Intended for healthcare professionals

CCBYNC Open access
Research

Strength of association between umbilical cord pH and perinatal and long term outcomes: systematic review and meta-analysis

BMJ 2010; 340 doi: https://doi.org/10.1136/bmj.c1471 (Published 13 May 2010) Cite this as: BMJ 2010;340:c1471
  1. Gemma L Malin, clinical research fellow1,
  2. Rachel K Morris, clinical research fellow1,
  3. Khalid S Khan, professor of obstetrics, gynaecology, and clinical epidemiology12
  1. 1Department of Obstetrics, School of Clinical and Experimental Medicine, College of Medical and Dental Sciences, University of Birmingham
  2. 2Department of Obstetrics and Gynaecology, Birmingham Women’s Foundation NHS Trust, Birmingham
  1. Correspondence to: G L Malin, Academic Department of Obstetrics and Gynaecology, Birmingham Women’s Hospital, Edgbaston, Birmingham B15 2TG g.l.malin{at}bham.ac.uk
  • Accepted 14 February 2010

Abstract

Objective To evaluate the association between umbilical cord pH at birth and long term outcomes.

Design Systematic review and meta-analysis.

Data sources Medline (1966-August 2008), Embase (1980-August 2008), the Cochrane Library (2008 issue 8), and Medion, without language restrictions; reference lists of selected articles; and contact with authors.

Study selection Studies in which cord pH at birth was compared with any neonatal or long term outcome. Cohort and case-control designs were included.

Results 51 articles totalling 481 753 infants met the selection criteria. Studies varied in design, quality, outcome definition, and results. Meta-analysis carried out within predefined groups showed that low arterial cord pH was significantly associated with neonatal mortality (odds ratio 16.9, 95% confidence interval 9.7 to 29.5, I2=0%), hypoxic ischaemic encephalopathy (13.8, 6.6 to 28.9, I2=0%), intraventricular haemorrhage or periventricular leucomalacia (2.9, 2.1 to 4.1, I2=0%), and cerebral palsy (2.3, 1.3 to 4.2, I2=0%).

Conclusions Low arterial cord pH showed strong, consistent, and temporal associations with clinically important neonatal outcomes that are biologically plausible. These data can be used to inform clinical management and justify the use of arterial cord pH as an important outcome measure alongside neonatal morbidity and mortality in obstetric trials.

Introduction

Perinatal asphyxia is a major cause of neonatal and childhood morbidity and mortality and has been associated with neonatal death1; hypoxic ischaemic encephalopathy and seizures2 3; intraventricular haemorrhage4; cerebral palsy5; and delayed development.5 Perinatal asphyxia is predicated by fetal acidosis, determined by umbilical cord pH at birth.6 Cerebral palsy is thought to occur more frequently at an arterial cord pH of <7.00 and a base deficit of ≥12 mmol/l.7 These criteria, however, have been derived through consensus, not through evaluation of collated summaries of evidence,7 leading to clinical uncertainty.8 This is because existing observational studies of the association between cord pH and outcomes have drawn inconsistent inferences, partly as a result of the different thresholds used to define abnormality, the variety of outcomes evaluated, and the different variables measured (arterial cord pH, venous cord pH, or base excess).9 10 11 It has been suggested that neonatal complications are associated with metabolic rather than respiratory acidosis.12 13 Respiratory acidosis arises in the early stages of impaired blood supply to the fetus; hypoxaemia and hypercapnia occur, leading to a reduction in pH with a normal base excess.6 If hypoxia is prolonged, anaerobic metabolism results and base excess rises secondary to the presence of lactic acidosis.

