Mr Mikael J Sømhovd, Department of Psychology, University of Copenhagen, Øster Farimagsgade 2a, 1353 Copenhagen K, Denmark. E-mail: email@example.com
Aim To determine if adolescents who are born very preterm (<32wks; of gestation) and/or with very low birthweight (VLBW; <1500g) have a higher risk of experiencing clinically significant anxiety problems.
Method We used a systematic review and meta-analysis. We searched the databases ISI Web of Knowledge, PubMed, PsycNET, Educational Resources Information Center (ERIC), Latin American and Caribbean Literature on the Health Sciences (LILACS), and Virtual Health Library (VHL) with equivalent search expressions (from the databases’ inception to June 2011). Also, we screened reference lists of identified articles. We selected case–control studies of adolescents 11 to 20 years old who were very preterm/VLBW and had a matched reference group born at term with normal birthweight that reported a validated anxiety outcome measure. For data extraction, two authors independently reviewed titles, abstracts, and full articles identified through the searches. Subsequently two authors independently extracted data.
Results We included six studies with 1519 adolescents (787 very preterm/VLBW, 732 comparisons). The general risk of developing clinically significant anxiety problems was nearly doubled (p<0.05) in the very preterm/VLBW population (OR 2.27, 95% confidence interval 1.15–4.47). The overall prevalences were 9.9% in the very preterm/VLBW group and 5.5% in the comparison group.
Interpretation Those born very preterm/VLBW have an increased risk of developing clinically significant anxiety problems in adolescence.
• It determines that there is an increased risk of clinically significant anxiety problems among those born with very low birthweight.
• It increases awareness of anxiety as a possible outcome of very low birthweight.
• It proposes further targeted research to reveal causal properties of the overrepresentation of anxiety.
Survival rates for those born with very low birthweight (VLBW; <1500g) and very preterm (VPB; <32wks of gestation) have steadily increased in recent decades.1 Somatic and neurological disorders,2 difficulties in the perceptual, cognitive, and intellectual domains,3–8 and emotional and personality disorders 9–13 are all more common in children, adolescents, and young adults born preterm. There is no standard intervention that prevents potential developmental disorders, and the purpose of follow-up programmes is to support these high-risk children and their families, to inform appropriate interventions, and to promote supportive care when possible.
Several systematic reviews have revealed more externalizing disorders, attention difficulties, and cognitive difficulties among those born with low birthweight than those born at term,14–16 and a recent review has suggested that those born preterm and with low birthweight (<2500g)17 have nearly a three times greater prevalence of psychiatric diagnoses than those born with normal birthweight and gestational age. However, another review of internalizing symptoms, which include those of anxiety and depression, found only a weak association with birthweight.18
Anxiety is the most prevalent psychiatric disorder in the general population.19–22 It has a profound impact on the quality of life,23,24 significant economic consequences,25 and is associated with decreased learning ability.26,27 Fortunately, anxiety is also effectively treatable.28,29 Within a developmental psychopathological framework, the period of adolescence is significant as many psychiatric symptoms, including anxiety, appear for the first time in this period, and several psychiatric disorders have their average age of debut during adolescence.30 The risk of having at least one psychiatric disorder during adolescence, particularly internalizing disorders, is also generally high compared with later in life.31
After adolescence, anxiety has high heterogenic continuity, often ‘transitioning’ into depression in adulthood. Evidence from clinical studies suggests that exposure to early life stress is associated with neurobiological changes in children and adults, which may underlie psychopathology, including anxiety.32
Studies assessing the association between birth status and behavioural and psychiatric outcomes usually encompass a wide range of outcomes, but targeted studies investigating anxiety in the VPB/VLBW population do not exist. Several of the methods used in measuring outcomes in the general psychiatric studies of VPB/VLBW cohorts, however, include subscales measuring anxiety.
In an effort to disentangle anxiety from other behavioural and psychiatric disorders and to synthesize results, we systematically accumulated the evidence. Studies of adolescent VLBW/VPB cohorts including an anxiety measure were assembled to assess whether adolescents who are born very preterm and/or with very low birthweight are at greater risk of developing anxiety than adolescents born at term.
