Impact of hypothermia on predictors of poor outcome: how do we decide to redirect care?

SUMMARY

Therapeutic hypothermia is now considered the standard of care for neonates with neonatal encephalopathy due to perinatal asphyxia. Outcomes following hypothermia treatment are favorable, as demonstrated in recent meta-analyses, but 45–50% of these neonates still suffer major disability or die due to global multi-organ injury or after redirection of care from life support due to severe brain injury. The ability to determine which patients are at highest risk of severe neurologic impairment and death and those in whom redirection of care should be considered is limited. This is especially true in the first few days after birth and in situations where the brain might be more significantly affected than other organ systems, making it difficult to discuss redirection of care. Clinical history, neurologic examination, serum biomarkers, neurophysiology [amplitude-integrated electroencephalography (aEEG) or EEG], near-infrared spectroscopy, and magnetic resonance imaging have all been studied as predictors of severe neurologic injury and poor outcome, although none is 100% predictive. Serial evaluation over time seems to be an important element to facilitate discussion regarding anticipated poor prognosis and decision-making for transition to comfort care. Thus far, brain monitoring in the form of aEEG and conventional EEG seem to be the best objective tools to identify the highest-risk patients. A delay or lack of recovery of the aEEG background during hypothermia treatment is an established important predictor of poor outcome (death or disability). This paper highlights the prognostic indicators that have been considered and focuses on aEEG as an important predictor of death or severe disability, which may facilitate conversations regarding redirection of care.

1. Introduction

Therapeutic hypothermia (TH) for the treatment of neonatal encephalopathy due to perinatal asphyxia is now considered standard of care and should be offered to qualifying neonates [1]. Multiple randomized controlled trials have demonstrated efficacy in the reduction in the risk of death or moderate–severe disability at 18–24 months of age [2–4]. School age outcomes from the major trials are also now being published and demonstrate that treatment with hypothermia, in at least one of the major trials, has sustained benefit at 6–7 years of age [5,6]. Unfortunately, despite treatment with therapeutic hypothermia, in the randomized controlled trials (RCTs) and in clinical practice, a significant number of neonates do not survive (as high as 30%) and 45–50% suffer the composite outcome of death or major disability [7]. Identification of those neonates that are “too severe” to benefit is difficult. Clinicians have the arduous task of differentiating those who may survive with moderate impairments from those who will survive with severe impairments or who are unlikely to survive to hospital discharge. Discriminating these groups of newborns is complicated by the fact that hypothermia has impacted the ability of key markers of severity of encephalopathy (neurologic exam) to predict outcome and changes the time-frame in which tools such as amplitude-integrated electroencephalography (aEEG) are most predictive of outcome. Magnetic resonance imaging (MRI) is known to retain its predictive abilities but this is sometimes delayed – presumably due to clinical instability or poor access to imaging – by as much as two weeks after birth, which does not help in conversations about possible outcomes with parents in the first few days of life. Clinicians likely use these “predictive tools” as criteria for entering end-of-life discussions with parents, as neonates rarely die while actively receiving intensive care. Although poorly reported in the literature, it appears that the majority of deaths in the setting of severe hypoxic–ischemic encephalopathy (HIE) can be attributed to end-of-life decisions and redirection of the goals of care to comfort measures [8,9]. This article reviews the risk factors for poor outcome – including death in the newborn period, many of which the clinician considers during the course of treatment – and presents the findings of a subset of an international cohort study of newborns treated with hypothermia as part of standard of care in which these risk factors were evaluated as predictors of poor outcome and death in the newborn period.

2. Use of clinical examination as a predictor of outcome

Clinical examination is critical in properly identifying newborns who qualify for TH. All of the randomized trials used some form of the Sarnat staging system in order to identify eligible patients. In 1976, Sarnat and Sarnat published an article in which they serially evaluated a small cohort of 21 neonates with perinatal asphyxia and encephalopathy [10]. Using a standardized neurologic exam and EEG, initially performed at 12–24 h after birth, and then performed serially during the hospitalization, they were able to classify neonates into three stages, describe the duration and evolution of each stage of encephalopathy, and determine the association between stage and evolution of encephalopathy and outcome at 6–12 months of age. In this very small cohort, neonates with Stage 2 encephalopathy who normalized their neurologic exam and EEG within 5 days (n = 8) tended to have a good outcome at 6–12 months of age, and those with features of stage 3 (n = 5) all had a poor outcome of death or severe impairment. The neurologic exam used in this landmark paper was adapted/modified in order to select patients for enrollment into the RCTs of TH. In order to be eligible for enrollment in an RCT, the neurologic exam within 6 h of birth needed to be consistent with Stage 2 or 3 encephalopathy. However, it is important to note that the exact timing of the initial exam in the Sarnat paper was not presented but was stated to occur at the time of admission. There are two important and overlooked findings of the Sarnat paper: (1) Some neonates were in Stage 1 for >6 h and then progressed to Stage 2 encephalopathy. These neonates would not have been enrolled in the RCTs and it is unclear what their risk of long-term neurodevelopmental impairments would be. (2) The study had relatively short-term follow-up ranging from 6 to 12 months of age, but only two of the surviving subjects were seen at 12 months of age.