On the basis of nine studies, a review14 claimed an association between fetal acidosis and neonatal death and cerebral palsy. This review was, however, carried out in the late 1990s, during which time new studies on the subject have been published and guidelines produced on the methodology and reporting of systematic reviews, including quality assessment of included studies, which were not in widespread use at the time of the review’s publication.15 Substantial uncertainty therefore remains about the value clinicians may attach to acidosis in the clinical management of neonates and the long term implications of a low arterial cord pH. However, cord pH is also commonly used as an outcome measure in obstetric clinical trials16 17 18 and it is one of the bench marks by which obstetricians judge their performance on the labour ward.19 A documented low cord pH is a factor that may be used to support medico-legal claims of harm during intrapartum events resulting in long term disability.20 It is therefore imperative that the validity of this association is supported with high quality evidence. We carried out a comprehensive systematic review of the literature to quantitatively establish the strength of association of acidosis at birth with neonatal mortality, morbidity, and long term outcomes and to assess if causal criteria were met.21

Methods

This systematic review was protocol driven using widely recommended methods for reviews22 23 24 and evaluation of causal associations.21 25 26 27 28 The review was carried out and has been reported according to MOOSE guidance.15

We searched Medline (1966-August 2008), Embase (1980-August 2008), the Cochrane Library (2008 issue 8), and Medion for relevant published articles. To identify “grey” literature we also searched SIGLE, Web of Science, the national research register, and medical conferences register. For the Medline search (also see web extra on bmj.com) we used a combination of MeSH headings, such as umbilical cord, hydrogen-ion concentration, or asphyxia neonatorum; keywords, such as umbilical artery pH and cord pH; and word variants using “OR” for capturing citations of the relevant text. To capture relevant outcomes we combined these using “AND” with a combination of MeSH headings, such as human development and infant mortality; keywords, such as developmental delay and handicap; and word variants. We restricted the search to human studies but applied no language restrictions. We adapted the Medline search strategy for use in other databases. We also hand searched recent major journals. Two investigators (RKM and GLM) carried out the search. Using Reference Manager 11.0, we constructed a comprehensive database collating all citations.

Study selection and data extraction

Two reviewers (RKM or GLM, partly in duplicate) scrutinised the database and obtained full articles of all citations potentially meeting the predefined selection criteria. Translations of articles in languages other than English were obtained. Two reviewers (GLM and RKM or ZL) included or excluded studies in accordance with the most recent guidance,23 using strict adherence to the following criteria: population—infants with cord blood obtained at birth; index test—cord blood examined for arterial or venous pH or base excess; outcome—any measure of compromise of neonatal or childhood wellbeing, such as mortality; neonatal morbidity, including hypoxic ischaemic encephalopathy, seizures, intraventricular haemorrhage, periventricular leucomalacia, and long term outcomes, including cerebral palsy; and study design—observational studies that allowed generation of a 2×2 table (true positives, false positives, false negatives, and true negatives) to compute an estimate of the association between test result and outcomes. We excluded studies with five or fewer cases, because of unreliability.

The manuscripts were examined for duplicated populations. If any were found we selected the most recent and complete version. No language restrictions were applied in study selection. We scrutinised the reference lists of selected studies and review articles and obtained additional relevant articles. Two researchers (GLM and RKM or ZL) used a data collection sheet to extract information in duplicate from the selected articles. Data were extracted on study characteristics, quality, and results, and entered onto an Excel spreadsheet. Data were used to construct 2×2 tables of the association between the cord blood variable at the threshold reported in the paper and the postnatal outcome for each infant. When data were thought to be relevant but 2×2 tables could not be constructed, we contacted the relevant authors. Difficulties in data extraction were resolved by discussion with a third reviewer (KSK).