Study eligibility criteria
We included case–control studies of children and adolescents assessed between 11 and 20 years of age that compared a group with VPB and/or VLBW with a matched group with normal birthweight and gestational age. In those studies of high methodological quality, comparison populations were matched by age and sex. Eligible studies also had to exclude children with severe learning disability,* cerebral palsy, or major psychiatric disorders, to reduce confounding by comorbidity to severe neurodevelopmental impairment. We excluded studies published only as abstracts, conference proceedings, and those involving fewer than 30 children or adolescents, and studies of low methodological quality (i.e. accumulating fewer than five stars on the Newcastle–Ottawa scale).33
We used the cut-off scores for a clinically significant level of anxiety as defined for each specific measure. To obtain as many eligible studies as possible for the synthesis, binary outcomes were accepted as most studies with eligible cohorts reported clinical-level anxiety symptoms as cut-off scores. In addition, the outcome measures were required to be well-established instruments in use for research and/or clinical purposes to measure anxiety. We, by default, accepted the validity of the cut-offs selected by the authors of the original papers. All types of informant were accepted; however, self-reports were our a priori preference.
ISI Web of Knowledge, PubMED, PsycNET, Educational Resources Information Center (ERIC), Latin American and Caribbean Literature on the Health Sciences (LILACS), and Virtual Health Library (VHL) were searched in June 2011 with equivalent search strings. All databases were searched without language limitations and from the databases’ inception to June 2011. The general search string (as deployed in the PubMed) was submitted as follows: ‘((“Infant, Low Birth Weight”[Mesh] OR “Infant, Very Low Birth Weight”[Mesh] OR “Infant, Extremely Low Birth Weight”[Mesh] OR “Premature Birth” [Mesh] OR “Obstetric Labor, Premature”[Mesh] OR “Premature Birth”[Mesh] OR “Infant, Premature”[Mesh]) AND (“Anxiety”[Mesh] OR “Anxiety Disorders”[Mesh] OR “Depression” [Mesh] OR “Depressive Disorder” [Mesh] OR “Psychiatry” [Mesh] OR “Child Psychiatry” [Mesh]))’.
Mesh is an abbreviation for Medical Subheadings and is the thesaurus of the US National Library of Medicine. The exact search strings according to the respective thesauruses from all databases are available on request from the corresponding author.
One author (MJS) and a student assistant independently reviewed titles of all citations returned from the search. If either reviewer considered a citation as potentially relevant, or if titles were not sufficiently clear, abstracts were also reviewed. If either reviewer deemed the citation relevant, the full-text article was retrieved and reviewed by the same two assessors.
Study quality and extraction of data
Study quality and risk of bias was assessed using the Newcastle–Ottawa scale,33 which was rated only by MJS. Two authors (MJS and BMH) independently extracted predefined data of years of study, study design, bias risk, characteristics of participants, and outcome measures. When possible, published data were used for the analysis, and authors were contacted for relevant data if the published material was not sufficient.
Synthesis of results and statistical analysis
Statistical analyses were conducted using the software program RevMan 5.1 (The Cochrane Collaboration, Nordic Cochrane Centre, Copenhagen, Denmark).34 We used odds ratios (ORs) to pool our dichotomous data counting children with clinically significant anxiety symptoms among the respective groups; hence, estimates are to be considered as unadjusted. A random effects model was chosen to account for both the random variability within studies and random variability in effects between studies owing to expected heterogeneity. Both clinical and statistical heterogeneity was considered. The chi-square test for statistical heterogeneity was deemed significant at p<0.10, and was considered a basic indication of heterogeneity. We also tried to estimate the level of heterogeneity, which was defined as low, moderate, and substantial if inconsistency (I2) was between 0 and 25%, 25 and 75%, and 75 and 100% respectively.
