Derivation and Validation of the CREST Model for Very Early Prediction of Circulatory Etiology Death in Patients Without ST-Segment–Elevation Myocardial Infarction After Cardiac Arrest

The outcomes of patients following successful cardiopulmonary resuscitation vary significantly among centers, and optimal treatment pathways for individual patients remain controversial.^1–3 Under current treatment paradigms, ≈40% of patients successfully resuscitated and admitted to an intensive care unit (ICU) survive to discharge, varying by case mix and the severity of their injuries.⁴ Two-thirds of in-hospital deaths are secondary to neurological injury and about one-third of patients die because of a circulatory etiology, including repeat cardiopulmonary arrest, progressive refractory shock, and multiorgan system failure.^5,6

Recent studies of electroencephalography performed after resuscitation from cardiac arrest suggest that accurate and very early assessment of neurological injury is possible. In animal models, quantitative electroencephalography assessment begins to stratify brain injury severity at 30 to 60 minutes after resuscitation.^7,8 Processed and raw electroencephalography modalities in observational human studies demonstrate the severity of neurological injury at 3 to 6 hours postresuscitation with 80% to 95% accuracy.^9–17 We anticipate that, in conjunction with a circulatory risk stratification model, these methods of assessing neurological injury may be useful to triage patients to more appropriate and effective individualized treatment pathways. For example, a patient with mild brain injury and high risk of circulatory-etiology death might be expected to benefit greatly from mechanical hemodynamic support or early coronary angiography and revascularization,¹⁰ whereas urgent coronary revascularization might not benefit a similar patient with severe brain injury.¹² Such hypotheses, crucial to developing effective new therapies and improving the cost-effectiveness of care, can only be tested if accurate risk assessment models are developed. Although preliminary work using processed electroencephalography for early stratification of brain injury severity has been published, and several sophisticated models for predicting survival and functional outcome proposed,^18,19 no practical tools are available to estimate the risk of circulatory-etiology death early after resuscitation, hampering our ability to weigh competing risks at the point of care.

Current guidelines recommend urgent coronary angiography for patients with ST-segment–elevation myocardial infarction (STEMI) following cardiopulmonary resuscitation,^20,21 but the management of those without STEMI, constituting the majority of postresuscitation ICU admissions, remains controversial.^1–3 Some experts propose early cardiac catheterization and percutaneous coronary intervention (PCI) in all patients without an obvious noncardiac etiology for the arrest,^22–24 whereas others practice the more conservative approach of delaying aggressive interventions until neurological viability is demonstrated.^3,25 The aim of this study was to develop and validate a simple bedside prediction tool to help determine the risk of circulatory-etiology death in survivors of cardiac arrest without STEMI, using data routinely available to clinicians at the time of hospital admission.

The data, analytic methods, and study materials will be made available by the corresponding author to other researchers for purposes of reproducing the results on reasonable request.

Between January 2003 and December 2013, 2791 comatose patients who survived to hospital admission following cardiac arrest were entered into the INTCAR database. Of these, 661 were excluded because they met criteria for STEMI and 709 were excluded because of in-hospital cardiac arrest. A further 465 patients were removed because of missing data (primarily, the admission echocardiogram) leaving 956 patients in the final cohort for analysis (Figure 1). The derivation cohort comprised 638 patients, whereas the validation cohort was formed of 318 patients. Baseline demographic and clinical characteristics of the patients are shown in Table 1. There were no significant differences between the 2 groups of patients. Of the patients in the derivation cohort, 121 (18.9%) met the end point of circulatory-etiology death. In the validation cohort, 60 (18.9%) experienced circulatory-etiology death.

In a univariate analysis, older age (P<0.0001), hypertension (P=0.0017), diabetes mellitus (P=0.0005), history of coronary artery disease (P<0.0001), nonshockable rhythm (P=0.046), longer ischemic time (P=0.022), the presence of shock at admission (P<0.0001), coronary angiography at any time during admission (P=0.0017), and severe left ventricular dysfunction (P<0.0001) were associated with circulatory-etiology death. There were no differences between those with and without circulatory-etiology death regarding rate of early angiography or PCI (early was defined as performed before awakening; Table 2).

