Our data suggest that a Thompson score of ≥ 7 at age 3–5 hours identifies all infants with an abnormal 6-hour aEEG with a specificity of 67%. The presence of moderate-severe encephalopathy at the same age had similar specificity and sensitivity but failed to identify one infant with an abnormal 6-hour aEEG. Individual clinical signs previously used as entry criteria in clinical cooling trials had low sensitivity and/or low specificity for an abnormal 6-hour aEEG. A Thompson score ≥ 5 at 1 or 3–5 hours identified all infants with an abnormal aEEG at 3 and/or 6 hours but with low specificity. A Thompson score of ≥ 7 at 3–5 hours predicted moderate-severe encephalopathy presenting within 72 hours after birth. However, the hypothermia and the sedating medication received by 83% (40/48) of the infants with moderate-severe encephalopathy may have exaggerated the abnormal signs.
We defined an abnormal aEEG as one that qualifies an infant for cooling according to published protocols [15–17]. This definition allows inclusion of infants with DNV. In a retrospective study of aEEG in cooled vs. normothermic infants, Thoresen et al. defined a normal aEEG as one with CNV or DNV . However, following their observation that three of the nine normothermic infants with DNV had a severely abnormal outcome vs. none of the eight cooled infants with DNV, they conceded it was reasonable to cool infants with DNV at 3–6 hours.
The predictive values for an abnormal outcome are lower for an aEEG at 3 vs. 6 hours  and an early abnormal assessment might inappropriately select infants for cooling therapy who were destined to be normal without cooling. Post-hoc analysis of the subgroup of infants with 1-hour assessments (Table 6) showed that an earlier Thompson score was more sensitive but less specific in predicting an abnormal aEEG at 3 and/or 6 hours (and therefore the need for cooling). If a single Thompson score threshold to be used at 1 or 3–5 hours is to be defined, a score of ≥ 7 had the best overall combination of high sensitivity and high diagnostic odds ratios of 35 and 26 respectively. The Thompson score of ≥ 5 at 1 or 3–5 hours identified all infants with an abnormal aEEG at 3 and/or 6 hours (and all the infants who were cooled). Although this threshold may be suitable to identify infants for referral or further assessment, the low specificity and LR makes it unsuitable as a sole indication for cooling.
Several cooling trials required the presence of a decreased level of consciousness and/or hypotonia as minimum evidence of moderate or severe encephalopathy before further assessment [15–17, 24, 25]. In our study, a significant proportion of the infants without either a decreased level of consciousness or hypotonia at 3–5 hours had an abnormal 6-hour aEEG. Our data suggest that the absence of specific individual clinical signs should not be grounds for excluding infants from aEEG assessment or cooling therapy.
Shalak et al. studied 50 infants with suspected intra-partum hypoxia . They determined the ability of the moderate-severe encephalopathy at age 5±3 hours and an abnormal fronto-parietal aEEG acquired within an hour of the examination, to predict moderate-severe encephalopathy persisting until the fifth day. The clinical assessments were performed slightly later than in our study and the aEEG position was fronto-parietal, but similar to our study, they found that 23% of infants without all the criteria for moderate-severe encephalopathy had an abnormal aEEG and progressed to moderate encephalopathy persisting on the fifth day. The sensitivity and specificity of early moderate-severe encephalopathy to predict moderate-severe encephalopathy persisting to the 5th day were both 78%, but the combination of an abnormal aEEG and encephalopathy including clinical signs of both mild and moderate encephalopathy increased the specificity to 94%. In our study, only one of the infants (6%) without all the criteria for moderate-severe encephalopathy at 3–5 hours had an abnormal aEEG at 6 hours and developed persistent moderate-severe encephalopathy by the fifth day. By comparison, none of the infants with a Thompson score < 7 had an abnormal aEEG at 6 hours, and moderate-severe encephalopathy did not persist to the fifth day in any of those infants. One infant with a Thompson score < 7 was cooled, but data from a secondary analysis of the CoolCap cooling trial data suggest that the hypothermia is unlikely to have influenced the improvement in grade of encephalopathy in this infant .
Sarkar et al. have questioned whether a 6-hour aEEG should be used to identify infants for cooling . In a retrospective analysis, they reported that 13 of 24 infants with a normal aEEG had abnormal magnetic resonance imaging (MRI), but the rates of abnormal outcome were similar between the cooled and normothermic groups. In a prospective study, Shankaran et al. found that a fronto-parietal aEEG at < 9 hours did not enhance the predictive value of HIE grade at < 6 hours . Twelve of 71 infants with moderate HIE had a CNV background at < 9 hours and three of these infants had an abnormal outcome. The data of both Sarkar and Shankaran suggests that a normal aEEG in the first 6–9 hours does not guarantee a normal outcome. However it is still unclear whether cooling infants with a normal aEEG is beneficial.
The Thompson score ≥ 7 at 3–5 hours identified more infants with moderate-severe encephalopathy than did the 6-hour aEEG, but the additional infants identified by the Thompson score were predominantly those who were not cooled. The presence of CNV or DNV background voltage on aEEG at 24 hours is significantly associated with normal outcomes in both cooled and normothermic infants[23, 29]. The finding of a normal 24-hour aEEG, very low or normal Thompson scores by day 7, and a median discharge age of 5 days, in the infants with moderate-severe encephalopathy who were not cooled suggests that these infants did not require cooling. Thus, although clinical assessment with either MSEG or the Thompson score identified significantly more infants with moderate-severe encephalopathy, this may not be clinically significant.
This study has several limitations. The exclusion of sick infants may alter the sensitivity and specificity of the Thompson score – the findings of this study cannot be applied to the infants whom were excluded. It is a significant weakness of the study that we did not determine the kappa coefficient for the multiple assessors (raters) in this particular study. However, all clinicians were trained, the kappa coefficient for two raters of the Thompson score in a previous study at one of the study sites (Groote Schuur Hospital) is known to be high , and the use of multiple raters replicates the setting in which this particular diagnostic approach would ultimately be implemented. The lack of data from age 1 hour in all infants compromised the strength of our conclusions regarding very early assessments. Although the primary objective was to study the prediction of an abnormal 6-hour aEEG, the availability of MRI or long-term follow up data would have allowed further validation and interpretation of the threshold Thompson scores, particularly in the infants with moderate-severe encephalopathy who were not cooled. A further important limitation is that the use of phenobarbitone and morphine may have resulted in overdiagnosis of moderate-severe encephalopathy in the infants who were cooled.
The strengths of our study are that we prospectively recruited infants with all grades of HIE and we obtained at least one clinical and one aEEG assessment by age 6 hours. We compared previously published neurological assessments to a validated aEEG assessment method as a gold standard. We blinded the clinicians’ assessments by comparing their clinical assessment with a later aEEG recording and neonatologists who were blind to the clinical assessments subsequently read the aEEG recordings.