Risk of bias across studies and sensitivity analysis
Publication bias was assessed both visually by examining the symmetry of a precision funnel plot, and formally by a tau-b correlation test35 and a weighted linear regression of effects.36 Sensitivity analyses considering age category (≤14y and >14y), sampling frames, and birthweight category were performed.
Initially, 303 non-duplicated titles and abstracts were identified (Fig. 1). After an initial screening, 273 publications were discarded as they clearly did not meet eligibility criteria. Thus, we reviewed the full texts of 30 studies and screened them for additional citations from their reference lists. No additional articles were identified. After the full-text review, 24 more studies were excluded, the most common reasons for exclusion being age of children, failure to report outcomes of interest, or unclear study design. Among these 24, five studies reported continuous effect sizes and hence binary data were requested from the authors. All authors answered and four of the datasets were acquired, one author replying that their dataset was old and inaccessible. However, only one of the extra datasets was included,37 because it was controlled against large normative samples coherent to the scale in use. Our hope was that it would comprise a level of detail enabling an extraction of a VLBW/VPB ‘free’ sample, making it suitable for inclusion in our data synthesis. Upon receiving the data, however, this was unfortunately not possible. As proxy and self-report about children have a low correlation, the only study exclusively containing self-reported data38 was excluded to avoid the potential theoretical heterogeneity, despite self-reported measures being the a priori preference. A total of six studies were therefore finally included,37,39–43 with 1519 children comprising 787 in the VLBW/VPB cohorts and 732 matched comparisons.
Characteristics of studies
The children’s ages ranged between 11 and 20 years. As summarized in Table I, the included studies reported data from five different measures. The studies all originated from developed countries, and all were published in English. Two of the six were from national cohorts,40,43 the remainder were hospital or regional cohorts. Further, two of the studies comprised cohorts of only those born with extremely low birthweight (<1000g). All six included studies were proxy reported by parents, or parents and teachers. The prevalence of anxiety symptoms for the individual studies was between 2.2 and 25% for the index cohorts, and 1.4 to 10.5% for the comparison groups. Prevalence details of individual studies are presented in Table I. There is, however, no apparent systematic dependency between prevalence and OR. All six studies achieved the minimum criterion of five stars on the Newcastle–Ottawa scale.
Table I. Central characteristics of included studies
Mean age (y)
Percentage above cut-off anxiety level (ELBW or VLBW vs comparison group)
aThe Child Behavior Checklist.61bThe Development and Well-Being Assessment.64cThe Child and Adolescent Psychiatric Assessment.60dSchedule for Affective Disorders and Schizophrenia.63eYoung Adult Behavior Checklist.62 ELBW, extremely low birthweight; VLBW, very low birthweight.
Compared with normal birth groups, children and adolescents between 11 and 20 years old with VLBW and/or VPB have a significantly increased risk of developing anxiety (OR 2.27, 95% confidence interval [CI] 1.15–4.47; p=0.018). We found moderate statistical heterogeneity (I2=55.94% and χ2=11.34; p=0.05).
Clinical heterogeneity stemming from differences in the various measures was also considered acceptable under a random model as all data included were extracted from scales validated against constructs from the International Classification of Diseases, Ninth Revision (ICD-9) or the DSM-IV.44 The results of the synthesis can be viewed graphically in Figure 2 wherein results from the individual studies can also be viewed.
Risk of bias across studies
We visually examined a precision funnel plot. All selected studies fell within the 0.01 limit and were spread fairly symmetrically, which is relevant to the synthesis estimate. Neither the Begg’s test (tau=0.2, p=0.57), nor the Egger’s regression asymmetry test for zero intercept (β=2.03, p=0.35) was statistically significant, which both indicate that publication bias is not reflected in the current selection. Hence, a trim and fill procedure was deemed redundant.
