Seizures in 204 comatose children: incidence and outcome

Purpose Seizures are common in comatose children, but may be clinically subtle or only manifest on continuous electroencephalographic monitoring (cEEG); any association with outcome remains uncertain. Methods cEEG (one to three channels) was performed for a median 42 h (range 2–630 h) in 204 unventilated and ventilated children aged ≤15 years (18 neonates, 61 infants) in coma with different aetiologies. Outcome at 1 month was independently determined and dichotomized for survivors into favourable (normal or moderate neurological handicap) and unfavourable (severe handicap or vegetative state). Results Of the 204 patients, 110 had clinical seizures (CS) before cEEG commenced. During cEEG, 74 patients (36 %, 95 % confidence interval, 95 % CI, 32–41 %) had electroencephalographic seizures (ES), the majority without clinical accompaniment (non-convulsive seizures, NCS). CS occurred before NCS in 69 of the 204 patients; 5 ventilated with NCS had no CS observed. Death (93/204; 46 %) was independently predicted by admission Paediatric Index of Mortality (PIM; adjusted odds ratio, aOR, 1.027, 95 % CI 1.012–1.042; p < 0.0005), Adelaide coma score (aOR 0.813, 95 % CI 0.700–0.943; p = 0.006), and EEG grade on admission (excess slow with >3 % fast, aOR 5.43, 95 % CI 1.90–15.6; excess slow with <3 % fast, aOR 8.71, 95 % CI 2.58–29.4; low amplitude, 10th centile <9 µV, aOR 3.78, 95 % CI 1.23–11.7; and burst suppression, aOR 10.68, 95 % CI 2.31–49.4) compared with normal cEEG, as well as absence of CS at any time (aOR 2.38, 95 % CI 1.18–4.81). Unfavourable outcome (29/111 survivors; 26 %) was independently predicted by the presence of ES (aOR 15.4, 95 % CI 4.7–49.7) and PIM (aOR 1.036, 95 % CI 1.013–1.059). Conclusion Seizures are common in comatose children, and are associated with an unfavourable outcome in survivors. cEEG allows the detection of subtle CS and NCS and is a prognostic tool.

Abstract Purpose: Seizures are common in comatose children, but may be clinically subtle or only manifest on continuous electroencephalographic monitoring (cEEG); any association with outcome remains uncertain. Methods: cEEG (one to three channels) was performed for a median

Introduction
Seizures may be difficult to detect in unconscious children, particularly if the episode is partial or subtle or when neuromuscular blocking (NMB) agents are used during mechanical ventilation [1]. Electroencephalography (EEG) may be helpful in this situation, but there are a range of possibilities when observing electroencephalographic seizure (ES) activity. For example, in the child who is able to move, if there is no convulsive activity, then the state could be defined as non-motor, non-convulsive seizures (NCS). Alternatively, there may be subtle clinical features of epileptic nystagmus or facial myoclonus [2][3][4]. Monitoring with continuous EEG (cEEG) in neonates and older children is sensitive for detection of ES, with few false-positives [5][6][7]. ES may be frequent, with a prolonged time-course [8][9][10][11][12][13][14][15][16], but their significance is unknown, so it not clear whether cEEG is advisable or anticonvulsant treatment should be given for subtle seizures, NCS and/or ES in comatose patients or those receiving NMB agents. In order to answer this question, we need to know whether there is a relationship between these events and eventual outcome, independent of underlying aetiology.
The purpose of this study was to examine the prevalence of ES in comatose children and to determine any association of subtle seizures or ES with mortality and morbidity.

Ethical approval
The research ethics committees of Guys Hospital Medical School and the Kenyan Medical Research Institute gave ethical approval for the work undertaken in their respective institutions. At the time of this observational study, individual written informed consent was not required to report clinically indicated monitoring or follow-up, but the investigators discussed the research with parents or guardians and obtained verbal consent to monitor EEG activity and to follow-up the patients as part of clinical care.

