Necrotizing enterocolitis (NEC) occurs in approximately 10% of extremely low birthweight (ELBW, ⩽1000 g) infants¹ and surgery is required in 50–70% of these infants.¹ Mortality for surgical NEC remains high (up to 40%). There are no specific therapeutic agents for NEC, the therapeutic aim is organ support, together with resection of gangrenous bowel.

Despite advances in neonatal intensive care improving survival of newborn infants, the long-term outlook for infants with NEC, both those treated conservatively and those requiring surgery, is not well defined. Recent studies have highlighted improved neurodevelopmental outcomes for ELBW infants,² but note that preterm birth contributes disproportionately to neonatal morbidity and subsequent neurodevelopmental disability.³ Recently, several studies have addressed the aetiology of poor neurodevelopmental outcome of premature or ELBW neonates^4,5 and investigated the effects of interventions on neurodevelopmental outcomes.⁶

A number of studies have assessed neurodevelopmental outcomes of very low birthweight (VLBW, ⩽1500g) or ELBW neonates with NEC,^7–9 many with small numbers of patients or from single centres.^10,11 Some have concluded that surgical NEC is not associated with a greater risk of poor neurodevelopmental outcome than prematurity alone.¹² Systematic reviews with or without meta-analysis allow the reader to judge the potential for generalisation to different populations.

Our aims were, therefore:

(1) to characterise the neurodevelopmental outcome of neonates with NEC;

(2) to investigate whether NEC increases the risk of neurodevelopmental impairment in VLBW neonates;

(3) to investigate whether stage of disease or need for surgery increase the risk of poor neurodevelopmental outcome.

METHODS

Literature search strategy

A systematic review of peer-reviewed literature, including PubMed, ISI Web of Science, Embase and the Cochrane database, was performed. The search strategies used the following keywords and MeSH terms: necrotizing enterocolitis, necrotising enterocolitis, outcome, neurodevelopmental outcome, neurodevelopment, neurodevel* and neurol*. Citation searches were performed in ISI Web of Science and hand searches of reference lists were performed. Studies were included if they met a predefined list of inclusion criteria (table 1) and were selected independently by CR and SE. There were no disagreements concerning eligibility of papers; however, there were several overlapping cohort studies from NIHCHD, so the two reviewers decided between themselves which was the most appropriate paper to include. Data were collected for a predefined list of variables and entered into Cochrane Collaboration Review Manager 4.2 independently by CR and SE.

Variables and definitions

The variables of interest were defined as follows. Necrotizing enterocolitis was defined as Bell’s stage II or above¹³ diagnosed clinically, radiologically or histologically; medical management included antibiotic usage, withdrawal of feeds and any resuscitative measures; surgery included laparotomy or peritoneal drainage. In one paper,¹⁴ infants were classified as stage II or stage III NEC; here, stage III was taken to imply surgical treatment, and stage II medical. Cerebral palsy was defined as a non-progressive neurological disorder characterised by abnormal limb tone (in one or more limbs) and inability to control movement and posture.⁹Visual impairment was defined as blindness or visual deficit in at least one eye. Deafness was defined as hearing impairment requiring hearing aids in at least one ear. Mental developmental impairment (MDI) and psychomotor developmental impairment (PDI) were defined as scores of <70 on the relevant Bayley II scale.¹⁵ MDI scores assess memory, problem solving, discrimination, classification, language and social skills. PDI scores assess muscle control, walking, running, jumping, use of writing implements and imitation of hand movements. In both scores, the normal range (corrected for age) is 100±15 (mean ±1 SD) so a score of <70 lies >2 SDs below the mean. Infants who are so impaired that testing is impossible are given a score of 49. In studies where Bayley scores were not assessed, patients with impaired mental development (including delayed language development) were grouped with MDI <70 and patients with psychomotor impairment grouped with PDI <70. Neurodevelopmental impairment (NDI) is a composite outcome based on cerebral palsy, bilateral blindness or deafness, PDI score <70, or MDI score <70. This gives an overall estimation of the number of infants with developmental deficit. Papers where Griffiths Developmental scales were used were assigned to the NDI group if the Griffiths GQ score was less than 2SD below the mean corrected for gestational age. In one paper,¹² cognitive impairment was marked by requirement for speech therapy and psychomotor development by performance on a self-developed drawing test. These results were grouped with MDI and PDI respectively.

Very low birth weight (VLBW) infants are those with birthweight ⩽1500 g and a subset of these are extremely low birth weight (ELBW) infants whose birthweight is ⩽1000 g.

Statistical comparisons

Review Manager 4.2 (the Cochrane Collaboration) was used to analyse data. Data are presented as odds ratio (95% confidence intervals) with p values shown for Z test for overall significance and I² statistic¹⁶ for heterogeneity.