Sensitivity analysis of the two age categories showed noticeable differences, as excluding the three cohorts comprising those younger than 15 years old40,43,44 (n=691, VLBW=346, comparisons=345) lowered the OR to 1.4 (95% CI 0.5–3.9), whereas excluding the cohorts of those above 15 years old increased the OR to 3.5 (95% CI 1.9–6.6). The rationale for setting the age cut-off at 15 years was that before 15 many would still be pre-pubertal. Additionally, a potential issue of heterogeneity in the synthesis was the inclusion of the two studies comprising extremely low birthweight cohorts.37,43 Excluding both of these studies (n=1069, VLBW=519, comparisons=550) increased the OR to 2.42 (95% CI 1.0–5.85). Yet another potential issue was the sampling frame. Included studies comprised two hospital cohorts,39,42 two regional cohorts,28,37 and two national cohorts.25,39 Excluding the two hospital cohorts increased the OR to 2.85 (n=761, VLBW=410, comparisons =351; CI 1.6–4.97). Excluding the two regional cohorts resulted in an OR of 2.38 (n=1302, VLBW=682, comparisons=620; p<0.06; CI 0.97–5.86). Finally, excluding the two national cohorts gave an OR of 1.95 (n=975, VLBW=482, comparisons=493; CI 0.7–5.41).
In this systematic review and meta-analysis, we found that adolescents born with very low birthweight and/or very preterm have more than a twofold risk of having anxiety symptoms in the clinical range compared with matched adolescents with normal birth status. However, it should be noted that the overall rates both in individual studies and in the synthesis are relatively low.
To our knowledge, this is the first meta-analysis directly assessing the relation between birth status and anxiety symptoms. Other systematic reviews and meta-analyses have indeed investigated cognitive and behavioural outcome of birth status, and suggest that internalizing disorders, which include anxiety, are overrepresented among school-age VLBW/VPB children (see, for example, Bhutta et al.14). Further, previous analysis has synthesized studies targeting school-age children whereas our analysis is based on studies targeting adolescents older than 11 years. The distinction between studies on children and adolescents in this context is important as adolescence is a crucial period when considering developmental psychopathology.30,31,45
Splitting up our age range, as we did in the sensitivity analysis above, suggests a noticeable impact on the predictions, with the younger age category demonstrating more clinically significant anxiety symptoms than the older cohorts. However, the analysis of the older age cohorts also indicated more anxiety in the VLBW/VPB groups compared with peers. Yet, as the sensitivity analysis suggests, caution should be taken when interpreting the results for these age effects. Sensitivity analysis suggested ignorable changes in ORs when excluding data from the extremely low birthweight cohorts. Our decision to include the full VLBW continuum in our analysis is also supported by studies reporting no differences in the risks of anxiety within the VLBW group.46 Finally, sensitivity for differences of sampling frames indicated no noticeable impact from blending/mixing hospital, regional, and national cohorts.
Even though four studies contained relevant data, and reported continuous effect sizes, they were excluded from our analysis as comparable data could not be acquired. Dahl et al. (2006) reported more overall behavioural problems, which included significantly more anxiety in a follow-up VLBW cohort between 13 and 18 years of age (n=99) compared with two normative samples.47 Levy-Shiff et al. reported a significant association between birth status and overanxious disorder in a cohort of 13- to 14-year-olds born with VLBW (n=90)48 compared with matched peers. Also, Saigal et al. reported a significant difference between 16- to 19-year-olds with extremely low birthweight (n=141) and matched peers on anxious disorders.49 Conversely, Hille et al. reported no significant differences in anxiety between 19-year-olds with VLBW/VPB (n=656) compared with scale norms.50 Obviously, a major limitation of our analysis is that we were unable to include data from the large study by Hille et al. This study reported equal levels of psychopathology in the VLBW/VPB cohort compared with peers; however, it suggested a ‘more prominent internalizing behaviour’ and difficulties in establishing social contacts in the former group. Accordingly, these studies mainly support our findings; however, the partly negative result in the study by Hille et al.50 would possibly moderate the results from our synthesis.
