A multicenter prospective study was performed in 3 academic hospitals in Spain. We included consecutive adult survivors of nontraumatic OHCA admitted to the intensive cardiac care unit who underwent TTM between January 2007 and July 2015 (only patients who survived until hospital admission and who underwent TTM where analyzed). The overall population was divided into 2 groups: data from patients admitted at 2 hospitals (Hospital General Universitario Gregorio Marañón and Hospital Universitario de Salamanca - derivation group) were used to create a predictive model, whereas data from patients treated at a third hospital (Hospital Universitario de Bellvitge - validation group) were used to conduct an external validation of the model.

All survivors of OHCA with a Glasgow Coma Scale score ≤ 8 were admitted to the intensive cardiac care unit, monitored and treated according to international clinical standards and the institutional protocols, including adequate sedation, paralysis to prevent shivering during hypothermia, mechanical ventilation and, if needed, vasoactive or inotropic support was administered to maintain a mean arterial blood pressure> 80 mmHg. When indicated (briefly, if no obvious noncardiac cause was present or if ongoing ischemia was suspected), early coronary angiography and percutaneous coronary intervention were performed. In a subset of patients, hypothermia was started before hospital arrival, but in most patients, it was initiated as soon as possible after hospital admission. The target temperature was 32°C to 34°C, maintained for 24hours. Three methods were used: nonautomatic temperature control with thermal blankets and ice (12.0% of patients), and automatic temperature control systems, either with surface cooling device (in 79.3% of patients) or with an endovascular system (in 8.7% of cases). Rewarming was done gradually, avoiding hyperthermia. At the end of rewarming, sedation was stopped and fever was prevented.

Data were collected prospectively using Utstein-syle recommendations,20 from prehospital and in-hospital patient care records. We collected demographic information and other general characteristics including vital signs, whether OHCA was witnessed, and the presence or absence of bystander cardiopulmonary resuscitations. Time from collapse to first cardiopulmonary resuscitation (which will be approximate in patients with unwitnessed arrest), time from collapse to ROSC, and time from ROSC to initiation of therapeutic hypothermia were registered. Arterial blood sampling was performed on admission. Cardiac arrest was dichotomized as shockable rhythms (ventricular fibrillation and ventricular tachycardia) and nonshockable rhythms (including asystole and pulseless electrical activity). Postresuscitation circulatory shock was defined as the occurrence of arterial hypotension (mean arterial pressure <60 mmHg or systolic blood pressure <90 mmHg) sustained for more than 60minutes following hospital admission, despite fluid resuscitation, leading to noradrenaline administration. Neurological outcome was determined using the Pittsburgh Cerebral Performance Category (CPC),21 in patients who remained alive at discharge. This scale was assessed at hospital discharge and at 6 months. Data on neurological status at 6 months were collected from medical reports performed by physicians responsible for outpatient monitoring (who were blinded to model prediction results). The primary endpoint was survival with favorable neurological outcome at 6 months, defined as a CPC level of 1 (good recovery) or 2 (moderate disability), whereas CPC 3 (severe disability), 4 (vegetative state), and 5 (death) were considered as an unfavorable outcome.

As a general rule, the criteria for withdrawal of life-sustaining treatment in our centers is bilateral absence of N20 evoked potentials and/or and unreactive malignant electroencephalogram pattern, in a patient who remains comatose ≥ 72hours from ROSC and ≥ 48hours after rewarming.

The study complied with the Declaration of Helsinki and was approved by the Ethics Committee of Hospital General Universitario Gregorio Marañón, Madrid, Spain.

Continuous variables are summarized as median and interquartile range, and differences were analyzed with the Student t test or with the Wilcoxon rank sum test (when not normally distributed). Categorical variables are presented as frequencies and percentages and were compared by the chi-square test or the Fisher exact test.

A multivariate logistic regression model was constructed with data from derivation group by backward stepwise variable selection. Variables with a P value of <0.2 in univariate analysis were considered as candidates for inclusion in the multivariate model. The Hosmer-Lemeshow test was used to determine goodness-of-fit. Possible medical relevant interaction terms between variables were tested. The screening accuracy of the model to identify poor outcome at 6 months (defined as death, vegetative state, or severe disability; CPC 3-5), from data available at hospital admission, was analyzed through odds ratio in multivariable logistic regression, the area under the curve (AUC), sensitivity-specificity coefficients, and the false-positive rate. The final model was evaluated, and external validation was performed using data from the validation group.

Data were collected locally, with removal of any identifying information, and were entered into a FileMaker database (FileMaker Inc, Santa Clara, California, United States). All tests were 2-sided and performed with Stata 12.1 (StataCorp, Lakeway, Texas, United States). Patients with a missing value in one of the variables of interest were reported and excluded from analysis.

Between January 2007 and July 2015, 300 patients were admitted to our institutions with a diagnosis of OHCA, persistent Glasgow Coma Scale score ≤ 8, and were treated with TTM. A total of 56 patients were excluded from the analysis: 33 had no lactate measurement at admission, 16 no information regarding time from collapse to ROSC, and 7 were lost to follow-up. Thus, 244 patients were available for analysis. As previously described, the population was divided into 2 groups: 153 patients in the derivation group, and 91 in the validation group.

Clinical features of the entire patient cohort are outlined in Table 1 and Table 2. There were several differences between the 2 groups: patients admitted in the derivation group were older, had less shockable rhythms, less cardiac etiology, and worse hemodynamic status at admission, a higher rate of out-of-hospital hypothermia, and shorter time to initiation of therapeutic hypothermia than those in the validation group.

