Depending on the length of intensive care unit (ICU) stay and subsequent neurological outcome, resources required by successfully resuscitated cardiac arrest patients might be enormous. For example, a quality-adjusted life year after successful cardiopulmonary resuscitation (CPR) may cost up to $225,000 US, which is about 143,200 euros.1 Therefore, the value of prehospital CPR efforts may be questioned,2 with a trend toward optimization of resources to maximize outcome and minimize costs.3 The possibility to predict outcome of a patient undergoing a prehospital CPR attempt on the basis of simple variables such as age, medical history, ischemia duration, and the initial electrocardiogram (ECG) rhythm could be very helpful for optimization of resource utilization. For economic, ethical, social, and legal reasons, it is necessary to determine the outcome as early as possible.4
Despite numerous efforts, there are no clinically validated algorithms to identify a cardiac arrest patient on the scene who should not be resuscitated because there is no chance for survival and recovery of a self-determined life.5–8 Any algorithm that might result in a rapid decision to initiate or continue CPR, or not to do so, has to be validated with excellent prediction parameters and meet the highest ethical standards.9
The purpose of the study was to create a CPR prediction algorithm that facilitates decision-making for prehospital CPR attempts to decrease the number of patients dying in the hospital or remaining in a vegetative state.
The original study was performed in 33 communities with 44 physician-manned emergency medical service units in Austria, Germany, and Switzerland between June 1999 and March 2002 and was designed as a blinded, prospective, multicenter, randomized, controlled clinical trial.10 The criteria for inclusion were out-of-hospital cardiac arrest in adult patients who presented with ventricular fibrillation, asystole, or pulseless electrical activity that required CPR with a vasopressor. Exclusion criteria were successful defibrillation without a vasopressor, documented terminal illness, no IV access, hemorrhagic shock, pregnancy, cardiac arrest after trauma, age younger than 18 yr, or presence of a do-not-resuscitate order. Patients presenting with pulseless electrical activity or asystole were randomized immediately, and patients with ventricular fibrillation were randomized after the first three defibrillation attempts failed. When a patient underwent randomization, either 1 mg of epinephrine or 40 IU of vasopressin was injected. If spontaneous circulation was not restored after the second administration of the study medication within 3 min, the patient was given additional injections of epinephrine at the discretion of the emergency medical service physician who was leading the CPR attempt. The primary end point of the study was hospital admission and the secondary end point was hospital discharge rate. There were no preassigned termination criteria. The emergency medical service physician was in charge of deciding whether to continue resuscitative efforts.10
In total, 1219 prehospital cardiac arrest patients were randomized, but 33 patients were excluded because of a missing study drug code. Another 20 patients could not be included because of missing follow-up data, resulting in 1166 included patients. Those 1166 patients were assigned to three groups for post hoc analysis: the first group (n = 786) contained the patients who died on scene, the second group included the patients who initially survived but died in the hospital (n = 265), and the third group comprised the patients who survived and could be discharged (n = 115; Table 1). Patients who survived to discharge (n = 92) were evaluated within cerebral performance categories (CPC)11 (1 = normal life possible; 2 = moderate disability, able to manage simple daily life tasks; 3 = conscious but dependent on the help of others; 4 = vegetative state). As in other CPR studies, CPC 1 and 2 were categorized as “good” and CPC 3 and 4 as “poor.”3,12
In the first step of analysis, patients’ characteristics were compared with the χ2 test, Fisher’s exact test, analysis of variance, or Student’s t-test, as appropriate. Logistic regression analysis was used to assess factors which seemed to be of prognostic significance, and a receiver operating characteristic curve (ROC) was created. The score developed was not externally validated. The data were analyzed with the statistical software package SPSS 15.0.0 (SPSS, Chicago, IL).
Analysis of clinical data, CPR treatment, initial ECG, assumed reason for cardiac arrest, and drugs that had been given apart from the study medication showed significant differences between the groups of patients who died on scene or in the hospital and those who could be discharged (Tables 1 and 2). According to previous studies, there was a survival advantage for younger patients, patients found in ventricular fibrillation, and those patients who had a witnessed arrest and received early basic and advanced life support (ALS). In terms of neurological outcome, the only difference between patients who were discharged with good or bad neurological outcome was ventricular fibrillation or asystole in the initial ECG, as well as sodium bicarbonate administration (Table 3).
