The association of cerebral palsy with birth asphyxia: a definitional quagmire


Professor Jonas H Ellenberg at Center for Clinical Epidemiology and Biostatistics, Perelman School of Medicine at the University of Pennsylvania School of Medicine, 423 Guardian Drive, 604 Blockley Hall, Philadelphia, PA 19104, USA. E-mail:


Aim  The aim of this study was to investigate whether current literature provides a useful body of evidence reflecting the proportion of cerebral palsy (CP) that is attributable to birth asphyxia.

Method  We identified 23 studies conducted between 1986 and 2010 that provided data on intrapartum risks of CP.

Results  The proportion of CP with birth asphyxia as a precursor (case exposure rate) varied from less than 3% to over 50% in the 23 studies reviewed. The studies were heterogeneous in many regards, including the definitions for birth asphyxia and the outcome of CP.

Interpretations  Current data do not support the belief, widely held in the medical and legal communities, that birth asphyxia can be recognized reliably and specifically, or that much of CP is due to birth asphyxia. The very high case exposure rates linking birth asphyxia to CP can probably be attributed to several factors: the fact that the clinical picture at birth cannot specifically identify birth asphyxia; the definition of CP employed; and confusion of proximal effects – results – with causes. Further research is needed.


Hypoxic–ischemic encephalopathy

What this paper adds

  •  Study design affects the level of association between birth asphyxia and CP.
  •  Non-specificity in identifying birth asphyxia, the definition of CP used, and confusion of proximal effects with causes has led to the high association of with CP with birth asphyxia.
  •  Current data do not support the argument that much CP us due to birth asphyxia.

Is asphyxia at birth the dominant cause of cerebral palsy (CP) or the antecedent of only a small proportion of the disability? There is wide disparity in estimates of the proportion of CP attributed to birth asphyxia. In 1986, Nelson and Ellenberg1 reported that, contrary to the then current (and much subsequent) literature, the proportion of CP that might reasonably be attributed to birth asphyxia was less than 10%. Investigators continue to revisit this issue. We examined the reported case exposure rates of birth asphyxia in later CP cases in peer-reviewed papers published in the last two decades, to determine whether the succeeding reports have improved the understanding of this association and how the design approaches of succeeding studies have influenced their results.


Using PubMed and within-paper reference citations, we identified publications of original data from 1986 onwards that provided data on intrapartum risks of CP. The selection criteria required that publications reported on both author-defined birth asphyxia and CP. There were no restrictions based on methods of data collection or study design. Studies limited to preterm infants were not reviewed. A possible asphyxial etiology of CP may be relevant chiefly to relatively mature fetuses; thus, in the case of studies reporting case exposure rates separately for term and preterm infants, the results for term infants are reported.

The measure of association we used was the proportion of all children with CP with an author-defined history of conditions that can interrupt oxygen supply or blood flow to the fetus or infant. This measure of asphyxial exposure, the case exposure rate, is not an attributable risk or a relative risk. Authors did not necessarily compute this measure, but provided in their manuscripts the data for such a computation. For reports that included information that allowed identification of CP cases with and without non-asphyxial etiologies, the case exposure rates were computed for the cerebral cases without other non-asphyxial etiologies. We examined several components of study design to determine whether such study characteristics influenced the reported level of association between birth asphyxia and CP. The characteristics studied included resource-poor nation status, age at CP diagnosis, median calendar year of study coverage, types of CP considered, data collection approach (prospective vs retrospective, population or registry based versus referral or case series based), gestational age or birthweight, the definition of birth asphyxia, and the definition of CP.

We scrutinized the definition of birth asphyxia in each reviewed study to identify whether the definitions included factors likely to be potential instigating causes, or proximal consequences, of asphyxial birth, or conditions capable of causing similar clinical manifestations in the absence of known asphyxia. Asphyxia was coded 0 if the definition focused on clinically recognized asphyxial complications of birth that can deprive the fetus of oxygen or blood flow, such as uterine rupture or pulseless prolapsed cord, and 1 if asphyxia was either undefined or defined using a diagnosis of hypoxic–ischemic encephalopathy (HIE) or neonatal seizures to indicate the occurrence of birth asphyxia. We also examined whether identified syndromes, postneonatal causes, brain malformations, or other non-asphyxial pathologies were excluded from definitions of CP. We dichotomized the definition of CP based on what the authors did or did not exclude in their definitions of the outcome, coding 0 if cases with at least some non-asphyxial pathologies were excluded and 1 if no exclusions were specified.