RESULTS

Searches performed according to strategy identified 1039 papers; 14 fulfilled inclusion criteria, 13 comparing VLBW or ELBW neonates with NEC to infants of similar age and gestation without NEC,^{8,9,10,11,12,14,17,18,19,20,21,22,23} and one study²⁴ comparing surgically treated infants with NEC to those treated medically so was included in this analysis only. Three of the 13^8,9,23 were excluded because of overlap with a more recent study.²¹ The study characteristics are shown in table 2. All studies were retrospective case control or cohort studies.

Altogether, 7843 children were included in the meta-analysis, of whom 821 had NEC. Median follow-up was 20 months (range 12 to 156). The median percentage of patients followed up was 90%.

The meta-analysis results are summarised in table 3. 20% of neonates with NEC developed cerebral palsy, 3% developed visual, 3% hearing, 36% cognitive and 35% psychomotor impairment. All outcomes were worse in neonates with stage III disease or who required surgery (surgical NEC group). There were significant (p<0.05) differences in the odds ratio of cerebral palsy (1.55 (1.19 to 2.03)), visual impairment (2.31, (1.04 to 5.11)), cognitive impairment (1.44 (1.24 to 1.68)) and psychomotor impairment (1.72 (1.35 to 2.19)) between infants with NEC and infants who did not have NEC. Neurodevelopmental impairment occurred in 45% of children who had had neonatal NEC, compared to 35% of children who had been VLBW or ELBW but did not have NEC (χ² p = 0.0003). Overall, infants with NEC were significantly more likely to have neurodevelopmental impairment (OR 1.58 (1.25 to 1.99), p = 0.0001) (table 3). These data are shown as a Forrest plot in fig 1A. When we analysed the cohort studies separately from the case-control studies, both analyses yielded similar odds ratios to the combined data, inevitably with wider confidence intervals (case-control 1.7 (0.92 to 3.21); cohort 1.56 (1.21 to 2.00)).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

 (A) Forrest plot: Neurodevelopmental Impairment NEC vs. No NEC. (B) Forrest plot: Neurodevelopmental Impairment surgical NEC vs. medical NEC. The Forrest plot is the graphical output of a meta-analysis. Each study is represented by a dot with 95% confidence intervals represented by horizontal lines. The size of the dot is proportional to the relative size of the study. The diamond represents the combined outcome, its width representing the 95% confidence interval. (n = number affected; N = total number of patients; OR =  odds ratio; CI =  confidence interval)

When the surgical NEC group was compared with those managed non-surgically (medical NEC group), they were significantly more likely to have cerebral palsy (OR 2.74 (1.44 to 5.21) p = 0.002) and psychomotor impairment (OR 1.85 (1.07 to 3.21), p = 0.03). There were non-significant trends towards an increased risk of hearing impairment and cognitive impairment. Overall the surgical NEC group was 2.34 times more likely to have neurodevelopmental impairment than the medical NEC group (95% CI 1.51 to 3.60, p = 0.0001). These data are shown in fig 1B/table 4. There were no significant differences between the medical NEC and No NEC groups for neurodevelopmental impairment (OR 1.02 (0.73 to 1.44), p = 0.89), thus supporting the finding that surgical NEC is specifically associated with a poorer neurodevelopmental outcome than either medically treated NEC or prematurity alone. There were no significant differences for any of the other outcomes between medical NEC and No NEC (results not shown), although there were fewer studies to compare for the other outcome measures.

DISCUSSION

A small number of papers have been published describing the neurodevelopmental outcome of survivors of NEC. Some compare children who had NEC with those born ELBW who did not have NEC, others compare the outcomes of babies treated conservatively with those receiving surgery. Meta-analyses can be usefully performed on non-randomised studies,²⁵ in which case caveats must be drawn in the interpretation of the results. Clearly, in comparing ELBW infants who had NEC with those that did not get NEC, we are comparing two populations rather than two groups of individuals randomly allocated from the same population.

Limitations of this type of study include the presence of confounding factors – it is impossible to quantify the effect of factors that are related to both NEC and neurodevelopmental impairment (eg prematurity); this is the most important threat to the validity of results from cohort studies.²⁵ Epidemiological studies are prone to publication bias as only statistically significant findings may be published.²⁶ Other problems which are likely to be greater in meta-analyses of retrospective, non-randomised studies include inadequate reporting of methods, variation in study design, variation in inclusion criteria and variation in presentation of results.²⁶ Another potential problem with meta-analysis of non-randomised studies is combining data from different types of study design. In this meta-analysis, we have combined data from cohort and case control studies, although where there were enough studies of each type to undertake two separate analyses, the results were consistent. We chose to combine studies by using odds ratios rather than relative risks as whereas it is appropriate to refer to odds ratios for both cohort and case-control studies, relative risk cannot be calculated for case-control studies. Interpretation of odds ratios in the same way as relative risks, however, can overestimate effect sizes where the event is frequent in the index group (as in our study) and the odds ratio is large²⁷ (although most odds ratios we obtained were <2.5). Another potential problem with the approach that we have used is the fact that different papers have used different tools for measurement of neurodevelopmental outcome, although most studies that we included used one of two scores–Bayley scores or Griffiths Quotients.