Study populations
Two prospectively recruited consecutive cohorts of unconscious children without previous acute neurological injury underwent cEEG monitoring. In the first (Guy's Hospital, London) all of the children were mechanically ventilated and received pancuronium for NMB. In the second (Kenya Medical Research Institute, Kilifi, Kenya) none of the children underwent mechanical ventilation since facilities were not available, and so they could be assessed clinically. Any child older than 1 month with an acute encephalopathy and a summated Adelaide coma score (ACS) [17] less than normal for age was eligible, and as there was sufficient monitoring equipment available, all had cEEG monitoring until they were able to localize a painful stimulus ([9 months) or flex to pain (\9 months) or death.

Clinical observations
Diagnosis followed standard definitions (see ''Aetiology of coma''). In unventilated patients, level of consciousness was assessed on admission and then 6-hourly using the ACS until the child was drinking. In ventilated patients, ACS was obtained on admission. Thereafter, clinical examination or detection of clinical seizures (CS) was usually precluded by the use of medication for sedation and NMB. In all patients, clinical signs suggestive of seizures, such as jerking, or increases in blood pressure or heart rate, were noted. In patients without mechanical ventilation, if ES were noted on the cEEG, the bedside attendant undertook a careful clinical examination. Evidence of subtle CS (e.g. changes in pupil response, blood pressure, or heart rate, or clonic movements of the digits, face or eyes) was recorded.

Electroencephalography
Two EEG devices and montages were used in these studies. In 52 patients managed at Guy's hospital, the output from a three-channel Oxford Medilog EEG machine [14,15] with disc electrodes placed at F4-P4 and F3-P3 and C3-C4 on the International 10-20 system (reference electrodes just behind the hairline) was displayed in real time at the bedside (Siemens monitor, Erlangen, Germany) and the raw EEG was replayed offline through a commercial system (Oxford Systems, Oxford, UK). This equipment did not include amplitude and frequency integration, and staff found it difficult to recognize seizures ( Fig. 1) in real time. Therefore, in the remaining 152 patients, including all 46 Kenyans, a oneor two-channel cerebral function analysing monitor (CFAM) [7,8] was used, calibrated using a conventional multichannel EEG. The disc electrodes were placed at P3-P4 (reference electrode just behind the hairline) for single channel monitoring (n = 118), or P3-P4 and F4-P4 for bilateral monitoring (n = 34) with data displayed on a paper trace with raw EEG displayed automatically at least every 10 min. A button could be pressed by attending staff to record raw EEG during increases in amplitude representing possible seizures (Fig. 2). Electrode impedance was maintained below 2 kOhm and mains artefact as low as possible [11,14].

Clinical management of seizures
The attending physician decided on the prescription of prophylactic or treatment doses of antiepileptic drugs. Rectal diazepam (0.5 mg/kg) or paraldehyde (0.2 ml/kg) was given to stop seizures (CS or ES) lasting longer than 5 min. Some children were prescribed prophylactic parenteral phenytoin (18 mg/kg) or phenobarbital (15-20 mg/kg) before cEEG commenced. Others received these anticonvulsants only when there was more than one CS or ES. If the episode continued, then repeat doses were given after 6-12 h.

Severity of illness and outcome assessment
The Paediatric Index of Mortality (PIM) [18] and ACS on admission were recorded. There was active surveillance for death during the study. Treatment was only withdrawn in the event of brain death. All children who survived initial hospitalization were followed-up at 1 month after admission to hospital by one of the paediatric neurologists in the study, who undertook a detailed history from the parents about any change from premorbid cognition and behaviour, as well as a detailed neurological examination. Using this information, outcome was classified on the Glasgow Outcome Scale by an investigator (B.G.R.N.) blinded to the admission clinical and cEEG data. Outcome was dichotomized: children who remained in a vegetative state, or had a movement disorder precluding ambulation, or severe learning difficulties, were classified as having an unfavourable outcome, with the remainder as having a favourable outcome. Where possible, and always if the neurological examination was abnormal, defined as reduced level of consciousness, or if abnormalities of motor, visual, auditory or cognitive function were found, further follow-up was arranged and longer term outcome assessed using the same methodology.

