Neurobehavioral consequences of chronic intrauterine opioid exposure in infants and preschool children: a systematic review and meta-analysis

The systematic review of the literature was conducted accordingly to the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines [46] and the Preferred Reporting Items for Systematic reviews and Meta-Analysis (PRISMA) guidelines [47].

For the purpose of this review, the meanings of the terms ‘opioid’ and ‘opiate’ were considered as largely synonymous, with opioid being used, as it has a broader definition. An Infant was defined as a child up to 2 years old, pre-school child as one between 3 and 5 years of age and a school child as one between 6 and 12 years of age. Neurobehavioral function was defined as ‘growth of perceptual, emotional, intellectual, and behavioural capabilities and functioning during childhood (prior to puberty) which includes development of language, symbolic thought, logic, memory, emotional awareness, empathy, a moral sense, and a sense of identity, including sex-role identity’ [48].

Only studies that recruited opioid users were included in the meta-analysis. Furthermore all trial methodologies, not only RCTs, were considered. Studies had to use a validated diagnostic system and explicitly define whether their participants were opioid/methadone dependent [49, 50].

We excluded studies that recruited mothers who were polydrug users during term pregnancy even though they might have also been taking opioids. Studies that only investigated the immediate effects of opioid use on neonates including neonatal abstinence syndrome and the neurological consequences of opioid exposure were also excluded. Sufficient study statistics not convertible to effect size (d) e.g. means, standard deviation, F, t, X ¹⁸ were also excluded, as well as studies with less than 15 in the total sample size.

Articles were identified using an electronic and hand strategy based search. A computer based search was performed using the following database: Cinahl, EMBASE, PsychINFO and MEDLINE between the periods of January 1995 to January 2012 (17 years). No language constraints were applied. Subject headings originally included ‘child, opioid, prenatal exposure and substance misuse’. (Refer to Additional file 1: Table S1)

This was followed with the term ‘neurobehavioral’ which was subsequently replaced with a succession of terms describing names of a list of cognitive and psychomotor tests and using wild cards.

Standard meta-analytic techniques were employed to this review [51]. Magnitude is indexed with the effect size d that is meant to reflect the degree to which the dependent variable is present in the sample group or the degree to which the null hypothesis is false [52]. In mathematical terms d is the difference between two group means standardised via pooled standard deviation units. Effect sizes (i.e. Cohen’s d statistics) were calculated for each neurobehavioral test and then adjusted for sampling bias [53]. A value of 0.80 is regarded as a large effect size, 0.5 as a medium effect and 0.2 small [54, 55]. Formulae were appropriately adjusted so that all derived statistics informally represented the same direction; that is the same polarity of performance when comparing groups. Negative scores always represented worse performance on the part of the opiate exposed group.

The multi-domain model is the most widely used model of infant-pre-school assessment. The theoretical basis of the model is that the Child Development is an interactively unfolding, continuous process that occurs in several distinct but interrelated domains. Traditionally these domains include (a) motor (fine and gross motor skills), (b) communication (receptive and expressive language), (c) cognition (problem solving skills), (d) adaptive competence (dressing, eating, toileting), and (e) personal-social competence [56, 57]. For this review all relevant test variables were coded into one of three neurobehavioral domains [56, 58].

In keeping with recommendations on meta-analytical research in neuropsychology, previous factor-analysis, where possible, informed the placement of measures into the aforementioned domains. This approach provides an objective alternative to the arbitrary grouping of neuropsychological variables on the basis of face validity or unconfirmed notions held within the existing literature [59]. Unfortunately the factor-analytical studies to date do not encompass all of the neuropsychological measures that were encountered in this comprehensive systematic review. To this end, there was also a reliance on authoritative texts and discussion with experts in the field of neuropsychology and/or cognitive measures to help organise remaining measures [57, 60] and, when necessary, we relied on the classification used by the authors of a given study [5, 42, 43] (Table 1).

To meet the assumption of independence, when multiple test variables in a study contributed to any one neuropsychological domain, the effect size for each measure was assessed separately and then the mean effect size of these measures were combined to assess the overall outcome in the respective area of functioning. Multiple measurements can increase the likelihood of Type 1 errors and so a p value over 0.01 will be interpreted with caution even though analysis will use a significant level of p < 0.05.

Tests for the presence and degree of heterogeneity were conducted using the Q statistic [55] and I² index [61] respectively. However, quantification of heterogeneity is only one component of a wider investigation of variability across studies, the most important being diversity in clinical and methodological domains, and the observed degree of inconsistency across studies with regards to the direction of effects [62]. As different scales were sometimes used by different studies, standardised mean difference (SMD) effect-size estimates were calculated. In case of significant heterogeneity, random effect models were applied [63, 64].