Study quality assessment

We assessed those articles meeting the selection criteria for methodological quality, defined as the confidence that the study design, conduct, and analysis minimised bias in estimation of the association. The articles were assessed using the complete STARD and QUADAS checklists.29 30 These are guidelines for reporting, and methodological quality of, studies on diagnostic accuracy. As the focus of this review was to determine association, the checklist items thought to be most relevant were selected for the purpose of defining the overall quality of included studies and to divide the studies according to quality for meta-analysis and metaregression, in accordance with published guidance.23 We did not assign a quality score as this has been shown to give flawed results.31 We considered the cohort study design to be superior to the case-control design, and we considered the cohort design as a covariate in metaregression. Because of the small number of studies included in each meta-analysis we did not think it appropriate to consider each remaining quality item as an individual covariate for metaregression,32 and we therefore divided the studies into two categories on the basis of several criteria. We rated a study as being of high quality if it had at least four of the following items: adequate description of population, adequate description of the test and outcome measure, consecutive recruitment, prospective recruitment, more than 90% completion of follow-up, appropriate outcome measurement, blinding of the investigators carrying out the outcome measure, and a statement on the use of intervention between the index test and outcome. When a study adhered to three or fewer of these criteria we considered it to be of medium or low quality.

Data synthesis

We used 2×2 tables to compute odds ratios and 95% confidence intervals for each pair of index test and outcome. The odds ratio was selected as the summary statistic as it is relatively constant regardless of the test threshold33 and is often used to show an epidemiological association.34 It represents the effect of the exposure on the odds in an unbiased fashion and enables linkage between the results of case-control and cohort studies.34 As the focus of this review was to establish the association between cord pH and adverse outcome rather than the value of cord pH as a prognostic test, we did not calculate other summary statistics (including sensitivity, specificity, and likelihood ratios). We calculated the estimated predictive interval for each meta-analysis, which relates to the effect of a new study that would be eligible for inclusion in meta-analysis and therefore allows the full uncertainty around inferences to be calculated, including both magnitude and consistency.35

When tables contained cells with a value of 0, we added 0.5 to these cells to allow the calculation of variances.36 Analyses were done for groups defined a priori according to index test (arterial cord pH, venous cord pH, and base excess). The main outcome measures considered were neonatal mortality, a composite measure of neonatal morbidity, and cerebral palsy. To maximise the number of events that could be included in the analysis and to avoid the need to select a single morbidity as a primary outcome measure, we used a composite outcome measure for neonatal morbidity. However, a hazard of composite outcome measures is the assumption that the significance of the result applies to all components.37 To deal with this, we analysed the component outcomes of hypoxic ischaemic encephalopathy, seizures, intraventricular haemorrhage, or periventricular leucomalacia. When the composite outcome measure was used, we took care to ensure that each individual was counted only once in each analysis, particularly when studies reported multiple outcomes for a single population. When multiple outcomes and test thresholds were reported, we attempted to select the most consistent threshold and outcome across the composite analysis—for example, where studies reported on hypoxic ischaemic encephalopathy, seizures, and intraventricular haemorrhage, we included hypoxic ischaemic encephalopathy and seizures in the meta-analysis as we believed this to be more clinically consistent and overall the more common outcome across the included studies.

We plotted odds ratios data in forest plots, assessing heterogeneity visually and statistically using the I2 statistic, a scale that measures the degree of heterogeneity across studies in a meta-analysis.38 Within the largest groups (arterial cord pH paired with neonatal mortality and neonatal morbidity) we explored the reasons for heterogeneity using metaregression, planned a priori in keeping with published recommendations.39 40 We considered study design (cohort v case-control), study quality (high v medium or low), and population risk (high v low/unselected or unreported) as potential sources of heterogeneity. We defined a population at high risk for complications on the basis of reported characteristics, including abnormalities on cardiotocograms, meconium liquor, a low Apgar score at birth, gestation less than 37 weeks, and birth weight less than 2500 g. It was not possible to use a low risk population as comparator because only one included study met the criteria for this definition.41 Given the reporting quality of the primary studies it was also not possible to explore other potential sources of heterogeneity. Metaregression analysis was univariate throughout, in accordance with published guidance that recommends allowing at least 10 studies per covariate explored.42 When a reason for heterogeneity was identified, we carried out subgroup analyses. As complications such as cerebral palsy and intraventricular haemorrhage have been shown to be increased in preterm and low birthweight infants we analysed studies with these populations separately and compared the results with that of a term population.