Analysis of ES episodes
An ES was defined as the presence of rhythmical spikeand-wave activity for more than 15 s, and was considered subclinical if there were no clinical manifestations. For the CFAM data, all traces were analysed independently for the presence of ES by S.G.B. and F.J.K.; consensus Subsequently the EEG became isoelectric and the child died a brain death

Statistical analyses
Differences in demographics and outcome measures between the three groups (UK neonates, UK infants and children, Kenyan infants and children) were tested using one-way parametric or Kruskal-Wallis ANOVA or the chisquared test as appropriate. The hypothesis test for proportions was used to compare prevalences. Logistic regression models were used to investigate potential predictors of two outcomes: death in the first month and unfavourable outcome at 1 month amongst the survivors, including clinically important potential predictors: cohort (UK or Kenya), aetiology, PIM score, ACS and EEG grade on admission, prophylactic anticonvulsant administration with benzodiazepines and/or phenobarbitone, CS at any time, and the presence (yes/no), number, total duration and single longest duration of ES. For prediction of unfavourable outcome, the duration of EEG monitoring was also explored as a potential confounder. Unadjusted and adjusted odds ratios (OR) with 95 % confidence intervals (CI) are presented in the tables. Cerebral malaria was chosen as the reference because seizures and coma are common, while outcome is variable, although the malaria parasite does not directly invade the brain parenchyma. Variables were added to multivariable models according to statistical significance in a forward step-wise fashion. All variables retained led to a significant change in log likelihood. Variables not initially significant were investigated after adjustment for other factors. The significance level for inclusion was primarily set at p \ 0.05.

Aetiology of coma
Aetiological categories included: cerebral malaria (defined as Plasmodium falciparum asexual parasitaemia in a normoglycaemic patient unable to localize a painful stimulus at least 6 h after last seizure [19]), meningitis (bacteria identified in the cerebrospinal fluid by microscopy, culture or antigen), encephalitis (febrile illness with previous viral infection or pleocytosis and no bacteria), Reye-like syndrome (coma with high blood ammonia), hypoxic-ischaemic encephalopathy (witnessed cardiopulmonary arrest), head injury (evidence of trauma to the head), hypertensive encephalopathy (coma with severe hypertension), thrombotic thrombocytopoenic purpura/haemolytic uraemic syndrome (coma associated with fragmented or burr red cells and thrombocytopaenia), systemic vasculitis (coma associated with laboratory findings diagnostic of autoimmunity), intracerebral haemorrhage (spontaneous coma with intracerebral bleeding on CT), drug toxicity and cavernous sinus thrombosis (  Table 2). In univariable logistic regression, in addition to absence of CS at any time, death was associated with the following findings on admission: aetiology, higher PIM score, lower ACS and lower cEEG grade (Table 2). These five variables were entered into the multivariable model.   (Table 3) in the ventilated cohort, in those with longer duration of monitoring, and in those with hypoxic-ischaemic aetiology, lower PIM score, lower cEEG grade, CS at any time, ES at any time, and larger number and longer duration of ES activity (Fig. 3) and longest duration ES. No child had a favourable outcome when there had been more than 139 seizures, or a total duration of ES of more than 759 min, or an individual ES of more than 360 min. There was no significant association with age, sex or the use of anticonvulsant prophylaxis. In the final model, the presence of ES and PIM score remained in the model for predicting unfavourable neurological outcome (Table 3) independent of cohort and duration of EEG monitoring. None of the other variables was statistically significant after taking these variables into account. Longer term outcome was available for all 71 UK survivors at a median of 6 months (range 1.5-144 months) and for 13 of 40 Kenyan survivors, including all five with an unfavourable outcome at 1 month, at a median of 12 months (range 1.5-40 months). Of the 29 with an unfavourable outcome at 1 month, on final follow-up at a median of 6 months (range 1.5-144 months), one had improved to moderate handicap, eight had died and the remainder stayed in the same outcome category. Although four of those with a favourable outcome at 1 month subsequently died, none deteriorated neurologically when assessed at a median age of 6 months (range 1.5-48 months).
In a multivariable model for prediction of death which included ES, the odds ratio and confidence intervals changed only marginally. Adding the predictors of death to a multivariable model for prediction of outcome in survivors did not change the model and minimally altered the odds ratio and confidence intervals for the effect of ES.