Research with statistically significant results are potentially more likely to be submitted and published than studies with non-significant results. The presence of such publication bias was assessed informally by visual inspection of funnel plots and formally by its statistical analogue, Fail Safe N, according to Orwin [65].

A Fail-Safe N is the number of non significant, unpublished, or missing studies that would need to be added to the meta-analysis in order to change the overall result from significance to non-significance. More than two studies are needed to enable a Fail-Safe N to be calculated.

Eligible research studies comprising a common dependent variable as well as statistics that can be transformed into effect sizes were systematically surveyed. Individual study results (typically means and standard deviations from each group) and relevant moderator variables considered as relevant by previous reviews (e.g. dosage of maternal methadone during pregnancy, gestational age, presence of Neonatal Abstinence Syndrome (NAS), quality of the study, and population studied) were used as moderators if needed during this review. They were abstracted, quantified,coded and assembled into a database and analysed using Comprehensive Meta-Analysis Version 2 [66]. The significance level was p = 0.01 and in Q statistics p = 0.10.

For all review questions, data were extracted by one reviewer and checked by another. Discrepancies were resolved by referral to the original studies and, if necessary, arbitration done by a third reviewer. Duplicate publications were actively screened for and, when found, the latest and most complete report was used. The Effective Public Health Practice Project (EPHPP) quality assessment checklist (amended) was used in this study [67]. For pragmatic reasons no papers were excluded on quality grounds as all papers were weak to moderate (Refer to Additional file 2: Table S2).

Ethical approval and informed patient consent was not required as this study was a literature review and had no direct patient contact or influence on patient care.

Combined searches yielded 1452 references. In total 65 articles were retrieved for further assessment from which 65 studies identified intrauterine exposure to opioids and reported health and developmental outcomes for the opioid exposed children. From these studies, 8 articles were found to investigate the cognitive, psychomotor and behavioral outcomes in opioid exposed infants, pre-school and school children when compared to healthy non-opioid exposed controls. Only 5 studies could be further utilised in this meta-analysis since 1 study measured motor rather than psychomotor skills [68] and 2 other studies had small sample sizes of < 15 [69, 70].

Furthermore 2 studies tested the same cohort during infancy and pre-school periods [6, 71] and another study tested the cohort during pre-school and school periods [72]. Considering that only one study measured outcomes during the school period [72] it was decided that further analysis should concentrate on the infancy and pre-school periods. During these periods there were 4 studies comparing opioid exposed infants with controls [6, 71, 73, 74] and 3 studies comparing opioid exposed pre-school children with controls [6, 71, 72]. In Hans et al. [74] infants in the cohort were tested at 1 and 2 years old allowing two observational points (Figure 1).

All studies were case controlled observational studies conducted with a population living in urbanised and low socioeconomic communities exposed to either heroin or methadone.

The global quality assessment reported four studies as of moderate quality and one as weak. The assessment for analysis performed was moderate for all studies but all other domains were reported as either weak or moderate (see Additional file 2: Table S2).

Cohort characteristics for the 4 studies comparing opioid exposed infants with controls describe a total number of 218 individuals tested compared to a total of 205 non opioid exposed controls. The mean infant age was 14.1 months (1.2 years). Cohort characteristics for the 3 studies comparing opioid exposed pre-school children with non opioid exposed controls describe a total number of 224 individuals tested compared to a total of 231 non opioid exposed controls. The mean age of the pre-school children tested was 50.7 months (4.2 years). General and specific characteristics of the included studies are shown on Tables 2&3.

There were six effect size measures possible (3 for the infant cohort and 3 for the pre-school cohort groups) from the selected studies. There were no effect sizes identified as greater than 2× inter-quartile range (25 and 75 percentile) from the nearest quartile (outliers) [75].

In this first ever quantitative review of the research literature on the neurobehavioral outcomes as a result of intra-uterine opioid exposure in infants and pre-school children the meta-analysis has determined what abilities, if any, were reliably found impaired across studies when compared with non-opioid exposed controls. Our findings indicate no significant impairments in cognitive, psychomotor or observed behavioural outcomes for chronic intra-uterine exposed infants and pre-school children, although in all domains there is a trend to poor outcomes in infants/children of opioid using mothers (Table 4).

The result of this systematic review is in accordance with Whitham [42] who conducted an open label non randomised flexible dosing longitudinal study with results showing that children prenatally exposed to illicit heroin and/or methadone did not differ to non-exposed infants and other children in cognitive, psychomotor and caregiver rated temperament outcomes. As a recent review observed, the conflicting results of traditional systematic reviews on this subject could be that most children in these studies were exposed to other drugs in addition to opioids such as methadone [5]. Another explanation given to the conflicting results may be that various studies were conducted using different neurobehavioral tests at different ages of development. Prenatal opioid exposure may affect children’s cognitive and psychomotor performance differently at different ages resulting in neurobehavioral outcomes that might improve or worsen over time [5]. This analysis could only be conducted during the two early stages of a child’s development (infancy and pre-school children). It was not possible to conduct a similar analysis on children aged between 6–12 years due to the presence of only one study meeting the criteria for analysis.