As clinical heterogeneity was present between studies, we used a random effects model throughout. The presence of publication bias was determined using the Harbord test.42 43 To assess the effect of the use of different thresholds to define low pH, we repeated the meta-analyses using the bivariate method, which takes into account any threshold effect.44 We additionally analysed subgroups within the outcome measures of neonatal mortality and neonatal morbidity according to the arterial pH threshold reported (<7.00, <7.10, and <7.20). We statistically analysed data using Meta-Disc 1.3 software,45 and Stata version 10.0 using the metan, metandi, and meta-bias commands.46 47 48

To assess a causal association for each outcome we considered Hill’s criteria.21 25 We examined strength of association using point estimates of odds ratios and consistency using the direction of association in the forest plots and heterogeneity statistics.

Results

An initial search of 5690 citations identified 51 primary articles1 4 5 10 11 41 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91, two after obtaining information from the authors (fig 1).92 93 The included studies totalled 481 753 infants. Overall, 190 2×2 tables were produced; 43 of the studies providing these data were eligible for inclusion in meta-analyses according to predefined outcome measures, totalling 479 383 infants and 70 2×2 tables. One included study was thought to contain duplication60; however, the outcomes reported were different and therefore both papers60 61 were included in the review in separate meta-analyses. When more than one 2×2 table was produced per study, double counting of infants in any one analysis was avoided.

Figure1

Fig 1 Study selection process

The included studies used various index tests (table 1): most papers (n=46) reported arterial cord pH, with thresholds ranging from 7.00 to 7.24. Arterial base excess was reported in four studies, with thresholds of 12-16 mmol/l.61 73 81 91 Five studies reported venous cord pH; three using a threshold of 7.20 and one a threshold of 7.10.55 69 78 85 93 Only one study utilised venous base excess.93 The cord pH level was obtained before the occurrence of the adverse outcome. A wide variety of outcome measures were reported, ranging from neonatal mortality and morbidity to long term outcomes, including cerebral palsy, unspecified neurological abnormality, IQs, and developmental assessments (for example, Bayley and Griffiths scores). When the same outcome was reported, the thresholds and ages at assessment varied between studies, therefore meta-analysis was not possible for long term outcomes other than cerebral palsy. The odds ratio point estimate varied widely. Thirty six studies reported on outcomes within the neonatal period only. Fifteen studies had long term follow-up; one did not report the age of ascertainment of the outcome. The age range reported in other studies was 1 to 8 years; the median across studies was 5 years.

Table 1

 Studies included in systematic review of association between cord pH at birth and neonatal and long term outcomes

View this table:

Quality assessment

Overall, study quality was variable (table 2). Most studies were retrospective but with a cohort design, allowing inferences on temporality of association. Over 80% of studies met the following quality items: appropriate outcome measure, more than 90% verification with outcome measure, and cohort design. Studies scored poorly for adequate description of index test and outcome measure, and consecutive recruitment. All included studies were unclear about the use of medical intervention between the index test and outcome measure. Overall, 45% (n=23) of studies were of high quality, 39% (n=20) of medium quality, and 16% (n=8) of low quality. Study design and quality did not seem to influence results on metaregression (table 3).

Table 2

 Methodological quality of studies in systematic review of association between low cord pH at birth and neonatal and long term morbidity

View this table:
Table 3

 Exploration of heterogeneity in estimation of association between low arterial cord pH at birth and neonatal mortality and morbidity

View this table:

Association between cord pH and neonatal mortality

Fifteen studies (13 cohort, two case-control) totalling 469 395 infants reported on the association between arterial cord pH and mortality. The Harbord test was not significant, suggesting no influence of small study effect (see web extra on bmj.com). All studies had an odds ratio point estimate greater than 1.0. Heterogeneity was significant (I2=61.0%) overall. Metaregression (table 3) identified population risk as a significant explanatory factor. Within subgroups (fig 2) the association was consistent across studies. The association of low arterial cord pH with neonatal mortality was stronger in the unselected population (odds ratio 16.9, 95% confidence interval 9.7 to 29.5, estimated predictive interval 5.0-57.3, I2=0.0%) than in the high risk population (4.2, 2.6 to 6.9, 1.3-13.7, I2=35.1%). The analysis for the unselected population was dominated by one large study,1 which was based on data from a regional perinatal register where cord pH was determined at most of the births. A sensitivity analysis excluding this study this did not significantly alter the point estimate (17.0, 4.4 to 65.5); however, the estimated predictive interval became wide (0.0-106 299.0)

Figure2

Fig 2 Association of low arterial cord pH with neonatal mortality. EPI=estimated predictive interval

Examination of the high risk population further showed that when studies reporting on a population of infants born at less than 32 weeks’ gestation, or with a birth weight of less than 2000 g, were analysed separately (seven studies), a significant association was found between cord pH and mortality (3.5, 2.3 to 5.4, I2=0.0%). When limited to a term (>37 weeks’ gestation) population (four studies), the association was also strong (9.3, 1.4 to 63.2, I2=84.0%); however, heterogeneity was significant and the predictive interval broad (0.0-38 169.8). Grouping the studies according to quality did not affect the significance or direction of association of the pooled result (fig 2). Exploration for a threshold effect (table 4) showed that for a cut-off pH threshold of 7.00 the association did not reach significance overall (6.1, 0.9 to 41.6) and the predictive interval was broad (0.0-20 406.6). The results for a threshold of 7.10 gave a similar point estimate but achieved significance (7.1, 3.3 to 15.3); however, the predictive interval remained broad (0.8-64.3) and crossed the line of no effect. For a threshold of 7.20, the odds ratio was lower (4.3, 2.2 to 8.7), with a broad predictive interval (0.5-40.6). Only one study examined all three thresholds,1 with the strongest association at threshold 7.00 (16.9, 9.2 to 31.1) and the weakest at 7.20 (3.1, 2.3 to 4.1).

Table 4

 Effect of using varying thresholds of arterial cord pH at birth on accuracy to predict neonatal morbidity and mortality

View this table:

Association between cord pH and neonatal morbidity

Thirty one studies compared arterial cord pH with a variety of neonatal outcomes (fig 3). One of these61 was excluded from meta-analysis of composite neonatal morbidity because it contained a duplicate population to another included study,60 therefore the dataset to assess the association of arterial pH with a composite measure of morbidity included 30 studies with 10 904 infants. The Harbord test was significant, suggesting the presence of a small study effect (see web extra on bmj.com). Only two studies had an odds ratio point estimate of less than 1.0, the rest showing an association. Significant heterogeneity was, however, present (I2=58.2%). Metaregression showed population to be an explanatory variable (table 3). Subgroup meta-analysis for a high risk population (fig 3) showed a weaker association (3.4, 2.3 to 4.9, 1.4-8.4, I2=26.4%) than in an unselected or undefined population (10.6, 4.7 to 24.1, 0.8-135.8, I2=66.4%). When analysed in subgroups according to quality, the direction of effect remained consistent and significant between the group of 12 high quality studies (6.6, 3.7 to 11.8, 1.3-32.6, I2=51.2%) and 18 low or medium quality studies (4.6, 2.6 to 8.0, 0.7-29.2, I2=61.5%). Exploration by subgroup analysis for a threshold effect (table 4) showed the most substantial association at a pH threshold of 7.00 (12.5, 6.1 to 25.6). The predictive interval was broad but did not cross the line of no effect (1.7-89.9). The results were similar for a threshold of 7.10 (2.4, 1.3 to 4.3) and 7.20 (2.2, 1.3 to 3.7).