Hypothermia has impacted the ability of stage of encephalopathy at presentation to predict outcome. Secondary analyses of the National Institute of Child Health and Human Development (NICHD) trial and the CoolCap trial both identified that the predictive ability of the initial stage of encephalopathy was altered by treatment with hypothermia and that improvement in stage of encephalopathy during treatment was important in identifying patients who were likely to have a good outcome [11,12]. In addition there appears to be an interaction between encephalopathy stage at the end of treatment and treatment itself, as the predictability of moderate encephalopathy at the completion of treatment was altered by hypothermia with treated infants with persistent moderate encephalopathy on day 4 doing better than those who were not treated. Specifically, in the CoolCap trial, cooled neonates with a persistent moderate encephalopathy (HIE grade 2) at day 4 tended to have a favorable outcome as compared to non-treated infants (24 of 31 subjects vs 10 of 29 subjects, P = 0.002) (Fig. 1). Improvement in grade of encephalopathy from grade 2 to grade 1, regardless of treatment, seemed to confer a favorable outcome compared to those whose stage of encephalopathy deteriorated [11]. In an analysis of serial modified Sarnat staging of newborns enrolled in the NICHD trial, the initial stage at <6 h was predictive of death or disability, but, in those who survived to complete 72 h of treatment, initial stage at <6 h was no longer a statistically significant predictor of outcome when the stage at 72 h was accounted for [12], again suggesting that evolution of stage during treatment is more important than the initial stage of encephalopathy. In predictive models, time to improvement in stage and time to reaching no or mild HIE were important predictors of death/disability with high area under the ROC curve (AUC) of ≥0.84 and performed better than stage at <6 h of age (AUC = 0.75). A lack of improvement in stage of encephalopathy during cooling carried higher odds of death or disability than those who improved in the first 24 h of treatment. This finding again highlights the value of serial neurologic examination over time. In this study the odds of the composite outcome of death or disability for those with severe encephalopathy at 24, 48 and 72 h increased sharply from an unadjusted odds ratio (OR) of 6.07 [95% confidence interval (CI): 3.11–11.82] at <6 h, to 55.93 (95% CI: 5.11–207.02) at 24 h, to 66 (95% CI: 17.52–252.19) at 48 h, to 81.99 (95% CI: 21.64–310.65) at 72 h. The odds of death for those with moderate or severe encephalopathy at each time-point evaluated during the course of hypothermia were not presented in either study, although patients died at various time-points during treatment.

Fig. 1.

Incidence of favorable and unfavorable outcome by stage of encephalopathy at pre-randomization and at day 4, in cooled and control neonates enrolled in the CoolCap Study. HIE, hypoxic–ischemic encephalopathy.

3. Neurophysiology

Electroencephalography has long been studied as a predictor of outcome in neonates with encephalopathy. Evolution of the EEG background over the first days to weeks after a perinatal insult carries important prognostic information that may be more objective and less subjective than the neurologic exam [10,13,14]. The advent of bedside aEEG allows neonatologists to continuously trend the background pattern during the course of TH without causing a significant burden on neonatal neurophysiologists. aEEG background patterns are based on voltage and pattern recognition criteria and use similar terminology and carry similar meaning to that encountered in EEG interpretation [15]. EEG and aEEG both provide a continuous real-time assessment of brain function and allow for the assessment of seizures, brain maturation, sleep–wake states, and general neurophysiologic health. Like the neurologic exam and stage of encephalopathy, after a hypoxic–ischemic insult, the EEG/aEEG background evolves and may either improve or worsen over the first days after birth [16].