The multivariable regression yielded a clinical predication model for circulatory-etiology death that included 5 variables: coronary artery disease (odds ratio [OR], 2.86; P<0.0001), nonshockable rhythm (OR, 1.75; P=0.017), initial ejection fraction <30% (OR, 2.11; P=0.0018), shock at presentation (OR, 2.27; P=0.0006), and ischemic time >25 minutes (OR, 1.42; P=0.13). These results are presented in Table 3. For simplicity, prediction accuracy was summarized by assigning each of the 5 variables 1 point, and then binning predicted and observed event rates by a cumulative risk index called the CREST score (coronary artery disease, initial heart rhythm, low ejection fraction, shock at the time of admission, and ischemic time >25 minutes). In the derivation cohort, the model showed adequate calibration as evidenced by a nonsignificant Hosmer-Lemeshow χ² test P=0.47 and good discrimination (area under the curve, 0.73). Similarly, in the validation cohort, the calibration was found to be adequate with a nonsignificant Hosmer-Lemeshow χ2 test (P=0.41) and the discrimination was also good (area under the curve, 0.68). A roughly linear increase in likelihood of circulatory-etiology death is seen with incremental increases in cumulative CREST scores in the derivation (Figure 2) and validation (Figure 3) cohorts.

A simple clinical prediction model using data easily obtained at the time of ICU admission accurately estimated the risk of circulatory-etiology death in a large registry population of patients resuscitated from out-of-hospital cardiac arrest and without STEMI. This tool was derived from a cohort of patients from the INTCAR registry and validated in a smaller, similar group of patients from the same registry. History of coronary artery disease, nonshockable rhythm, initial ejection fraction <30%, shock at the time of presentation, and total ischemic time >25 minutes accurately and pragmatically predicted the risk of circulatory-etiology death. Such a tool may be especially useful at the bedside, ideally in conjunction with an assessment of the severity of brain injury, to triage patients to individualized treatment pathways after cardiac arrest, and to select the most appropriate patients for clinical trials of hemodynamic support, coronary revascularization, or other novel therapies to improve outcomes after cardiac arrest.

A recent Institute of Medicine report called for intensified research into postresuscitation cardiac arrest care.³¹ We believe the need for a sophisticated and early assessment of circulatory and neurological risks in this population is great. This position is supported by a recent trial of targeted temperature management in survivors of cardiac arrest, in which comatose patients with widely varying severity of neurological and cardiac injury were randomly assigned to different temperature targets, finding equivalence in outcomes at 2 levels of mild hypothermia: 36°C or 33°C.³² Subgroup analyses of that trial have generated further uncertainty with the identification of potential harm with moderate hypothermia in survivors of cardiac arrest with shock,^33–35 whereas other investigators showed more or less benefit of hypothermia in certain subgroups of survivors of cardiac arrest based on the duration of ischemia.^36,37 Others still have suggested potential benefit from therapeutic hypothermia in patients with cardiogenic shock.³⁸ It is reasonable to conjecture that subgroups of patients with a different severity of neurological or circulatory injury might have a different response to therapy, and could benefit from individualized treatment regimens. Such regimens can only be developed when an accurate, practical, and early approach to risk assessment and stratification is available.

An ad hoc approach to treatment of survivors with cardiac arrest likely results in both under- and overtreatment with coronary reperfusion strategies or mechanical circulatory support. In one study, as many as 69% of patients, determined by early processed electroencephalography to have a mild brain injury, were felt to receive submaximal circulatory support, and 20% of these patients died a circulatory-etiology death.¹⁰ Other patients who likely had a more severe brain injury were treated with urgent revascularization but later died a neurological death.¹⁰ As the number of patients resuscitated from cardiac arrest increases because of the increasing rates of bystander cardiopulmonary resuscitation and improved cardiopulmonary resuscitation techniques,³⁹ it will become more important to match aggressive care with the appropriate type and severity of injury. We believe that early assessment of both circulatory and neurological risks will allow for the most effective triage to individualized treatment pathways.