Among the patients with an extracardiac cause (39 patients), a diverse group of etiologies was included: 15 were in relation to severe hydroelectrolytic disorders (mainly hypokalemia), 9 patients due to acute respiratory insufficiency of noncardiological origin, 6 patients in relation to acute cerebrovascular disease (all of them died during admission), 6 due to extensive pulmonary thromboembolism and, finally, 3 patients in whom OHCA was associated with cocaine intoxication.

During admission, 129 out of the 244 patients (52.9%) died [81 patients (52.9%) in the derivation group and 48 patients (52.7%) in the validation group]. The most common cause of death was postanoxic brain injury (70.6%), followed by cardiogenic shock (18.8%), infectious complications (5.9%) and other causes in 4.7% of the patients. During the 6-month follow-up period, 4 additional patients died (all of them in the derivation group). Overall survival at 6 months was 45.5% (44.4% in the derivation group, and 47.3% in the validation group). Cumulative survival is shown in Figure 1. Regarding neurological outcome at 6 months, 107 of the 111 survivors (96.4%) had no significant neurological sequelae (98 patients with CPC=1, and 9 patients with CPC=2), while 4 patients had significant neurological damage (3 patients with CPC=3, and 1 patient with CPC=4). By groups, 65 of the 68 survivors in the derivation group (95.6%) and 42 of the 43 survivors in the validation group (97.7%) had no significant neurological sequelae (Figure 2).

Figure 1.

Cumulative survival.

(0.05MB).

Figure 2.

Patient flowchart. CPC, Pittsburgh Cerebral Performance Category; OHCA, out-of-hospital cardiac arrest; T-ROSC, time to return of spontaneous circulation.

(0.18MB).

In the derivation group, 5 variables available at admission were associated with 6-month outcome (death or CPC> 2): Shockable rhythm, Age, Lactate levels, Time Elapsed to ROSC, and Diabetes (SALTED), Table 3. No significant interactions were found between the selected variables. We constructed a receiver operator curve (AUC, 0.90; 95% confidence interval [95%CI], 0.85-0.95) and calculate sensitivity (79.6%) and specificity (84.6%) for the prediction model, with a false-positive rate of 12.5%. This model demonstrated a good fit with a Hosmer-Lemeshow value of 0.97. When we applied this prediction model to the validation group, the receiver operator curve showed an AUC of 0.82 (95%CI, 0.73-0.91), with a sensitivity of 73.5% and a specificity of 78.6% (Figure 3).

Figure 3.

Receiver operator characteristic curves for predictive model using derivation and validation groups. The curves are based on diagnosis-prediction models of poor prognosis at 6 months incorporating 5 variables: Shockable rhythm, Age, Lactate levels, Time Elapsed to return of spontaneous circulation, and Diabetes (SALTED).

(0.09MB).

Although left ventricular ejection fraction at admission was significantly lower in patients with worse evolution (38.7% vs 43.1%; P=.01), this variable was eliminated from the multivariable logistic regression model during its development (backward stepwise variable selection).

The model was designed to estimate the long-term prognosis (6 months) and yet a high percentage of patients died during hospital admission (52.9%), which could be an important source of error. Therefore, the model was also applied to the subgroup of patients who survived at hospital discharge (115 patients). That analysis showed that the model maintained an adequate diagnostic power, with an AUC of 0.86 (95%CI, 0.79-0.93) (Table 1 of the supplementary material and Figure of the supplementary material).

Analysis of the etiology of cardiac arrest showed that patients with extracardiac origin had a much worse outcome (CPC ≤ 2 at 6 months in 12.8% compared with 49.8% in the remaining patients, P <.001). However, when the proposed prognostic model was applied to the subgroup of patients with extracardiac origin, the receiver operator curve maintained an AUC of 0.86 (95%CI, 0.71-1.00). The model was recalculated after exclusion of patients with an extracardiac origin, following the same principles described for the main model (in this case, derivation group of 118 patients and validation group of 87 patients). This subanalysis yielded fairly similar results to those of the main model, in which the same variables were maintained as independent prognostic predictors. The main difference was that the “nonshockable rhythm” variable became much more relevant in the estimation of prognosis: adjusted odds ratio 43.01 (95%CI, 39.43-47.08) vs 14.86 (95%CI, 3.96-55.79) in the main model. These characteristics of this model, as well as its behavior in external validation (AUC, 0.84; 95%CI, 0.75-0.93) when it was applied to the validation group, are detailed in Table 2 of the supplementary material.

The predictive model was obtained with data from 153 patients treated at 2 hospitals and has subsequently been subjected to external validation in a second group of 91 patients admitted to a third hospital. Although there were significant differences between the patients in the 2 groups, the model retained a high diagnostic power.

The usefulness of spectral analysis of fibrillation records in the prognostic assessment of these patients has recently been described.34 This parameter also has the advantage of being available at patient admission, although it was not assessed in our analysis since our study included patients with shockable and nonshockable rhythms.

Probably the most important limitation of this study is the relatively high false-positive rate (12.5%) in a model that aims to determine which patients will have an unfavorable outcome, which means that a few patients may be misclassified by the model as having poor prognosis when their recovery at 6 months might be good. However it is important to emphasize that this estimation should not be taken as a definitive conclusion on the neurological status of the patient, but only as supplementary information to help in initial decision-making and as a support to inform relatives. The study was conducted at 3 large academic centers, and so the results may not be generalizable to other hospitals. Therefore, further studies in different populations are needed to contrast this model. Another limitation is the difficulty of obtaining precise information regarding the time of cardiac arrest, the initiation of resuscitation and its duration, as this information was obtained from prehospital emergency services and witnesses. In survivors of witnessed cardiac arrest (91% of patients in our database), time to ROSC is a good estimation of the duration of circulatory arrest, while in those with unwitnessed cardiac arrest, it may underestimate the true duration of circulatory collapse.