We conducted a retrospective cohort study of all consecutive patients who had suffered cardiac arrest outside the hospital and were subsequently admitted to the adult intensive care unit in the Caen University Hospital from November 2002 to July 2006. Patients who died or awoke within the first three days of admission were excluded from this analysis.

The institutional review board was consulted and considered that this protocol did not require Ethical approval given the observational and non-interventional nature of the study investigating routine care. Clinical variables collected at baseline were: age, sex, medical history (cardiovascular, respiratory, neurologic, metabolic diseases), cause of the arrest (cardiac, respiratory, other or unknown), time between arrest and cardiopulmonary resuscitation (<3 min, 3–5 min, >5 min), cardiac rhythm before cardiopulmonary resuscitation (ventricular fibrillation or tachycardia, asystole, pulseless rhythm), duration of cardiopulmonary resuscitation (<5 min, 5–15 min, >15 min), and scoring of disease severity within the first day in ICU as assessed by admission Simplified Acute Physiology Score type II (SAPS II) [18] and Acute Physiology and Chronic Health Evaluation (APACHE) II score [19].

Clinical procedures performed at days one and three were: pupillary light reflex (present/absent), motor response to painful stimulation (extensor or absent response/other response), corneal reflex (present/absent), tonic-clonic seizures (present/absent) and myoclonus (present/absent).

EEG and SSEPs were routinely performed in our center. For the purpose of this study, a neurophysiologist expert (O.E.) retrospectively reviewed all EEGs to allow similar definitions and increase homogeneity of the material. He was blinded to the clinical features at the time of recordings and to the outcome. EEG was recorded on at least a 10-channels system with needle electrodes using 10–20 international system (Fp1, Fp2, C3, C4, T5, T6, O1, and O2). The EEG patterns were classified at day one and day three according to the classification system of Synek et al [20, 21] [see Additional file 1]. EEG results were dichotomized as malignant and non-malignant, including benign and uncertain patterns. SSEPs were systematically performed the third day after cardiac resuscitation. However, if SSEPs recording was due on a weekend day, the recording was postponed to Monday. SSEps were recorded on Nicolet Viking IV using 6 channels: erb'point; C6sp; C'3 or C'4 controlateral to the stimulated hand and Fpz (ipsilateral ear was used as reference). The two remaining channels served as channels control: C'3 – C'4 (or C'3–C'4) on which N20 amplitude was measured and Fpz-C'3 (or Fpz-C'4) in order to check for long-latency component using larger time-window. Absence of early cortical responses to somatosensory evoked potentials were asserted only if the 3 following conditions were present: (i) correct peripheral (N10) and medullar (N13) component, (ii) no deflexion higher than 0.5 μV on C3–C'4 (or C'3–C'4) (iii) no late component on Fpz-C'3 (or Fpz-C'4). Two groups of patients were defined for statistical analysis according to SSEP results: Group 1, patients with bilateral absence of early cortical responses (unfavorable result of SSEPs) and Group 2, patients with uni- or bilateral presence of early cortical responses (favorable result of SSEPs).

Neurological status at six months after cardiac arrest was recorded using the five grade Glasgow-Pittsburgh Cerebral Performance Category (GP-CPC) scale [22] [see Additional file 2]. Neurological outcome was classified as favorable (GP-CPC 1, 2, and 3) and unfavorable (GP-CPC 4 and 5).

All patients were treated without restriction for the first three days following cardiac arrest. They received standard intensive care management and monitoring, including induced hypothermia as recommended [23, 24]. Propofol, midazolam and sufentanyl were used as standard sedative and were stopped shortly before clinical assessment and EEG recording when this was possible. Hypothermia was induced using a endovascular cooling catheter (Icy™, Alsius, Irvine, CA, USA), inserted in the inferior vena cava via the femoral vein and connected to a cooling device (Coolgard 3000™, Alsius, Irvine, CA, USA), and was maintained during 24 hours. For patients which underwent cooling after resuscitation, clinical variables and neurophysiological tests were performed after rewarning. In our practice, bilateral lack of cortical response to SSEPs leads to active care withdrawal.

Quantitative variables were expressed as means ± standard deviation (SD) with their 95% confidence intervals (CI). Qualitative variables were expressed as percentages with their 95% CIs.