Despite these limitations, we believe that our results show that neonates who have NEC can expect a significantly worse neurodevelopmental outcome that those VLBW neonates who do not have NEC. Although the risk seems small (OR 1.6), the risk of poor neurodevelopmental outcome is already great in VLBW infants, so any significantly increased risk is important. This results in 45% of NEC survivors having poor neurodevelopmental outcome. If a baby with NEC needs surgery, their neurological outcome becomes even worse (OR 2.34) compared to medically-treated infants. This finding does not imply that one should not treat a neonate with perforated NEC surgically; it only implies that babies sick enough to require surgery have a worse outcome than those not sick enough to require surgery. In this study, we only included infants that had Bell’s stage II or III NEC, as the signs and symptoms for stage I NEC are too non-specific to make this a meaningful group for comparison. However, even the diagnosis of stage II and stage III NEC can be equivocal and we acknowledge that this is another potential shortcoming of this study.

Factors causing poor neurodevelopmental outcome in premature infants are complex. Neurogenesis, neuronal maturation and synaptogenesis contribute to brain development in the 2^nd and 3^rd trimester of pregnancy²⁸ so extremely premature infants have a deficit in brain maturation. Postnatal continuation of these processes requires adequate nutrition, and may be adversely affected by many factors associated with prematurity and surgery. Indeed, a specific negative impact of NEC on cerebral growth is suggested by the finding that NEC is a significant predictor of smaller head circumference in ELBW infants.²⁹

Damage to existing cerebral tissue is another likely contributor to poor neurological outcome. This can be caused by infection or inflammation, respiratory insufficiency, hypotension, acidosis, fluctuations in glycaemic control, disseminated intravascular coagulation, anaesthesia and transport. In particular, infection and sepsis, which may result from intestinal perforation and/or gangrene, have been highlighted as independent risk factors.^9,30 They cause increased cytokines, which are implicated in the pathogenesis of periventricular leukomalacia, a major determinant of adverse neurological outcome.³¹ Surgery for NEC also causes a cytokine surge which could be responsible for white matter damage, and it would be interesting to determine if a surgical treatment with a lower degree of tissue trauma (peritoneal drainage) results in a better neurological outcome than laparotomy. Poor outcome may be independently associated with surgery, with or without NEC–28% of surviving ELBW infants who required surgery of any sort have a poor sensorieneural outcome.³² In favour of the severity of disease, rather than surgery per se causing poor neurodevelopmental outcome, Adesanya et al.⁷ found that neurodevelopmental impairment was worse in infants who had perforated NEC than those with spontaneous bowel perforation.

Most of the studies had good rates of follow up suggesting that the results are representative, although a bias in overestimation of adverse outcomes has been found previously, as children with poor neurodevelopmental outcome are more likely to be retained for follow-up.^7,33–35 We analysed cohorts that were born over a wide period of time (1975–2002), which includes babies born before surfactant and high frequency oscillation were introduced. Although the mortality of ELBW neonates has improved, the effects on neurodevelopmental outcome are less well defined.^2,5,36

Our findings on the overall rates of poor neurodevelopmental outcome in ELBW infants are comparable to recent population-based prospective studies which demonstrated that 10% of extremely premature infants had severe motor disability, 2% were blind, 3% were deaf and 49% had some disability³⁷ and that these disabilities persisted into childhood.³⁸ The largest study in our review, that from the National Institute of Child Health,²¹ was the most complete in terms of available data, therefore in comparisons where only a few studies were included there is a risk that the results of smaller studies are subsumed. However, this paper reported on outcomes from sixteen different centres in the US and many of the other papers are from different countries, so pooling the results makes the results more applicable to patients outside the USA.

If neonates with NEC are at increased risk of adverse neurodevelopmental outcome, two questions arise: can we predict which babies will have an adverse outcome, and can we do anything to prevent it? Logistic regression and neural networks had low sensitivity and specificity for predicting major handicap in ELBW infants³⁹ and although severely abnormal ultrasound has a high predictive value for adverse neurological outcome,²³ up to 30% of ELBW survivors with normal neonatal ultrasound are found to have impairment at 18–22 months.⁴⁰ MRI at school age has recently been suggested to be better predictor of IQ and motor performance than ultrasound,⁴¹ but whether neonatal MRI is a better predictor than ultrasound is not known.

Now that we know that infants with NEC are at increased risk of neurodevelopmental impairment we should ensure that neurodevelopmental assessment plays a part in the follow-up of any new treatment. It is also important that parents and professionals are aware of the increased long term risks in these critically ill neonates.