There are several studies of aEEG and its predictive value for the composite outcome of death or disability, both in non-hypothermia [17,18] treated and hypothermia treated neonates [19–23]. The vast majority of these studies are cohort studies, and only one study compares hypothermia-treated and control babies who were enrolled in a randomized trial of hypothermia treatment [17]. Unfortunately, in the few RCTs that utilized aEEG, aEEG was used to select patients for randomization and was not necessarily continued for the full course of treatment. A report from the TOBY study group evaluated the pre-randomization aEEG background pattern and its positive predictive value (PPV) for the composite outcome of death or severe disability at 18 months of age [20]. Amplitude-integrated EEG obtained in the first 6 h after birth was reviewed and scored for the presence of severely abnormal background pattern defined as burst suppression (BS), continuous low-voltage (CLV), or flat trace (FT). The length of the aEEG recordings analyzed was not reported but needed to be at least 30 min in duration. The PPV of the pre-randomization recordings was lower in treated subjects (0.51) and slightly higher in control subjects (0.59) but the difference was not statistically significant. The authors postulate that hypothermia treatment lowers the PPV of a very early aEEG due to its beneficial effect on neurologic outcome. The PPV for the outcome of death was not reported.

Thoresen et al. performed the best study of aEEG background during hypothermia as a predictor of composite outcome of death or disability at 18 months of age [19]. This cohort study included 74 neonates, 43 treated with hypothermia, 31 normothermia, who met entry criteria for the CoolCap trial and were enrolled in either an RCT of hypothermia, a feasibility study, or in a registry of neonates with HIE. Amplitude-integrated EEG was recorded continuously from just after birth and through the treatment period. The PPV of an abnormal trace (BS, CLV, FT) at 3–6 h after birth was 0.59 in treated neonates and 0.84 in normothermic neonates. The PPV in treated patients increased steadily over time and at 24–36 h was 0.92 and at 48–60 h was the same as in the normothermic patients with PPV = 1. This study highlights the importance of evaluation over time and that pre-randomization (birth to 6 h) predictors, such as Apgar scores, clinical encephalopathy and aEEG background, are not the best predictors of outcome.

4. Magnetic resonance imaging

Magnetic resonance imaging is a robust predictor of neurodevelopmental outcome in both hypothermia-treated and non-treated neonates with HIE. Compared to non-treated neonates, those who receive treatment are more likely to have normal conventional imaging at the completion of therapy [24,25]. Regardless of therapy, injury to the deep gray nuclei, as assessed using conventional qualitative and advanced quantitative MRI techniques, has a strong association with adverse outcome, although most studies are of MRI at 7–14 days after birth [26,27]. Advanced MRI techniques, such as diffusion tensor imaging (DTI) and proton magnetic resonance spectroscopic imaging (¹H-MRS), can be used to study brain injuries associated with HIE, and can detect abnormalities that may be missed by conventional MRI alone, especially if imaging is performed in the first several days after birth [28]. DTI can detect changes caused by alterations of tissue microstructure; parameters typically used to measure these alterations include fractional anisotropy (FA), an indicator of directionality of water motion (and indirectly of axonal organization), and the apparent diffusion coefficient (ADC), the average magnitude of water motion during each DTI acquisition. ¹H-MRS can characterize the metabolic stress associated with hypoxia–ischemia. Measured metabolites include: lactate (Lac), a marker of anaerobic metabolism; N-acetyl aspartate (NAA), a marker for neuronal maturation/integrity; choline (Cho), a marker of membrane turnover; creatine plus phosphocreatine (Cr), a measure of brain energy metabolism, and glutamate, a measure of excitotoxicity [29]. In neonates who do not receive TH, serial imaging over the first 14 days of life demonstrates characteristic changes in FA, ADC and metabolites [28]. There is an acute drop in both ADC and FA values, followed by a period of pseudonormalization of ADC values, which occurs around the second week after birth (7–10 days of life). In the setting of TH, there is a similar decrease in diffusivity in the first few days after birth but there is a slight delay in pseudonormalization [30]. Changes in metabolic ratios (like Lac:NAA, which increases, and NAA:Cho, which decreases) can be detected as early as the first 2 days of life, and continue to worsen for 4–5 days, although there may be some subjects in whom early imaging may result in false negatives [28,31]. In neonates with HIE who did not receive TH, ¹H-MRS findings of the deep gray nuclei are strongly associated with outcome [32].