Statistical analysis

The use of a synthetic review process (meta-analysis) to create a combined estimate from multiple studies generally requires that the studies are comparable with regard to their cohorts and study designs; otherwise it is not clear how useful such an estimate might be. We assessed the variability of the observed results using a measure of variability (I2), a measure of study inconsistency, and also Cochran’s Q statistic, an assessment of whether the studies are all measuring the same effect.2 Exact binomial confidence intervals were calculated for the point estimates of the case exposure rate.3,4 Multivariate regression with stepwise forward selection using SAS software (SAS Institute, Cary, NC, USA) allowed assessment of the importance of design characteristics in the reported level of the case exposure rate. Analytic procedures for synthesis were carried out using StatsDirect software (StatsDirect Ltd, Altrincham, Cheshire, UK).


The literature search yielded 71 records. Seven records involved second reports on the same database and were excluded. Of the remaining 64 reports, 23 met the inclusion/exclusion criteria and serve as the basis for this report (Fig. 1). Three of these reports5–7 covered successive birth cohorts and were considered separate reports. Five reports included information that allowed for exclusion of non-asphyxial etiologies.1,5,8–10

Figure 1.

 Flow diagram detailing the literature search process for papers that allowed examination of the association of birth asphyxia and cerebral palsy BA, birth asphyxia; CP, cerebral palsy.

The case exposure rates for birth asphyxia varied widely, from less than 3% to over 50%. The point estimates and confidence intervals from the 23 reports are presented in Figure 2.1,5–26 The number of children with CP in these cohorts ranged from 41 to 1000.

Figure 2.

 Percentage of cerebral palsy cases with birth asphyxia (case exposure rate). Sorted by point estimates of case exposure rate. Confidence intervals are shown as bands around the point estimates.1,5–26

Univariate analysis

Study design characteristics

Except where indicated, the univariate associations of study design characteristics with reported level of the case exposure rate had a computed p-value greater than 0.10. Studies that were reasonably close to population or registry-based studies and studies using prospective (vs retrospective) data collection had lower case exposure rates (p=0.047 and p=0.089 respectively). The reports reviewed covered median time frames from 1962 through to 2002. Nine studies originated in resurce-poor countries. Those studies from resource-poor countries and those studies using later cohorts reported higher case exposure rates (p<0.001 and p=0.013 respectively). Almost all studies included all CP subtypes and one20reported only on spastic quadriplegia. The minimum age reported for diagnosis of CP was 3 years 6 months and above in nine studies, and included cases diagnosed at less than 3 years 6 months, or was unspecified in 14. For half of the studies, data allowed for the derivation of a case exposure rate for term children. The remaining studies were reported for all maturities.

Study definitions

The reports differed in definitions for CP and for birth asphyxia, and some papers did not provide definitions.

Cerebral palsy

We were not always able to determine whether or to what extent CP cases with non-asphyxial pathologies known to be related to markedly heightened risk of CP were excluded from the assessment of case exposure rate. Among the 23 studies, nine specified no exclusions (two of these nine studies excluded only CP cases that were postnatally acquired) and 14 specified some exclusions. The reports that excluded or allowed us to exclude CP cases that also had other potential non-asphyxial etiologies had lower case exposure rates (p=0.006).