The variables such as age, body mass index, gender, time until initiation of ALS, ventricular fibrillation, asystole, pulseless electrical activity, additional medication, preexisting morbidities, number of defibrillation attempts, witnessed arrest, and bystander CPR were used for a final logistic regression model (Table 4). However, this model did not produce good predictive values. Cox and Snell (0.098) and Nagelkerke R2 (0.207) statistics for evaluation of explanatory variables in predicting nondischarge were comparatively low. The Hosmer-Lemeshow test (P = 0.318) did not indicate a lack of fit of the overall model, but prediction of death or survival under a classification table showed either good sensitivity and bad specificity or vice versa. Discrimination of the scoring was also assessed using an ROC curve (Fig. 1); the ROC curve based on logistic regression (Table 4) showed an area under the curve of 0.795 (0.751-0.839) at a confidence interval of 95%. It was significantly different from random prediction (0.5; P < 0). Prediction models including only significant variables using a “forward-qualified method” did not provide improved results. A potential discriminator for predicting death was 0.5, revealing a sensitivity of 99.8% and a specificity of 2.9%.
In Europe and the United States, about 480,000 patients per year die in hospitals after initially successful CPR. This is reflected in approximate costs of $2.16 billion US, about 1.38 billion euros based on a conservative calculation of $3000 US in costs per day and an average ICU stay of 3 days.
ICUs are the most expensive areas of any hospital and may account for approximately 20% of the total costs of hospitals.1,2 Therefore, it is absolutely necessary to monitor resources in ICUs very closely. Furthermore, through the aging of the Baby Boom population, the average age of intensive care patients is increasing close to or even beyond the 70s, whereas budgets get more and more restrictive. If intensive care budgets became more restrained, the increasing demand for cost-intensive therapeutic actions will lead to worse quality of treatment, economic crisis, or restriction of treatment. This raises the question of whether we can still afford maximum treatment for each critically ill or injured patient.
We hoped to address these problems by creating an algorithm that facilitates decision-making for prehospital CPR attempts. However, with our simple parameters for a prehospital scoring algorithm, it was not possible to predict either the survivors (to optimize therapy) or the patients who would die (to discuss early withdrawal of therapy). At a discrimination value (threshold) of 0.5, dying patients were predicted well, with 953 of 955 nonsurviving patients (99.8%) being recognized correctly. At the same time, 101 of 104 survivors (97.1%) would not have been classified correctly as survivors, indicating a sensitivity of 99.8% but a specificity of only 2.9%. At a discrimination value of 0.99, almost all survivors would have been recognized correctly (103 of 104; 99.0%) but at the same time, 883 of 955 patients (92.5%) who died would not have been classified correctly. This indicates a sensitivity for survival of 99%, which is satisfactory from an ethical point of view, but with a specificity of only 8%, which is of no economic advantage.
Unlike established prediction models in major trauma, a valid prediction of CPR results on the basis of our data is not possible. For German-speaking European emergency medical services, we were able to show that prediction of long-term survival is not possible in a prehospital setting, which is in agreement with the Quality Standards subcommittee of the American Academy of Neurology.5 Although we could not produce a prediction algorithm, we could show that a substantial number of patients who were found in very disadvantageous circumstances still survived, and even regained a self-determined life.
Pupillary light response, corneal reflexes, motor response to pain, seizures, serum neuron-specific enolase, and somatosensory-evoked potentials can reliably assist in accurately predicting poor outcome in comatose patients after CPR.13 This indicates that a decision about withdrawal of life support is better made in the clinical setting, with more time available and better logistical support. For example, the marker neuron-specific enolase has a sensitivity of 91% and a specificity of 100%.8 Somatosensory-evoked potentials have a sensitivity of 94% and a specificity of 97%.14 A combination of neuron-specific enolase and somatosensory-evoked potentials reliably predicts poor outcome in postanoxic coma as early as 24 h after CPR.13 The predictions of emergency physicians asked to judge survival chances of CPR patients based on emergency medical services data showed a sensitivity of only 70% and a specificity 65%; sensitivity improved to 80% after 60 min in the hospital and availability of further laboratory data, but specificity decreased to 48%.14 Estimations of physicians asked to predict survival chances for patients resuscitated in the hospital were even worse than pure chance.15 Similar to our study, attempts to predict survival with cardiac ALS scores (variables: witnessed cardiac arrest, immediate start of CPR after collapse, initial ECG rhythm, and time until arrival of paramedics) were disappointing as well.6 Furthermore, decisions about which treatment strategy to use are made much more independently in the hospital than in an emergency setting. Being pressured for time, having limited information available, and being stressed by relatives on the scene may impair the power of judgment of the emergency medical services personnel.