Our study is the first to show that patients can be stratified according to risk of circulatory-etiology death early after resuscitation. Studies of postresuscitation cardiac arrest care have largely focused on neurological or functional outcomes and overall survival. Yet circulatory collapse and progressive multiorgan system failure represent up to one-third of deaths in patients who are admitted after cardiac arrest,⁵ and may be amenable to treatment with existing therapies. Early recognition of the patients at highest risk for specific poor outcomes (neurological versus circulatory-etiology death) may allow for intensified, targeted support, resulting in improved outcomes and a greater efficiency of care. Prior studies have used presenting rhythm as a crude measure of cardiac risk and have shown that clinicians empirically incorporate risk of ischemic heart disease and prognosis into decisions regarding early aggressive postresuscitation care.^3,10 Different prediction methods have been developed to assess overall risk of poor outcomes,⁴⁰ but grouping all etiologies of patients with poor outcomes together limits what we may learn from their outcomes, obscuring fundamental differences in pathophysiology and real targets for therapeutic intervention. Current guidelines state that coronary angiography and PCI may be reasonable in select (eg, electrically or hemodynamically unstable) patients with suspected cardiac-etiology arrest in the absence of STEMI.¹⁹ However, some argue that all patients with presumed cardiac etiologies for cardiopulmonary arrest should undergo emergent cardiac catheterization based on the high likelihood of finding a culprit lesion, and the potential for revascularization to improve both immediate hemodynamics and long-term cardiac function.^20,25 Conversely, there could be risks associated with the interhospital transfer of unstable patients, although this may be offset by the benefits of moving to a higher-volume receiving center.⁴¹ Our results suggest the CREST tool could be a more objective measure to assist in the early assignment of patients to appropriate therapies and interventions, or away from potentially harmful or futile interventions.

Clinicians seeking to use the CREST score in clinical practice should be aware that the INTCAR definition of shock differed from current American College of Cardiology terminology. In this study, shock was defined as hypotension persisting despite vasopressors, or the need for intra-aortic balloon counterpulsation, a significantly more severe definition than that used by the American College of Cardiology, which recognizes any patient requiring vasopressors to maintain systolic blood pressure >90 mm Hg as manifesting shock.

The results of this study should be interpreted within the context of its limitations. First, the design is retrospective and therefore subject to selection bias, information bias, and other confounders. Our data depend on an accurate determination of cause of death by bedside clinicians, which may not always be clear. Data were collected over a long period of time and from many institutions, potentially introducing bias from variations in standards of practice over time and among sites. Although it is the intention of the Registry to enroll consecutive patients, we cannot verify that all participating sites have abided by this, with resulting risk of selection bias. Over 450 patients were excluded from the analysis, many because of the lack of an admission echocardiogram or an unknown ischemic time when the arrest was unwitnessed; these omissions may further limit the generalizability of this study or create selection bias. Although early echocardiography is useful to assist with risk stratification, a delay in otherwise appropriate urgent coronary angiography for the purposes of obtaining echocardiography could be dangerous, and is not advised.

Our risk prediction model addresses only circulatory-etiology death, which accounts for ≈20% of the overall outcomes of patients admitted after resuscitation. Although the performance of the tool is only moderate, we anticipate the development of tools with better performance, and an area under the curve in this range is consistent with other widely used risk prediction tools, such as the CHADS2vasc score, to predict thromboembolic stroke risk in atrial fibrillation⁴² and the Pooled Cohort Equations for assessing the 10-year risk of developing atherosclerotic cardiovascular disease.⁴³ Neurological-etiology death is more common, and many other variables contribute to outcome in these complex patients. As with any scoring system, regardless of its predictive accuracy, the score can only assess risk for a particular outcome and may not apply to prediction of outcomes at an individual patient level; it should, therefore, be used only in conjunction with other available information at the bedside. Finally, because of the design of INTCAR, only patients who survived to ICU admission were considered for inclusion, thus potentially leading to an underestimation of the risk of resuscitated patients still in the emergency department, some of whom may have had life support measures withdrawn before inclusion in the Registry. As such, and because excluding patients without admission echocardiogram may have created selection bias, the CREST model requires prospective validation in an emergency department cohort. Despite these limitations, the CREST model is unique in conception and timing, and our scoring tool has a potentially important role for triage at the point of care.

Our derivation and validation of the CREST score in a diverse cohort of patients from European and American sites and a mixture of centers ranging from academic to community hospitals, suggests wide applicability, generalizability, and consistent performance. Analysis was performed on all initial survivors of out-of-hospital cardiac arrest without STEMI, regardless of clinical condition. In addition, the score is easily calculated with data available at or soon after admission to the ICU, does not require mathematical manipulations, and is therefore easy for clinicians to use.