There are two predominant scoring systems used to evaluate MRI in neonates with HIE. Both systems, Rutherford et al. [25] and Barkovich et al. [33], consider the distribution and extent of injury, with particular attention to involvement of the cortical gray matter, the basal ganglia and thalami, and the white matter. The Rutherford system scores each region independently, whereas the Barkovich system applies the presumed pathophysiologic mechanism of injury, acute profound vs partial prolonged, to the corresponding pattern of injury of deep gray nuclei (basal ganglia/thalami) vs watershed predominant injury. Pattern and extent of injury remain important predictors of outcome even after treatment with hypothermia, as evidenced by sub-studies of three of the RCTs of hypothermia (25,34,35). The TOBY study group evaluated MRIs of 131 of the 321 randomized patients and identified that moderate–severe injury to the basal ganglia and thalamus, severe white matter injury or abnormal signal in the posterior limb of the internal capsule retained good predictive value in hypothermia-treated babies [25]. All of the 17 who died in this sub-study of the TOBY trial had moderate–severe injury in the basal ganglia or were noted to have abnormal signal in the internal capsule. Overall the presence of these abnormalities was 84% accurate in identifying those who died or who had severe disability at 18 months of age. Limitations to this study included that MRI was not performed in all patients enrolled in the trial; only conventional imaging was evaluated; and the age range (2–30 days) at which imaging was performed was variable, making it difficult to understand how MRI can be used early after birth in order to discuss redirection of care with families. A sub-set of the ICE Trial demonstrated similar findings of the TOBY trial in terms of MRI injury to the basal ganglia having high specificity and a PPV of 0.88 on conventional imaging and 0.89 on diffusion-weighted imaging for death or severe disability [34]. The ICE Trial sub-study had limitations similar to those of the TOBY sub-study in that MRI was performed in only 177 of the 221 randomized patients, but its strength was that MRI was uniformly performed in the first 10 days of after birth. The association between MRI findings and the outcome of death was not reported in either study. Finally the NICHD trial reported on the ability of several scoring systems to predict outcome in 136 of 208 randomized patients [35]. This study demonstrated good association between the scoring systems and the composite outcome but was again limited by significant variability in the timing of imaging (mean 15 ± 12 days) with only 39 patients being imaged at age <7 days. None of the sub-studies of the large RCTs were able to evaluate the association of imaging findings during the course of TH and outcome. There is only one small cohort study (n = 12) of serial MR imaging early in the course of TH. This study aimed to image neonates on day of life (DOL) 1, DOL 2–3, DOL 8–13, and again at 1 month of age. Evidence of severe irreversible hypoxic–ischemic injury to the basal ganglia and thalami was identified as early as DOL 2 (e.g. within 48 h). Using the Barkovich system to score images, three of the four with severe injury to the basal ganglia and thalami died prior to hospital discharge [36]. Overall, MRI is a robust predictor of outcome even after treatment with TH, but further work needs to be done to evaluate the ability of early MRI, in the first days after birth and during TH, to predict death or severe disability.

5. In clinical practice: association of aEEG background evolution with death in an international cohort of neonates treated with hypothermia

Neonates treated with hypothermia therapy for neonatal encephalopathy as part of standard clinical practice between 2007 and 2010 at UCSF Benioff Children’s Hospital and Wilhelmina Hospital in Utrecht, The Netherlands, were retrospectively studied in order to evaluate the association between clinical markers of severity of encephalopathy, aEEG background pattern during hypothermia and MRI findings with death in the neonatal period. The medical records of 141 (UCSF n = 77 and Utrecht n = 64) neonates were reviewed. Institutional review boards at each institution approved the review of the records. Sequential multivariate logistic regression and receiver operator characteristic (ROC) curve analysis was performed to determine the association between the predictors and death with the goal of simulating the information available to clinicians during the course of hypothermia. Predictors analyzed included the following: (1) clinical factors identified just after birth were categorized into “asphyxia factor” and considered present if the neonate had an Apgar score of <3 at 10 min, and/or need for adrenalin during resuscitation, and/or an umbilical cord or first newborn blood gas pH <6.8; (2) abnormal aEEG background pattern (BS, CLV, or FT) at 6–12 h after birth; and (3) persistent abnormal aEEG background pattern at 24 h of hypothermia therapy.