Discussion
In this study ES occurred in over one-third of comatose children, and the prevalence of ES was related to the presence of CS and the underlying aetiology. Our data on prevalence of ES is consistent with those of previous studies of CS in unconscious adults and children [20][21][22][23][24][25][26][27][28][29][30][31][32] and of ES in paediatric [10,33,34] and adult intensive care [35], but not that of a recent study which detected ES in only 7 % of patients [36]. Independently of aetiology, PIM, ACS and the presence of CS, normal background cEEG activity predicted survival, while the presence of ES activity predicted an unfavourable outcome in survivors, although, as in an adult study [37], good outcomes may be seen after very prolonged seizures.
Most of the children had CS before EEG monitoring, but five (2 %), all ventilated, had only ES, i.e. NCS. This is a similar proportion to those found in children and adults who were in nonconvulsive status epilepticus on EEG during the first 3 days after presentation in coma [9,38]. Among the nonventilated children, all who had ES had CS. It would seem prudent to monitor EEG to detect NCS in children who remain unconscious or in children with an acute encephalopathy and CS requiring NMB agents for ventilation [9], or in the absence of a reliable history, e.g. after non-accidental injury.
These results must be interpreted cautiously. The criteria for admission to the high-dependency unit and PICU differed, although our cohort had a variety of diagnoses, so we could assess whether the effect of seizures was independent of aetiology. Since we monitored only one to three channels in these cohorts, a proportion of seizures may have been missed [5]. Patients with traumatic brain injury or intracranial haemorrhage secondary to arteriovenous malformation may have unilateral discharges, while those with status epilepticus in the context of polymerase-c mutations [39] or posterior reversible encephalopathy syndrome [40] typically have occipital involvement. These conditions should be borne in mind when the choice of electrode placement for longer term single or dual channel EEG monitoring is decided. The use of cEEG should be evaluated by a comparison between standard EEG and cEEG at the first opportunity. Although detection of any seizure in a patient may not be better, two or more channels may detect more seizures, particularly in unilateral disease [41]. Further studies comparing one-and two-channel amplitude integrated displays with raw EEG are warranted but as seizure detection is the priority, there may be an advantage in setting up a simple single-channel amplitude integrated EEG monitor with raw EEG display on admission, as ES are detected in the first hour in the majority of patients with seizures [33]. In our data, there was no evidence that monitoring three channels with the Oxford Medilog had any advantage over monitoring one or two channels with the CFAM which displays raw EEG regularly and on demand in addition to the advantage of easier review in real time allowed by the amplitude integrated display [10]. The effect on outcome of ES not manifest as rhythmical spike-and-wave detectable with limited channel EEG monitoring should be the subject of prospective studies with multichannel cEEG. We found an association between CS with survival, at face value surprising but previously documented after head injury [42], perhaps because CS precipitate admission. ES did not predict death either, perhaps because death is related to acute intracranial hypertension independent of seizures. We were only able to determine outcome in all the children at 1 month, and the number of surviving patients was relatively small for multivariable logistic regression. Despite these limitations, our data demonstrate an association between ES and unfavourable neurological outcome in survivors independent of age, aetiology and cohort. Further prospective studies with long-term follow-up are required, however, before definitive recommendations on monitoring and treatment can be made. Routine [3] or cEEG [9] with video recording detects NCS or nonconvulsive status epilepticus in at least onethird of PICU patients presenting with CSE. There is concern that ongoing ES activity is under-recognised in these patients, who are usually sedated and sometimes given NMB agents for ventilation [3,9]. In children in coma, Jette et al. [9] suggested excluding NCS even if there are no CS, and our data provide further evidence for this recommendation. However, multichannel cEEG is labour-intensive and expensive, treatment protocols have not been defined, and there are few data on any association with neurological outcome in coma with acute symptomatic CSE. Further research looking at the reliability of one-or two-channel cEEG monitoring devices in diagnosing NCS is required as the technical issues are resolved and our data add to the evidence that outcome may be predicted on the PICU [10].