Further work is needed in this area, including prospective application of the CREST score, combining the CREST model with a neurological risk stratification tool,¹⁰ and prospective assessment of how interventions such as early coronary angiography, PCI, or mechanical circulatory support modify the risk of circulatory-etiology death in individual risk groups. Future iterations of the tool might incorporate additional factors such as known chest pain preceding the arrest, preexisting cardiomyopathy, initial pH, or serum lactate levels. Nonetheless, the CREST model in its present form is validated and provides clinical information useful for early triage purposes at the point of care.

The authors thank the patients, their families, and the dedicated data abstractors for making this project possible.

Minneapolis Heart Institute, MN (n=280): Michael Mooney, MD; Maine Medical Center, Tufts University, Portland (n=277): Teresa May, DO; Ullevål University Hospital, Ullevål University Hospital, Oslo, Norway (n=204): Kjetil Sunde, MD; Vanderbilt University Medical Center, Nashville, TN (n=194): John McPherson, MD; Lehigh Valley Health Network, Allentown, PA (n=169): Nainesh Patel, MD; Uppsala University Hospital, Sweden (n=166): Sten Rubertsson, MD; Cardiocenter, General Teaching Hospital, Prague, Czech Republic (n=151): Ondrej Smid, MD; Stavanger University Hospital, Norway (n=147): Eldar Soreide, MD; Eastern Maine Medical Center, Bangor (n=147): Robert Hand, MD; Skåne University Hospital, Lunds Universitet, Sweden (n=137): Malin Rundgren, MD; Landspitali University Hospital, Reykjavik, Iceland (n=120): Felix Valsson, MD; St John’s Mercy Medical Center, St. Louis, MO (n=113): Farid Sadaka, MD; Maastricht University Medical Center, Netherlands (n=94): Bas Bekkers, MD; Rigshospitalets Heart Center, Copenhagen, Denmark (n=61): Michael Wanscher, MD; Blekingesjukhuset, Karlskrona, Sweden (n=46): Eva-Lotta Lindell, MD;Ochsner Baptist Medical Center, New Orleans, LA (n=40): Paul McMullan, MD; Sarver Heart Center, University of Arizona, Tucson (n=36): Karl Kern, MD; Falu Hospital, Sweden (n=35): Pehr Guldbrand, MD; Halmstad Regional Hospital, Sweden (n=34): Anders Torstensson, MD; Kristianstad Central Hospital, Sweden (n=33): Krystyna Dybkowska, MD; Kalmar Hospital, Sweden (n=33): Johan Israelsson, MD; Gentofte Hospital, Hellerup, Denmark (n=25): Ulrik Skram, MD; Central Maine Medical Center, Lewiston, ME (n=25): Michelle Guzowski, MD; Asklepios Kliniken, Langen, Germany (n=23): Hans-Bernd Hopf, MD; Helsingborgs Lasarett, Sweden (n=23): Niklas Nielsen, MD; Örebro University Hospital, Sweden (n=22): Stefan Persson, MD; Swedish Medical Center, Englewood, CO (n=22): Ira Chang, MD; Östersund Hospital, Sweden (n=22): Line Samuelsson, MD; Danderyd Hospital, Sweden (n=20): Eva Oddby, MD; Karlstad Central Hospital, Sweden (n=18): Kristina Savolainen, MD; Kungälv Hospital, Sweden (n=16): Richard Zätterman, MD; Kärnsjukhuset Skövde, Sweden (n=13): Daniel Rodriguez, MD; Columbia University, New York, NY (n=12): Stephan Mayer, MD; Evangelisches Krankehnhaus, Bonn, Germany (n=11): Markus Födisch, MD; Sjukhuset i Lidköping, Sweden (n=8): Beata Oscarsson, MD; Centrallasarettet, Västerås, Sweden (n=3): Håkan Scheer, MD; Sahlgrenska Universitetssjukhuset/Östra, Sweden (n=3): Roman Sarbinowski, MD; Intensivårds-Avdelningen, Lasarettet i Ystad, Sweden (n=2): Ulf Hyddmark, MD; Lariboisière Hospital, Paris, France (n=2): Nicolas Deye, MD; IVA, Länssjukhuset Ryhov, Sweden (n=1): Anna Lindbom, MD; Santa Chiara Hospital, Trento, Italy (n=1): Claudia Armani, MD; Södersjukhuset, Sweden (n=1): Sune Forsberg, MD; Mora Lasarett, Sweden (n=1): Anders B. Ericsson, MD.