The studies included in the meta-analysis had differences in the measurement of exposure to opiates. Most studies involved the use of illicit heroin wherein the dosage is highly variable and based on a large number of factors, while two of the studies involved methadone prescribed in a controlled environment. None of the studies commented on the use of illicit drugs and alcohol which may also have a bearing on the outcomes of interest.

The results of our analysis must be cautiously interpreted bearing its limitations in mind. The inclusion criteria used for this meta-analysis was very stringent so as to exclude neurobehavioral effects as a result of intra-uterine maternal polydrug use and the potential confusion of neurobehavioral outcomes associated with the neonatal abstinent syndrome and other opioid withdrawal presentations during the neonatal period (Jones et al., 2010). The main limitation of using a meta-analytic technique is the small number of primary studies available for analysis and also their small sample size. This limits the generalisability of the result [75] and the small sample size of individual studies means they may miss an increased risk of the occurrence of relatively rare outcomes like ADHD, autism or psychosis. The quality assessment of the individual studies would also raise some concerns about the generalisability of the findings as all were reported as of moderate or weak quality.

Meta-analysis tends to present results as composite scores for broad neurobehavioral functions using different neuropsychological tests. This is a convenient way to summarise findings but it combines data from tests potentially exploring different neuropsychological processes (e.g. memory tests assessing immediate or delayed recall, learning or recognition) and possibly generating results of questionable theoretical relevance. This study attempted to minimise this by utilising neuropsychological domains agreed by consensus and used in systematic reviews on chronic substance use effects [24, 76].

Even though the meta-analysis grouped together studies that used the same rating scales on a cohort who clearly were exposed to opioids, it was not possible from the information presented in the studies to exclude other confounding effects such as dosage of maternal opioid use during pregnancy [5], timing during pregnancy when the fetus was exposed to opioids with potential resulting behavioral teratogenicity [77], the differential effects of exposure to different opioids [42] and/or gender specific neurobehavioral influences [71], other illicit drugs including cocaine and alcohol use during pregnancy [5]. Each included study reported that there were no polydrug using mothers within their cohorts but it is not possible to be certain that this was achieved.

This meta-analysis helps in supporting certain clinical observations in this population. The observed, if any, neurobehavioral outcomes in infants and pre-school children prenatally exposed to opioids are very often attributed to substance exposure. However it is important to examine the contribution of other influences on a child’s development. Ongoing maternal depressive illness is correlated with poorer cognitive and motor development and increase in teacher and parent rated behavior problems in pre-school children [78, 79]. Poverty and low socio-economic status is inversely related to children’s developmental performance [80, 81]. A study examining the relationship between birth weight and cognitive functioning among children in South Australia indicated that cognition at 2 years of age was significantly related to birth weight [82]. However the magnitude of the association attenuated over time became non-significant in childhood. Factors that became significantly associated with neurobehavioral outcomes included low socio-economic status, low maternal IQ, poor quality of the home environment and child’s lead exposure. Overall it is increasingly becoming evident that the risk factors that can predict poor neurobehavioral outcomes is not the drug fuelled lifestyle or actual substance exposure during pregnancy but the presence of multiple, inter-related and weighted variables cumulatively influencing neurobehavioral outcomes [3]. The risk factors were: maternal mental health, maternal attitudes toward parenting and maternal child–parent interaction, maternal education, parental occupation, minority status, stressful life events and family size with not one risk factor contributing exclusively to one cognitive or other neurobehavioral outcomes.

In many countries, including the UK, pharmacological maintenance with methadone is the first line of treatment for pregnant opioid dependent women [17, 83]. Whilst treatment with methadone during pregnancy results in fewer complications for both mother and infant when compared with the use of illicit opioids such as heroin, its use in pregnancy is associated with high rates of neonatal abstinence syndrome [84, 85]with its treatment involving using another opioid, morphine [86]. Prenatal exposure to opioids also significantly increases the risk of low birth weight and small head circumference as shown in the cohort of children in the studies selected for this review. However this analysis did not observe any increased risk in neurobehavioral problems in opioid exposed infants and pre-school children compared to non-exposed peers suggesting that there is no neurobehavioral sequelae to the chronic prenatal and, if treated for NAS with opioids, also postnatal, exposure to opioids with any effects being short term and/or reversible.