Figure3

Fig 3 Association of low arterial cord pH with neonatal morbidity. HIE=hypoxic ischaemic encephalopathy; IVH=intraventricular haemorrhage; PVL=periventricular leucomalacia; NS=not stated; WM=cerebral white matter; EPI=estimated predictive interval

When components of the composite outcome were analysed (fig 3), the odds ratio for an association between arterial cord pH and hypoxic ischaemic encephalopathy was 13.8 (6.6 to 28.9, 5.2-36.4, I2=0.0%), between arterial cord pH and seizures was 8.1 (3.0 to 21.9, 0.4-153.6, I2=66.3%), and between intraventricular haemorrhage or periventricular leucomalacia was 2.9 (2.1 to 4.1, 2.0-4.3, I2=0.0%). Only two of nine studies reporting this outcome were not limited to a preterm (<32 weeks) or low birthweight (<2000 g) population. Excluding these two studies did not affect the strength or significance of the association.

Five studies reported on the association between venous cord pH and neonatal morbidity.55 69 78 85 93 Meta-analysis showed a significant association (4.0, 1.2 to 13.3, I2=44.5%). However the predictive interval crossed the line of no effect (0.1-125.3). Four studies examined the association between arterial base excess and neonatal morbidity, which was similar to venous cord pH and also had an estimated predictive interval that crossed the line of no effect (2.5, 1.3 to 4.8, 0.6-10.4, I2=0.0%).61 73 81 91 Owing to the small number of studies reporting this variable, it was not possible to combine the cord pH and base excess to compare the difference between respiratory and metabolic acidosis.

Association between cord pH and cerebral palsy

Seven studies totalling 1117 infants examined the association between arterial cord pH and cerebral palsy.4 5 49 68 74 75 81 Of those, two had an odds ratio point estimate of less than 1.0, but the rest showed an association of low pH with cerebral palsy (table 4). Only two studies gave the criteria used to diagnose cerebral palsy.49 75 The overall association was odds ratio 2.3 (1.3 to 4.2, 1.1-5.0, I2=0.0%l fig 4). It was not possible to explore the threshold effect by subgroup analysis owing to the small number of studies reporting this outcome, therefore bivariate meta-analysis was carried out, which did not change the overall results (2.1, 1.2 to 4.1).

Figure4

Fig 4 Association of low arterial cord pH with cerebral palsy. NS=not stated; EPI=estimated predictive interval

Discussion

Low arterial umbilical cord pH had a strong, consistent, and temporal association with neonatal mortality and morbidity (composite of hypoxic ischaemic encephalopathy, seizures, and intraventricular haemorrhage or periventricular leucomalacia) and long term outcome (cerebral palsy). In all of the associations between arterial cord pH and outcome explored, with the exception of composite morbidity in a low risk population and seizures, the estimated predictive interval suggested that a future study would have the same direction and significance of effect observed. The associations observed are biologically plausible.6 94 95 96

Strengths and limitations of the review

This review provides the best available evidence, at the time of publication, of the association between cord pH at birth and outcomes. The strength of our review and the validity of our inferences lie in the methods used. We complied with existing guidelines for the reporting of systematic reviews of diagnostic22 97 and observational studies evaluating causal association.15 25 We used the most up to date techniques for carrying out and interpreting meta-analysis.35 44 An extensive literature search was done in relevant databases with no language restrictions applied. The studies we pooled had heterogeneity in terms of quality, population risk, threshold of cord pH used, and ascertainment of neonatal outcome. We carried out recommended analyses to tackle this problem, including bivariate and subgroup meta-analysis to take into account the threshold effect,44 metaregression analysis to explore reasons for heterogeneity,39 and component outcome analysis to examine the suitability of the composite morbidity outcome. This did not significantly affect our results. Recent work suggesting the differences between standard and bivariate techniques may be small.98 The observed associations were qualitatively in the same direction, and statistical heterogeneity, when present, arose from variation in strength of associations from study to study rather than opposition in direction of association. We are therefore confident that our inference concerning a causal association between low arterial cord pH at birth and neonatal mortality and a variety of neonatal morbidities merits consideration.