All analyses were replicated for day one and day three variables. Firstly, we investigated the univariate associations between clinical and EEG results (predictors) and outcomes (pejorative results of SSEPs and 6-month poor neurological outcome) using Fisher exact tests. Secondly, we constructed a multivariable model predicting the probability of outcome by performing a backward logistic regression that included variables associated at p < 0.25 in univariate analysis. Thirdly, we computed a score, based on the point system developed by Sullivan [25], using the beta coefficients of all risk factors that were significantly associated with outcomes in the multivariable analysis. The beta coefficients were multiplied by 10 to round them to the nearest integer and then were summed. Fourthly, we plotted Receiver Operating Characteristics (ROC) curves to estimate the capacity of the score to predict outcome using the Area Under Curve (c-index), sensitivity (Se) and specificity (Sp). We used SPSS 14.0 (Chicago, IL) for data analysis. All tests were two-sided and a p-value < 0.05 was considered statistically significant.

Of the 109 consecutive patients resuscitated after a cardiac arrest outside the hospital and admitted to our intensive care unit during the study period, 66 fulfilled inclusion criteria for analysis: 34 (52%) in group 1 (unfavorable result of SSEPs) and 32 (48%) in group 2 (favorable result of SSEPs), as shown in Figure 1. Patient baseline characteristics are presented in Table 1. Fifteen patients (22%) underwent cooling after resuscitation (6 in group 1 and 9 in group 2). Only asystole as cardiac rhythm before CPR (odds ratio [OR] = 0.3; 95% CI = 0.1–1; p = 0.04), and duration of CPR > 15 min (OR = 0.3; 95% CI = 0.1–0.9; p = 0.03) were significantly associated with group 1. The neurological outcome at six months was favorable in 13 (20%) patients, all in group 2 (Figure 1).

In univariate analysis (Table 2), absence of corneal and pupillary reflexes, myoclonus, and extensor or absent motor response to painful stimulation was significantly associated with group 1 at day one and day three. A malignant EEG was also significantly associated with group 1 at day three. In multivariate logistic-regression analysis (Table 2), absence of corneal reflex, myoclonus, extensor or absent motor response to painful stimulation were independently associated with a unfavorable result of SSEPs, at both day one and day three. In addition, absence of pupillary reflex and a malignant EEG were also independently associated with group 1 at day three.

At days one and three, the prediction models included three and five variables, respectively: myoclonus (30 points), extensor or absent motor responses to painful stimulation (17 points), and absence of corneal reflex (10 points) for day one, and extensor or absent motor responses to pain (28 points), myoclonus (25 points), absence of pupillary (21 points) and corneal (14 points) reflexes, and a malignant EEG (11 points) for day three.

As shown in Figures 2a and 2b respectively, the ROC curves for unfavorable result of SSEPs indicated a score ≥ 88.5 points with a positive predictive value of 100% and a sensitivity of 18% (95% CI = 8–27) at day three, but failed to determine a score with a positive predictive value of 100% at day one. The scores with the highest specificity and sensitivity were ≥ 28.5 points (day one; Sp = 75% [95% CI = 65–85], Se = 88% [95% CI = 80–96]) and ≥ 40.5 points (day three; Sp = 84% [95% CI = 76–93], Se = 85% [95% CI = 77–94]), with a positive predictive value of 79% and 85%, respectively. At days one and three, these scores predicted absence of early cortical responses to SSEPs with a false positive rate of 21% (day one) and 15% (day three).

As shown Figures 3a and 3b, the ROC curves for unfavorable outcome determined a score ≥ 52 points at day one and a score ≥ 69 points at day three, with a positive predictive value of 100% and a sensitivity of 35.8% (95% CI = 24–47) and 32% (95%CI = 21–43), respectively. An unfavorable outcome was predicted in more than a quarter of the studied population with no false positives. The scores with the highest specificity and sensitivity were ≥ 22 points (day one; Sp = 92% [95% CI = 86–99], Se = 79% [95% CI = 70–89]) and ≥ 32 points (day three; Sp = 85% [95% CI = 76–93], Se = 85% [95% CI = 76–94]) with a positive predictive value of 98% and 96%, respectively. The presence of two or more predictors at day one or day three predicted an unfavorable outcome with a false positive rate of 2% and 4%, respectively.

Among 32 patients with a favorable result of SSEPs, 19 patients had an unfavorable neurological outcome at six months (Figure 1). As shown in Figure 4, the scoring system with 100% specificity at day one (score ≥ 52 points) and at day three (score ≥ 69 points) predicted death or vegetative state in two and one patients, respectively. At day three, the score (≥ 32 points) with the highest sensitivity and specificity accurately predicted an unfavorable outcome in 9/19 (47%) patients, but failed in two.