Five reports permitted examination of the impact of including or excluding individuals with non-asphyxial explanations for their CP. Table I shows that in these five reports, case exposure rate estimates were markedly lower when cases with potential non-asphyxial pathologies were excluded. The five reports shown in Table I provided information that allowed estimation of the case exposure rates with and without exclusion from the computations of those cases of CP that had other, non-asphyxial pathology that could have been related to the origin of CP. Some were not specific (e.g. Hagberg et al.5 described such cases as having ‘no predisposing factors,’ and Strijbis et al.10 described these cases as those with ‘associated CP pathology’). Shevell et al.8 described such cases as not having cerebral dysgenesis, intracerebral hemorrhage, vasculopathy, infection, trauma, atrophy, or toxins. In our previous work1 we defined such cases as those with congenital malformations, microcephaly, very low birthweight (≤2000g), or alternative explanations for the child’s CP. Torfs et al.’s9 data allowed for identification of children with birth asphyxia and CP who also had minor or major birth defects. Our own work1 would have yielded an estimate of case exposure of 21.2% (40/189) not accounting for other pathologies, and yielded instead a case exposure rate of 8.9%. Strijbis et al.10 reported that 13 out of 15 of children with CP had major potential antenatal or pediatric pathologies which, when excluded, reduced their reported case exposure rate from 28.3 to 4.3%. Case exposure rates were markedly higher in the data from those reports that did not specify exclusions of potential non-asphyxial etiologies of CP from the analytic considerations.

Table I. Impact of inclusion of other pathologies on the estimate of cerebral palsy (CP) cases with birth asphyxia (case exposure rate)
AuthorTotal CP (1)Nr. with asphyxia and CPCase exposure rate (%)
Where CP defined without excluding other pathology (2)Where CP defined excluding other pathology (3)Where CP defined without excluding other pathology (4)Where CP defined excluding other pathology (5)
  1. Case exposure rates given as column (2)/column (1) and column (3)/column (1) for computations without and with exclusions of other pathologies respectively.

Shevell et al.8217472721.712.4
Nelson and Ellenberg1189401721.28.9
Hagberg et al.5132372328.017.4
Torfs et al.9419122.02.4
Strijbis et al.104613228.34.3

Birth asphyxia

Five of the 23 studies lacked any definition of birth asphyxia. Twelve studies defined birth asphyxia as the presence of a diagnosis of HIE with abnormal neurologic signs in the newborn infant, including neonatal seizures. Only six studies defined birth asphyxia as conditions of birth capable of interfering with oxygen supply or perfusion (‘sentinel events’), not extending the definition of birth asphyxia to include HIE with neonatal signs or newborn seizures. The reports that excluded or allowed us to exclude CP cases that also had other potential non-asphyxial etiologies had lower case exposure rates (p=0.057).

Multivariate analysis

Multivariate analysis followed the consideration of univariate assessment of components of study design as factors associated with the level of case exposure. Developing nation status was the only statistical significant predictor when considered in tandem with other design components (p<0.001). Of the nine studies reported from resource-poor nations, six included potential non-asphyxial etiologies in their definitions of CP, and eight either did not define birth asphyxia or defined it using HIE, indicating a confounding between resource-poor nation status and the definitions of both birth asphyxia and CP.

Synthesis of results

We assessed the variability of the observed results using a measure of variability (I2), resulting in a value of 98.8% (confidence interval 98.6–98.9%), where 0% indicates no observed heterogeneity in the effect being measured. Cochran’s Q statistic, also an assessment of whether the studies are all measuring the same effect, rejected the null hypothesis with a p-value of less than 0.001. Similar results were found when reports from only resource-poor nations were considered. Based on this marked evidence of heterogeneity of results and study design factors, no meta-analyses were undertaken.


Our review of the literature of the past two decades indicates the persistence of a wide disparity in the proportion of CP case exposure rates for birth asphyxia, with estimates ranging from less than 3% to more than 50%. Why is disagreement among studies in the proportion of CP attributed to birth asphyxia so large? There appear to be several interrelated factors leading to questionable definitions for both birth asphyxia and CP. One is the non-specificity of the clinical picture at birth, the second is the confusion of proximal effects/results with causes, and the third is the definition of the outcome, CP. These varying characteristics of the studies do not permit data synthesis.