This review has several limitations. Despite our efforts to search for published and unpublished literature, the Harbord test suggested that the meta-analysis of the relation of arterial cord pH and neonatal morbidity may be affected by small study bias. The quality of the primary studies varied. We used both metaregression and subgroup analysis to explore for the effect of this problem, which showed that study quality did not affect our inferences. The poor reporting of population characteristics limited the subgroup analysis according to risk factors: a large number of the papers with an “unselected” population did not fully report characteristics such as birth weight or gestational age, making it difficult to extrapolate our findings to the general obstetric population. Only one paper included in the meta-analysis specified that it was limited to a low risk term population. With regard to the index test examined, only a small number of papers reported base excess, which meant that it was not possible to compare metabolic acidosis with respiratory acidosis. We found that a high base excess was associated with neonatal morbidity, but in clinical practice the pH and base excess level would be considered together in any infant, and we are unable to comment on this. Additionally, we were only able to base our analysis on the thresholds reported in the primary studies, which limited our exploration of the effect of varying pH levels. Some of these issues may be dealt with using meta-analysis of individual patient data.99

Venous cord pH and arterial base excess showed weaker associations with neonatal morbidity than that of arterial cord pH, with estimated predictive intervals that crossed the line of no effect. This analysis was, however, based on a small number of studies, and as no other subgroup analysis could be done, a direct comparison with arterial cord pH was not possible. It seems likely from our results that arterial cord pH has a stronger association with outcome than venous cord pH.

Evidence supporting causality of association of low arterial cord pH with adverse outcomes

Hypoxic ischaemia initiates energy depletion, the accumulation of extracellular glutamate, and activation of receptors, leading to a deleterious cascade of events resulting in neuronal death.100 However, different areas of the brain are susceptible to injury at various stages in development, and the consequences of injury are unpredictable.101 This would support our findings of a strong association with hypoxic ischaemic encephalopathy but a weaker association with cerebral palsy. Only 10% of infants with evidence of hypoxic ischaemic encephalopathy develop cerebral palsy, but the reasons for this are not yet fully understood.102 This issue is further highlighted by the findings of the term breech trial, where long term follow-up did not show any difference in neurodevelopmental delay, despite an apparent increase in neonatal morbidity in the vaginal breech group.103

It is difficult to comment on the specificity with which cord pH is associated with the outcomes examined within our review. All of the outcomes may arise from a variety of causes—for example, in addition to hypoxia, neonatal seizures may be caused by congenital brain anomaly, infection, and metabolic disorders. Although some of the included studies excluded infants with congenital anomaly, others did not specify whether they had excluded from their analysis infants with an outcome related to another cause. This may explain why the strength of association of arterial cord pH with neonatal mortality and morbidity was stronger in the unselected group than in the high risk group. The infants in the high risk group had other factors (for example, prematurity, low birth weight) that may have contributed to the adverse outcome, thus diluting the strength of association. This is supported by the higher number of deaths in the high risk population, with a higher specificity for low arterial cord pH and death in the unselected population.

We explored for the presence of a dose-response relation by examining the strength of association between arterial cord pH and neonatal mortality and morbidity at different pH thresholds. Although the meta-analysis did not show a clear dose-response relation, and the estimated predictive intervals were broad, the association for both neonatal mortality and neonatal morbidity was weakest at the highest threshold. Only one study1 explored the association of arterial cord pH and neonatal mortality at all three thresholds examined. A clear dose-response relation was apparent within this study, with the strongest association at a threshold of 7.00 and the weakest at a threshold of 7.20. Comparison within a study may be more valid than comparisons between studies as it avoids confounding by other study level factors.

