Our study suggests that a gray-whitematter ratio (GWR) below 1.16 may be a specific parameter for prediction of poor outcome in patients treated with hypothermia. This corroborates previous findings that were obtained in normothermia and mixed patient cohorts (normothermia and hypothermia). None of 23 patients with a GWR < 1.16 survived with good outcome.
The majority of patients with GWR < 1.16 had at least one other parameter indicative of poor outcome (NSE > 97 μg/L and/or bilateral absent SSEP). Therefore, GWR did not substantially increase the sensitivity for poor outcome detection, if one accepts poor outcome prediction by a single parameter. However, because individual patients may survive despite poor outcome prediction by single parameters, it has been argued that poor outcome prediction should be based on a multi-parameter approach [
6]. We demonstrated in a subgroup of 62 patients who had received NSE, SSEP and CCT that a GWR < 1.16 increased the sensitivity of poor outcome detection (from 43% (17/40) to 53% (n = 21/40)) if at least two pathologic test results were required for poor outcome prediction. These results suggest that GWR is a useful additional parameter that may increase the level of certainty for poor outcome prediction in comatose patients after cardiac arrest.
GWR cut-off value and timing of CCT
Our cut-off value of 1.16 is similar to previously published values. Torbey et al. found a GWR < 1.18 to be 100% specific of poor outcome [
11]. The authors obtained GWR at the basal ganglia level in normothermic patients. None of the eleven patients with GWR < 1.18 survived. In the largest cohort so far by Metter et al., a GWR < 1.15 was found to be 100% specific for poor outcome [
17]. CCT was obtained within 24 hours and the majority of patients with GWR < 1.15 had been treated with hypothermia. Metter et al. concluded that a GWR < 1.2 had the best overall predictive value, but this threshold misclassified two patients. A of GWR < 1.2 in our patient cohort would have misclassified six patients as having poor outcome. Hence, our data suggest that a threshold of 1.2 can be associated with a substantial number of erroneous poor outcome predictions.
In the study by Metter et al. the number of patients who survived with little or no neurological deficit was small [
17]. This outcome distribution may have decreased the likelihood of detecting rare cases that survive with good neurological outcome despite a low GWR. In our study, the proportion of patients surviving with little or no neurological deficit was higher (38%, n = 37).
Inamasu et al. reported a high sensitivity and specificity of the ‘loss of boundary’ (LOB) sign and ‘sulcal effacement’ sign obtained prospectively very early after cardiac arrest [
16]. Only one of 52 patients with LOB sign and none of 20 patients with sulcal effacement sign survived with good outcome. These data corroborate the high specificity of early CT signs for poor outcome prediction.
Wu et al. investigated a semi-automated procedure, which used co-registration of CT to a brain atlas and determination of Hounsfield units for several brain regions [
15]. In a large cohort, the authors demonstrate that whole brain Hounsfield units obtained from early CCT (most within 24 hours) improved the predictive accuracy of a clinical examination. In this study, only 12% of patients had received hypothermia treatment, and therefore results should be interpreted with care for this patient group.
The majority of CCTs in our study were performed on admission, frequently within a few hours after cardiac arrest. Metter et al. did not find a significant association of attenuation with time since cardiac arrest within a narrow time frame of 24 hours [
17]. The distribution of GWR over time in our study, however, suggests that GWR may decrease over the days following cardiac arrest for patients with severe hypoxic-ischemic encephalopathy (Figure
2). Similarly, Wu et al. found a significant decrease in the density of the putamen for CT performed within 24–72 hours as compared to within the first 24 hours after cardiac arrest [
15]. No difference was found for other regions, but only few CTs (4%) were performed after day two, hence the power to detect a difference was low.
In our study, sensitivity of poor outcome detection by GWR < 1.16 was 23% for CCT performed within six hours, but increased to 60% for CCT performed between six hours and seven days. This finding, however, may be biased by patient selection and should be interpreted with care. Nonetheless our data argue for a prospective study investigating the prognostic value of GWR beyond six hours after cardiac arrest.
Since many CTs in our study were performed within the first few hours following admission, it is unlikely that hypothermia had substantially influenced the GWR in this subset of patients. However, as hypothermia changes the course of pathophysiological events leading to definite tissue damage in the days after cardiac arrest, it is still possible that hypothermia influences the relationship between early GWR and neurological outcome. Survival with good outcome may be possible in hypothermia patients despite a more severe early injury. Hence, re-evaluation of the prognostic value of GWR for hypothermia patients is mandatory, before GWR can be integrated into pathways for management of these patients.
Comparison with other outcome parameters
The lower sensitivity of GWR < 1.16 (35% as compared to 53% for NSE and 55% SSEP) and the fact that the vast majority of patients with GWR < 1.16 had either high NSE levels or bilateral absent SSEP may suggest that GWR < 1.16 identifies patients with more severe hypoxic-ischemic encephalopathy than NSE or SSEP. Most of the CCTs in our study were performed within a few hours, but SSEP and NSE were obtained around day three after cardiac arrest. The rather low sensitivity of GWR for the entire study cohort could therefore also be explained by the early time point. Indeed, our data suggest that the sensitivity of GWR increases substantially if CCT is performed > 24 h after admission.
If one accepts a prediction approach that uses a single parameter for poor outcome prediction, the combination of NSE and SSEP increases the sensitivity for detection of poor outcome. Apparently, a relevant number of patients with poor outcome are detected by NSE but not by SSEP and vice versa. If NSE and SSEP are performed, GWR < 1.16 did not significantly increase the number of patients for which poor outcome was detected. However, if one prefers a prediction approach that requires a second confirmatory parameter, GWR has an additional predictive value.
Limitations of our study
Due to the retrospective design of our study, we could only investigate the subset of patients who had received CCT (n = 111, i.e. 31% of all cardiac arrest patients). CCT was generally ordered by treating physicians if the cause of the cardiac arrest was deemed uncertain or if intracranial complications were suspected. Our results should therefore be extended to the entire cardiac arrest population with care and corroboration by a prospective study avoiding selection bias is desirable. The same caveat applies to the comparison of different prognostic parameters, which could only be performed in the subset of patients who had received NSE, SSEP and CCT.
The results of NSE, SEP and CCT were available to the treating physicians and influenced decisions on limitation of treatment. Therefore, we cannot exclude the possibility of a self-fulfilling prophecy. GWR was calculated for this study and not known to treating physicians. However, reduced GWR reflects brain edema and treatment decisions were influenced by CT if brain edema was reported by the radiologist. As a number of recent studies have indicated limited reliability of prognostic parameters in hypothermia patients, treatment withdrawal within the first days after cardiac arrest is restricted to patients with confirmed brain death at our institution. In the remainder, treatment is continued up to day seven [
19]. Hence, a major effect of a self-fulfilling prophecy from treatment withdrawal within the first days after cardiac arrest is unlikely in our study. Nonetheless, unexpected late recovery may occur in individual patients [
5], and we cannot fully rule out that treatment withdrawal based on over-interpretation of prognostic parameters has prevented the detection of such exceptional cases.