Non-specificity of ‘birth asphyxia’ syndrome

When a massive placental abruption occurs in the delivery of a late preterm or term infant, the infant may be born pale, limp, apneic, and unresponsive to stimulation. In that situation, it would seem legitimate to conclude that interruption of oxygen supply and ischemia were the cause of the clinical findings. When another infant is born pale, limp, apneic, and unresponsive, it may seem to attendants that the infant must also have suffered an interruption of oxygen supply or ischemia, even if no such asphyxial events were observed. However, newborn infants have a limited range of responses. Intrauterine exposure to inflammation or vasculopathy can mimic all features of the clinical picture of birth asphyxia, even producing abnormal fetal heart rate patterns and severe acidosis. Infants who are seriously affected by either asphyxial or inflammatory pathology, or perhaps other antecedents, may exhibit depression of consciousness in the first hours and days of life and may have neonatal seizures; they are at risk for long-term neurologic disability, including CP. Two reports indicate that placental infarction, especially in infants who are growth restricted, is a potential cause of neonatal encephalopathy and later CP, especially spastic quadriplegic CP, thus mimicking asphyxial birth.15,27 The clinical picture of encephalopathy in the neonate is associated with increased risk of adverse outcome, but it is not specific as to cause.

Many investigators have developed animal models of birth asphyxia, and this direction of research has revealed interesting and important information. Animal models cannot, however, create the complex web of interactions in the human experience. It is generally agreed that the etiology of CP is multifactorial, with substantial interactions among risk factors.28 Yet most animal models examine only a single factor, most commonly hypoxia–ischemia. Furthermore, experimental models cannot identify what proportion of neonatal encephalopathy occurs in human infants with asphyxial births; that information can be derived only from clinical studies in human populations.

The experience of controlled prospective studies in representative populations has been uniform in observing that most infants with neonatal encephalopathy, and the subgroup of those infants who go on to demonstrate CP, did not experience acute asphyxial events at birth.29–36

Definition of cerebral palsy

According to one consensus document, ‘Cerebral palsy describes a group of permanent disorders of the development of movement and posture, causing activity limitation, that are attributed to no progressive disturbances that occurred in the developing fetal or infant brain.’37 This formal definition is not an etiologic one and is silent on inclusions or exclusions. Another consensus document excludes major congenital malformations, antenatal infection and growth restriction that suggest a cause for CP other than acute intrapartum hypoxia.38 Of the controllable factors in the design of studies of association between birth asphyxia and CP, more inclusive definitions were associated with markedly higher rates of attribution of CP to birth asphyxia. Definitions of CP varied among the papers, with nine of the 23 reports offering no information on exclusion of other non-asphyxial etiologies. Five reports did provide information that permitted an assessment of the impact of including, among CP cases, individuals with both birth asphyxia and non-asphyxial pathologies (Table I). These reports demonstrated that such inclusion leads to markedly heightened levels of the proportion of CP with birth asphyxia as a precursor.

Causes versus proximal effects or joint consequences

Hypoxia during birth is seldom measured directly, but is inferred from clinical observations. Neurologic depression in term-born neonates has a variety of potential causes whose manifestations are clinically similar; for example, inflammatory conditions are common antecedents of low Apgar scores, depressed respiration, and other clinical signs that can also occur in asphyxia.39–42 If the role of birth asphyxia as an initiating factor on the causal pathway to CP is to be correctly assessed, a surrogate must be used that is relatively specific to birth asphyxia and not itself an early symptom of the developing disorder. An important and recurring issue is the conflating of proximal effects or joint consequences of a factor underlying both, with causes. For example, fetal monitoring abnormalities, low Apgar scores, and neonatal seizures are often taken as indicators of asphyxia, although none of these signs is specific to asphyxial birth; all are often related to underlying placental pathology.43 The most common antecedent of low Apgar scores is maternal fever in labor or a diagnosis of chorioamnionitis.44,45 Neonatal seizures are commonly a result of perinatal stroke or inflammation.46,47 None of these findings – low abnormal fetal heart rate patterns, low Apgar scores, or neonatal seizures – has birth asphyxia as its only or even its major cause. As already mentioned, the behavioral repertoire of a newborn infant is limited, and many different antecedents elicit a similar pattern of response.