As conclusions in the individual studies fluctuated, and as our own analysis suggested that at least two of the studies included37,42 yielded ORs below 1, a meta-analysis of the data as opposed to reviewing the results of single studies seemed rational. No discriminative characteristics of these two studies were apparent, and our publication bias analysis concluded that the included studies reflected no apparent file drawer issues. Comparability of the different cut-offs will always be a major limitation in this type of analysis. Even though the measures we used are all well validated to measure anxiety symptoms, they will include a considerable amount of measurement error for latent constructs like anxiety. Therefore, we chose a random effects model to account for both the random variability within studies and random variability in effects between studies caused by expected heterogeneity. The cut-off scores set in the individual studies may also be a source for bias.
The suggested higher prevalence of clinically significant anxiety symptoms among VLBW/VPB adolescents may be due to several factors. Firstly, anxiety may be a consequence of brain dysfunctions directly due to birth circumstances. Although children with clear-cut brain damage or major mental impairment were excluded from the studies, subtle brain anomalies among eligible children may still underpin dysfunctions in circuits leading to anxiety, and have developmental consequences. In particular, impairments of pre-frontal cortical function might be an explanation of the increased rates of anxiety as the pre-frontal/amygdala circuit is involved in anxiety and anxiety development.51 The pre-frontal/amygdala hypothesis in our context is also bolstered by research suggesting that impairments in executive and attention functions are overrepresented among the those born preterm;39,52 these functions are dependent on the pre-frontal functioning.
Potential dysfunction in neuroendocrine regulation central to the fear response and control might also be an explanation for increased anxiety among those with VLBW/VPB. High levels of perinatal stress are associated with preterm birth53 as well as with low birthweight,54 and high cortisol levels in infancy are also a precursor for anxiety later in life.55 Maternal anxiety is both associated with preterm birth and anxiety in children.
The link between maternal anxiety and state of distress is not only mediated through biological pathways, but has a substantial environmental component.56 Therefore, it seems reasonable to expect that environmental factors may be influenced by dramatic birth circumstances like preterm birth, or birth of a child with very low birthweight, as studies show that parents after preterm births report higher levels of symptoms of post-traumatic stress syndrome,57 which acts as a predicator for the development of anxiety in their children.58
Considering the coexistence of anxiety symptoms among adolescents born preterm and the high load of both biological and environmental risk factors, it seems warranted to target parents alongside an increased awareness of anxiety and distress challenges among children. Targetting the parents should include not only educating them about their offspring’s potential special needs, but also meeting the parents’ own needs during birth and in its aftermath while still in hospital settings, and in a longer perspective.
Strengths and limitations
The strengths of our meta-analysis include the comprehensive searches and the screening process. The inclusion criteria applied were relatively narrow, thereby lowering the number of possible confounders and strengthening the assumption of theoretical homogeneity to the extent possible when working with observational studies. Additionally, the use of the Newcastle–Ottawa criteria for a pre-synthesis assessment of selected studies indicated reasonable quality of the included studies. It is a limitation that we could not perform a subgroup analysis for sex owing to incomplete data in most of the studies. Anxiety disorders are known to have different prevalence rates in males and females, with females generally reporting more anxious symptoms than males.59 Another major limitation is that the synthesis of observational studies poses a greater risk of selection bias than, for example, randomized controlled trials. Also, one could speculate that anxiety in this population is not necessarily an equivalent phenomenon with anxiety in the general population owing to a possibly larger biological component, which compromises the comparability to references.
This systematic review indicates that children with VLBW/VPB have a higher incidence of clinically significant anxiety levels. The mechanisms are likely to involve biological vulnerability factors as well as environmental factors known to increase anxiety. Hence, care should be taken to meet the immediate needs of the child and parents in the dramatic context of preterm birth as well as their long-term needs. Furthermore, more targeted research is needed on the link between anxiety and preterm birth. In particular, research is needed to determine whether anxiety has a different aetiology among those born very preterm or at very low birthweight compared with the general population.
North American usage: mental retardation.
We acknowledge Maja Killengreen for her participation in the initial search and selection of studies. We also thank Abigail Cora Smith and Nicole Lønfeldt for proofreading the final manuscript.