Conclusions and practice implications

Our review did not explore the use of cord pH as a prognostic test, and therefore the extent to which our results can be used to counsel parents and target interventions for infants with a low cord pH at birth is limited. To explore this matter, further research is required. A recent review highlighted the current deficits in prognosis research and suggested that when high quality primary studies exist, meta-analysis of individual patient data enables the prognostic value of a test to be assessed at an individual level.104 Our review, however, highlights the lack of high quality primary studies, and therefore a large prospective cohort study with long term follow-up and accounting for potential confounding factors is required. Such a study must include evaluation of outcomes relevant to the individual and society, such as impaired development and use of healthcare and educational resources, before the evaluation of use of cord pH as a prognostic test and its cost effectiveness. This will enable further exploration of the threshold effect and the use of combined cord pH and base excess to predict outcome.

Cord pH is currently assessed in infants believed to be at high risk for neonatal asphyxia. Our results suggest, however, that the strength of association with cord pH and outcome is not limited to this high risk population. Therefore future research should assess the use of cord pH across neonatal populations, particularly exploring the cost effectiveness of testing all neonates. Systematic reviews to establish the association of other neonatal wellbeing measures with long term outcomes are ongoing in our department.

Based on our review, increased initial surveillance of neonates born with a low arterial cord pH, regardless of their clinical condition, is warranted as the odds of complications have been shown to be higher in this group. The avoidance of a low cord pH at birth should continue to be a target for day to day obstetric practice. Our findings justify the use of arterial cord pH as an important outcome measure alongside neonatal morbidity and mortality in obstetric clinical trials. It is difficult to draw strong conclusions on the necessity of long term follow-up for babies with low arterial cord pH, as the observed association with cerebral palsy was in a limited number of primary studies and the association, although statistically significant, was only moderately strong and based on studies of varying quality. We therefore support the use of long term follow-up with childhood assessments in obstetric trials, as this may show developmental anomalies and allow evaluation of cord pH as a predictive test for these outcomes.

What is already known on this topic

  • Umbilical cord pH at birth is frequently used to measure perinatal asphyxia

  • Neonatal and childhood mortality and morbidity, including cerebral palsy, are often attributed to fetal acidosis, as defined by a low cord pH at birth

  • Existing reports of the association between cord pH and adverse outcome are conflicting

What this study adds

  • Low cord pH is substantially associated with neonatal mortality and morbidity and cerebral palsy in childhood

  • These outcomes justify the increased surveillance of infants born with a low cord pH

  • Further research is, however, needed to explore the cost effectiveness of doing this test in all neonates

Notes

Cite this as: BMJ 2010;340:c1471

Footnotes

  • Zainab Laftah helped with data extraction. Richard Riley provided statistical expertise and advice.

  • Contributors: GLM designed the review; extracted, analysed, and interpreted the data, drafted the article; and is responsible for the integrity of the work. She is the guarantor. RKM extracted and interpreted the data, revised the article critically for intellectual content, and approved the final draft for publication. KSK conceived the review, helped analyse and interpret the data, revised the article critically for intellectual content, and approved the final draft for publication.

  • Funding: GLM is funded by the Mary Crosse fellowship, Birmingham Women’s Foundation Trust. RKM is funded by a Medical Research Council/Royal College of Obstetrics and Gynaecology clinical research training fellowship (RRAK12923).

  • Competing interests: All authors have completed the unified competing interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare (1) no financial support for the submitted work from anyone other than their employer; (2) no financial relationships with commercial entities that might have an interest in the submitted work; (3) no spouses, partners, or children with relationships with commercial entities that might have an interest in the submitted work; and (4) no non-financial interests that may be relevant to the submitted work.

  • Ethical approval: Not required.

  • Data sharing: The odds ratios for outcomes included in table 1 but not reported in the paper are available from the author on request.

This is an open-access article distributed under the terms of the Creative Commons Attribution Non-commercial License, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the license. See: http://creativecommons.org/licenses/by-nc/2.0/ and http://creativecommons.org/licenses/by-nc/2.0/legalcode.

References

View Abstract