Compared with reliable markers such as neuron-specific enolase, measured by a blood sample,16 many of our analyzed variables are less accurate. The time from cardiovascular collapse until start of CPR was found to be estimated as longer than it actually was.17,18 Also, fine ventricular fibrillation can be misdiagnosed as pulseless electrical activity or asystole.19 Furthermore, ECG rhythms may proceed to a different rhythm within a short time20; thus, asystole as an initial rhythm might have been ventricular fibrillation a few seconds before. In this case, the parameter “initial ECG” rhythm would most likely produce unacceptable prediction values. This could explain why survival of patients with witnessed cardiac arrest and ventricular fibrillation ranged between 5.3% and 37.7%.21 One reason for the low sensitivity for hospital discharge analysis is the small number of patients (92/1166 = 7.9%) who could be evaluated in our study. And finally, there may be fundamental variations in the quality of CPR, which may have a significant effect on outcome.22
We compared our data with a Canadian study investigating the predictive value of an absence of return of spontaneous circulation, no shocks administered, and cardiac arrest that was not witnessed by the emergency medical services.7 Termination of CPR efforts in our study would have been recommended less often than in the Canadian study (34.9% vs 63.5%), because return of spontaneous circulation could be restored more often (45.4% vs 5.5%). Almost all patients presenting with the Canadian criteria to discontinue CPR (408/410 = 99.5%) were declared dead on the scene by our emergency medical service physician managing the CPR attempt. Only two (2/410 = 0.5%) patients who fulfilled the Canadian criteria7 were brought to the hospital in our study, where they died. This indicates that even without a prediction model, the ability of our emergency medical service physicians to discontinue CPR in obviously unsuccessful situations was efficient enough. This suggests that the value of a prediction model is more important in a basic life support emergency medical services system such as in the Canadian investigation,7 whereas in a cardiac ALS system as in the European vasopressin trial,10 the value of a prediction model is shifted into the hospital. In our original study, about 40% of patients survived at least short term, and in futile cases, the emergency medical services physician terminated CPR efforts on the scene.
Limitations need to be noted. Our retrospective analysis represents nonrandomized groups and the number of discharged survivors is small (92), but excellent in international comparison, and can be explained by the relatively open inclusion criteria of the original study. We found that good neurological outcome was more likely after ventricular fibrillation and less likely after asystole in the initial ECG; however, these facts were known before. The higher significance (P = 0.009) of treatment with sodium bicarbonate in bad neurological outcome seems to be an epiphenomenon because of higher likelihood of sodium bicarbonate administration during prolonged resuscitation efforts. Another caveat to this study is that at the time that the study was conducted, emergency medical services personnel followed the 2000 CPR guidelines, which are no longer considered to be optimal.23 Since the completion of this study, the 2005 CPR guidelines24 have been published and results might be different.25
In conclusion, in out-of-hospital patients with cardiac arrest, predicting survival based on variables documented in the field did not allow a valid forecast and should be considered primarily after hospital admission.
