted cardiopulmonary resuscitation and survival in cardiac arrest. Circulation. 2001; 104: 2513–2516.
Jakobsson J, Nyquist O, Rehnqvist N. Effects of early defibrillation of out-of-hospital cardiac arrest patients by ambulance personnel. Eur Heart J. 1987; 8: 1189–1194.
This study evaluated the recognition of an out-of-hospital cardiac arrest by ambulance dispatchers. Survival after 90 days was significantly lower if the dispatcher did not suspect that the emergency call concerned a cardiac arrest. Not recognizing a cardiac arrest resulted in not offering telephone cardiopulmonary resuscitation instructions, and the ambulance arrived a mean of 1.40 minutes later on scene. These findings directly confirm the importance of telephone cardiopulmonary resuscitation and indirectly of early defibrillation. The main reason for not suspecting the cardiac arrest was not complying with the dispatch protocol, especially asking if and how the patient was breathing. We also identified words that callers use to describe a patient in cardiac arrest. Such findings may aid in developing a more sophisticated protocol to make recognition of a cardiac arrest faster and more accurate. Early recognition by dispatchers is an essential key in the chain of survival after out-of-hospital cardiac arrest, which has been underestimated until recently.
The online-only Data Supplement is available with this article at http://circ.ahajournals.org/cgi/content/full/CIRCULATIONAHA.108.768325/DC1.