Death in the newborn period occurred in a total of 36 (25%) neonates, 18 at each institution. The median age at death was 3 days (range: 1–18). There were no significant differences between the sites in the proportion of neonates with an “asphyxia factor” identified (47% at each institution, P = 0.99), nor in those with a persistently abnormal aEEG background at 24 h of therapy (27% vs 41%, P = 0.08). An abnormal aEEG backround pattern at 6–12 h after birth was more frequent among subjects from Utrecht (39% vs 57%, P = 0.04). MRI was performed in 16 of 36 (44%) subjects who died in the neonatal period and all had moderate–severe brain injury. The severity of brain injury on MRI was not included in the logistic models, as MRI was not uniformly performed in subjects who died. The odds of death for each predictor in the multivariate model were as follows: the presence of an “asphyxia factor” [OR: 5.08 (95% CI: 1.24–20.8), P = 0.02]; abnormal aEEG at 6–12 h post birth [OR: 4.30 (95% CI: 0.36–51.02), P = 0.25]; persistently abnormal aEEG background at 24 h of treatment [OR: 39.44 (95% CI: 6.93–224), P < 0.0001]. Table 1 shows the sequential ROC analysis. The data incorporated in each ROC model simulate the predictors known to the clinician at each time-point when they are potentially considering the risk of death or poor outcome in their patients. Whereas the presence of an “asphyxia factor” carries a relatively large odds of dying in the neonatal period, the ROC analysis identifies that by itself the “asphyxia factor” is not a robust predictor of outcome as the AUC is only 0.68. As time passes and more information in the form of the aEEG background pattern is gathered, the AUC increases both steadily and statistically.

Table 1.

Area under the receiver operator characteristic curve (AUC) at each time-point for the prediction of death

Time point during therapy	AUC (95% CI)	P-valuea
Just after birth (n = 141)
Asphyxia factorb	0.68 (0.60–0.77)
First 6–12 h (n = 130)
Asphyxia factor + abnormal initial aEEG BGP	0.84 (0.77–0.91)	<0.0001
At 24 h (n = 123)
Asphyxia factor + abnormal initial aEEG BGP persistent abnormal aEEG BGP	0.94 (0.91–0.98)	0.0004

CI, confidence interval; aEEG, amplitude-integrated electroencephalography; BGP, background pattern.

^aCompares the AUC from each progressive time-point to the one prior.

^bApgar score of <3 at 10 min of life, receiving adrenalin just after birth, or an umbilical cord or first newborn blood gas pH <6.8.

These data are limited by their retrospective nature and by the concern for a self-fulfilling prophecy. The time-points after birth used in this study to determine the association between the predictors and death were chosen by the authors and are not necessarily what is done in clinical practice. However, as aEEG background has been discussed as a predictor of outcome, many clinicians may be considering the severity of the aEEG background and whether there is improvement in that background pattern over the first days after birth as they prepare to counsel families regarding the severity of the encephalopathy and possible outcomes. These data and other retrospective cohort studies will likely be the best data available regarding the prognosis of death as it would be unethical to delay conversations with families and not offer withdrawal of care in situations in which there is no clinical improvement or recovery of the aEEG/EEG background in neonates with severe encephalopathy after the first several days of life.

6. Limitations that impact clinical decision-making

Our ability as clinicians to accurately identify neonates with certain poor or devastating prognosis is limited by the types of tools that are currently available and by the amount of subjective interpretation that is required. Bias also limits the ability to interpret the results of most cohort studies that discuss the ability of these tools to predict death in the neonatal period, as most deaths result after redirection of care to comfort measures only, creating the concern for a self-fulfilling prophecy. In addition, studies of predictors of death or poor outcome may be influenced by the culture in which the study occurs. In some cultures, withdrawal of care is not presented as an option to families; in others, it is only offered to families of those neonates who are neurologically devastated; and in others, withdrawal of care is offered for a range of possible outcomes along the moderate–severe spectrum.

Practice points.

• Despite hypothermia treatment, a subset of neonates remains at risk of death or significant motor or cognitive disabilities.

• Conversations with parents and decision-making regarding withdrawal of intensive care for those neonates with anticipated hopeless prognosis should not be delayed by TH. Whereas uncertainty always exists in clinical care, if there are signs of persistent severe encephalopathy, parents should be informed of the concern for a poor outcome.

• Serial clinical evaluation to assess the risk of death or severe disability should be performed and should include serial neurologic examination to monitor evolution of encephalopathy and evaluation of the aEEG or EEG background over time (at 24, 48, 72 h) with specific attention for improvement or worsening of the background.

• MRI remains predictive of outcome even after treatment with hypothermia. MRI is often performed after the completion of hypothermia, but in some centers it can be delayed 14 days or longer after birth. Conversations regarding the risk of death or poor outcome should not be delayed while awaiting imaging to be completed. Imaging – although informative – can also be considered adjunctive to other risk factors such as aEEG background. MRI can be informative even if performed with the first several days after birth if severe brain injury is identified.

Research directions.

• Reliable biomarkers of severe injury that are easy to test for could improve the ability of physicians to identify neonates with a poor prognosis.

• Large prospective cohorts of neonates from multiple centers that utilize similar criteria for TH and similar practices for monitoring and imaging may be better able to study the information clinicians incorporate into determining prognosis and discussions regarding redirection of care.