1. Lee KH, Angus DC, Abramson NS. Cardiopulmonary resuscitation: what cost to cheat death? Crit Care Med 1996;24:2046–52
2. Fischer M, Krep H, Wierich D, Heister U, Hoeft A, Edwards S, Castrillo-Riesgo LG, Krafft T. [Comparison of the emergency medical services systems of Birmingham and Bonn: process efficacy and cost effectiveness]. Anasthesiol Intensivmed Notfallmed Schmerzther 2003;38:630–42
3. Davis DP, Dunford JV, Poste JC, Ochs M, Holbrook T, Fortlage D, Size MJ, Kennedy F, Hoyt DB. The impact of hypoxia and hyperventilation on outcome after paramedic rapid sequence intubation of severely head-injured patients. J Trauma 2004;57:1–8; discussion 8–10
4. Berek K, Jeschow M, Aichner F. The prognostication of cerebral hypoxia after out-of-hospital cardiac arrest in adults. Eur Neurol 1997;37:135–45
5. Wijdicks EF, Hijdra A, Young GB, Bassetti CL, Wiebe S. Practice parameter: prediction of outcome in comatose survivors after cardiopulmonary resuscitation (an evidence-based review): report of the Quality Standards Subcommittee of the American Academy of Neurology. Neurology 2006;67:203–10
6. Larsen MP, Eisenberg MS, Cummins RO, Hallstrom AP. Predicting survival from out-of-hospital cardiac arrest: a graphic model. Ann Emerg Med 1993;22:1652–8
7. Morrison LJ, Visentin LM, Kiss A, Theriault R, Eby D, Vermeulen M, Sherbino J, Verbeek PR. Validation of a rule for termination of resuscitation in out-of-hospital cardiac arrest. N Engl J Med 2006;355:478–87
8. Zingler VC, Krumm B, Bertsch T, Fassbender K, Pohlmann-Eden B. Early prediction of neurological outcome after cardiopulmonary resuscitation: a multimodal approach combining neurobiochemical and electrophysiological investigations may provide high prognostic certainty in patients after cardiac arrest. Eur Neurol 2003;49:79–84
9. Mohr M, Kettler D. Ethical aspects of resuscitation. Br J Anaesth 1997;79:253–9
10. Wenzel V, Krismer AC, Arntz HR, Sitter H, Stadlbauer KH, Lindner KH. A comparison of vasopressin and epinephrine for out-of-hospital cardiopulmonary resuscitation. N Engl J Med 2004;350:105–13
11. A randomized clinical trial of calcium entry blocker administration to comatose survivors of cardiac arrest. Design, methods, and patient characteristics. The Brain Resuscitation Clinical Trial II Study Group. Control Clin Trials 1991;12:525–45
12. Hypothermia after Cardiac Arrest Study Group. Mild therapeutic hypothermia to improve the neurologic outcome after cardiac arrest. N Engl J Med 2002;346:549–56
13. Zandbergen EG, Hijdra A, Koelman JH, Hart AA, Vos PE, Verbeek MM, de Haan RJ. Prediction of poor outcome within the first 3 days of postanoxic coma. Neurology 2006;66:62–8
14. Madl C, Kramer L, Domanovits H, Woolard RH, Gervais H, Gendo A, Eisenhuber E, Grimm G, Sterz F. Improved outcome prediction in unconscious cardiac arrest survivors with sensory evoked potentials compared with clinical assessment. Crit Care Med 2000;28:721–6
15. Ebell MH, Bergus GR, Warbasse L, Bloomer R. The inability of physicians to predict the outcome of in-hospital resuscitation. J Gen Intern Med 1996;11:16–22
16. Fogel W, Krieger D, Veith M, Adams HP, Hund E, Storch-Hagenlocher B, Buggle F, Mathias D, Hacke W. Serum neuron-specific enolase as early predictor of outcome after cardiac arrest. Crit Care Med 1997;25:1133–8
17. Harvey AL, Gerard WC, Rice GF Jr, Finch H. Actual vs perceived EMS response time. Prehosp Emerg Care 1999;3:11–4
18. Morrison LJ, Angelini MP, Vermeulen MJ, Schwartz B. Measuring the EMS patient access time interval and the impact of responding to high-rise buildings. Prehosp Emerg Care 2005;9:14–8
19. Ornato JP, Gonzales ER, Garnett R. Atrial fibrillation with high degree atrioventricular block masquerading as ventricular fibrillation masquerading as asystole during cardiac arrest. Crit Care Med 1987;15:334–5
20. Callans DJ. Out-of-hospital cardiac arrest—the solution is shocking. N Engl J Med 2004;351:632–4
21. Fredriksson M, Aune S, Thoren AB, Herlitz J. In-hospital cardiac arrest—an Utstein style report of seven years experience from the Sahlgrenska University Hospital. Resuscitation 2006;68:351–8
22. Wik L, Kramer-Johansen J, Myklebust H, Sorebo H, Svensson L, Fellows B, Steen PA. Quality of cardiopulmonary resuscitation during out-of-hospital cardiac arrest. JAMA 2005;293:299–304
23. Guidelines 2000 for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care. Part 6: advanced cardiovascular life support: 7C: a guide to the International ACLS algorithms. The American Heart Association in collaboration with the International Liaison Committee on Resuscitation. Circulation 2000;102:I142–57
24. 2005 American Heart Association Guidelines for Cardiopulmonary Resuscitation and Emergency Cardiovascular Care. Circulation 2005;112:IV1–203
© 2009 International Anesthesia Research Society
25. Ewy GA. Cardiac resuscitation—when is enough enough? N Engl J Med 2006;355:510–2