So long as effective neonatal resuscitation is available, low Apgar scores are a result of prior causes. Neonatal seizures are early manifestations of brain injury, sometimes relatively mild and reversible, sometimes severe and associated with long-term disability. These signs are compatible with the possibility that the brain damage underlying CP is already present in many of these infants. The confusion of causes with proximal effects was especially notable in studies that used encephalopathy in the neonate, commonly called HIE, or neonatal seizures as indicators of hypoxic or ischemic births. As Kurinczuk et al.48 stress, the inclusion of intrapartum complications and of proximal effects into the criteria for birth asphyxia is ‘a tautology indeed.’

The 23 studies reviewed were investigations of association, not of causation. If care is taken to create definitions of interruption of oxygen flow or blood flow to the infant (‘sentinel events’) and definitions of CP that are free of observable etiologies other than birth asphyxia, such studies could provide hints of causation. The data from four papers (Nelson and Ellenberg,1 Torfs et al.,9 Lie et al.,11 Nelson and Grether14) with a definition of CP that excluded non-asphyxial etiologies and had definitions of birth asphyxia that used only clinically recognized acute asphyxial events of birth to define birth asphyxia, produced among the five (of seven) lowest case exposure rates (12% or less) for CP and birth asphyxia (Fig. 1).

Estimates of the proportion of CP with birth asphyxia were high in reports from low-resource countries. Of the nine studies from low-resource areas, six did not specify non-asphyxial exclusions to the diagnosis of CP and seven had inappropriate definitions of asphyxia. If other potential non-asphyxial etiologies were in these datasets, then the dramatic results in Table I would indicate that it is the definitions in these reports and not the nature of the populations that produced the high case exposure rates. Birth asphyxia may truly be more common as a cause of CP in regions of limited resources, but we could not distinguish this possibility because of the weaknesses of the definitions offered in these reports.


The current data do not support the belief, widely held in the medical and legal communities, that birth asphyxia can be recognized reliably and specifically on the basis of clinical signs such as aberrant fetal heart rate patterns, Apgar scores, respiratory depression, neonatal seizures, or acidosis, or that most CP is due to birth asphyxia. Overly inclusive definitions were associated with high rates of attribution of CP to birth asphyxia. A meta-analysis of these studies to derive a combined estimate of case exposure rate is not appropriate given the heterogeneity of designs and results among the reports. Future research on the etiology of CP should aim towards the following definition of CP: a group of permanent disorders of the development of movement and posture, causing activity limitation that are not attributed to progressive disturbances that occurred in the developing fetal or infant brain.37 When the purpose of the definition is to provide an estimate of the contribution of an asphyxial etiology to CP, cases with known or probable non-asphyxial etiologies such as brain malformations, death of a co-twin, or metabolic or neuromuscular disease should be excluded.

The following definition of birth asphyxia should also be pursued: a group of factors related to interruption of oxygen supply during the immediate perinatal period. An examination of an asphyxial etiology should not include clinical findings that are etiologically non-specific – potential results of a variety of etiologic factors (such as abnormal fetal heart rate monitoring findings or meconium in the amniotic fluid) – or that may themselves be signs of early brain injury (such as neonatal seizures or neonatal encephalopathy). Although not optimal, the best identifier now available at a population level for asphyxial birth is the occurrence of sentinel events such as uterine rupture, major placental abruption, or cord prolapse.

It is likely that placental disorders underlie much of what has been referred to in the past by vague terms such as ‘maternal deprivation of supply,’‘chronic hypoxia,’ and ‘placental insufficiency.’ Studies that include maternal and pregnancy factors, outcomes, and placental histology are needed to identify specific placental lesions, their etiologies, and their (probably multiple) consequences.

The ultimate demonstration of an asphyxial etiology of CP would be that an intervention designed to improve oxygenation of the fetus during birth results in a decrease in CP. To date there has been no such evidence.


The authors thank Ms Donna Zikowitz for her diligence in working on revisions of this manuscript and Ms Colleen Brensinger for her statistical programming support. This work was supported in part by the Intramural Research Program of the National Institute of Neurological Disorders and Stroke, and the Division of Biostatistics at the Perelman School of Medicine at the University of Pennsylvania. There was no involvement by either of these organizations in study design, data collection, data analysis, manuscript preparation, and/or publication decisions. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, or the University of Pennsylvania nor does mention of trade names, commercial products, or organizations imply endorsement by the US government or the University of Pennsylvania.