Introduction
Traumatic brain injury (TBI) is the most common cause of death and disability in children and adults (Ghaiar, 2001). Each year, approximately .4% of children will sustain a TBI that can lead to lasting complications (Max, 2014). The public health burden of TBI is especially high in children and adolescents (Thurman, 2016). Rates of disability are unknown but may be around 20% for those who are hospitalized (Thurman, 2016). Some estimates suggest that around one-third of individuals will sustain a TBI between birth and age 25 (McKinlay et al., 2008). Approximately 5.3 million Americans are currently living with TBI related difficulties (CDC, 2013).
The leading cause of TBI in children under 5 years old is falls, and motor vehicle accidents are the leading cause for adolescents over 15 years old (Thurman, 2016). The majority of TBIs are handled in an outpatient setting, while around 38% are admitted to the hospital for observation (McKinlay et al., 2008). Severity of TBI is usually assessed in the acute phase, normally within a few hours to a few days after the injury. Severity of injury is often indicative of long-term outcomes, with more severe injury predicting worse outcomes (Thurman, 2016). The Glasgow Coma Scale is one of the most commonly used ratings of TBI severity, and rates individuals based on eye opening responses, verbal responses, and motor responses (Jones, 1979). A lower score indicates a more severe injury. Other indicators of severity include the incidence and duration of loss of consciousness and posttraumatic amnesia (Thurman, 2016). 80-90% of pediatric TBI in the United States are mild (Faul et al., 2010).
TBI often leads to alteration in brain function that contribute to difficulties in cognition or sensorimotor difficulties (Bruns & Hauser, 2003). Some studies have found that injury severity was related to IQ following TBI (Anderson et al., 2000). This study found that more severe injury predicted a lower IQ. A younger age at injury was similarly linked to worse prognosis for the severe TBI group only. This study with IQ emphasizes that there are potentially devastating outcomes following pediatric TBI that are possibly moderated by severity of injury and age at injury. Additional studies have demonstrated that, even for children who had above average premorbid functioning, TBI can severely alter the developmental trajectory (Gamino et al., 2009; Babikian et al., 2015). Children who sustain TBI may appear to demonstrate cognitive recovery back to baseline, but often plateau and fail to meet developmental milestones (Gamino et al., 2009). This phenomenon has been coined “neurocognitive stall.” Additionally, many children and adolescents also experience significant, chronic behavioral problems following TBI (Babikian et al., 2015).
Children with TBI are more likely to be socially ostracized and to experience general psychopathology than their peers (Yeates et al., 2013). Some estimates suggest that there may be more frequent pre-existing psychiatric disorders in mild TBI than in severe TBI (Bloom et al., 2001). It is possible that children who are highly impulsive or more likely to engage in risk-taking behaviors often have diagnoses of externalizing disorders such as ADHD (Gerring et al., 1998). Epidemiological data following a birth cohort supports that children who have TBI are more likely to have these behavioral difficulties pre-injury (Bijur et al., 1988). A novel psychiatric disorder (NPD) refers to the development of any psychiatric disorder following TBI that was not present prior to injury (Max, 2014). The rates of NPD have varied from 10-100% of children included in studies, and may be higher in children with severe TBI at 24 months after an injury (Brown et al., 1981; Max et al., 1997; Max et al., 1998). One of the problems in determining most accurate estimate of NPD after TBI is the variability in diagnostic approaches. The gold standard for NPD assessment is for trained professionals to conduct a semi-structured interview that is supported by questionnaires to generate the psychiatric diagnoses (Max, 2014). However, some studies only include questionnaires that are only completed by a parent. An additional problem is the variability in control group characteristics, with some studies only comparing psychiatric difficulties across TBI severity groups and not including a typically developing control group. The current trend is to include a group of children who sustained a traumatic injury that did not include the head, often referred to as orthopedic injury controls. [Dr. Burlingame, I am working on finding literature to address the efficacy and rationale for including these control groups, and how it may affect generalizability. As of yet, I have not had to time to do an effective literature search for this data.] Perhaps due to the heterogeneous damage from a TBI, NPDs commonly include personality change (not a personality disorder), internalizing disorders (such as anxiety and depression), and externalizing disorders (such as ADHD and oppositional defiant disorder) (Bloom et al., 2001; Max et al., 2000).
Specific Aims and Hypotheses
The objective of this meta-analysis is to synthesize the available evidence concerning psychiatric status following pediatric traumatic brain injury. Given this objective, the following aims and hypotheses are made:
Aim 1: Determine the overall effect size of psychiatric status in children with TBI compared to non-head injured peers. These will be examined by injury severity group.
Hypothesis 1: A significant, detectable dose-response effect will be observed in psychiatric status in relation to injury severity.
Aim 2: Determine the effect of time since injury on psychiatric status effect size following pediatric traumatic brain injury.
Hypothesis 2: Time since injury will act as a significant moderator of novel diagnosis of psychiatric disorders and contribute to the severity of internalizing and externalizing problems. Specifically, there will be an increase in psychiatric outcome variables related to an increase in time since injury.
Aim 3: To determine the effect of age at injury on psychiatric status effect sizes following pediatric traumatic brain injury.
Hypothesis 3: Age at injury will act as a significant moderator of novel diagnosis of psychiatric disorders and contribute to the severity of internalizing and externalizing problems. Specifically, a younger age at injury will predict worse outcomes and more severe ratings or a higher incidence of novel psychiatric disorders.
Methods
The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines were used to ensure clarity and ease of reproducibility (Moher et al., 2009).
Identification and Selection of Source Studies
Electronic databases Web of Science, PubMed, and PsycInfo were searched for relevant articles. Web of Science was searched first, followed by PubMed, and PsycInfo was the last database searched. A search was conducted to find articles relating to pediatric TBI and psychiatric status using the search terms “(pediatric OR child OR adolescen*) AND (TBI OR “head injury” OR “brain injury” OR “traumatic brain injury”) AND (psychiat* OR depress* OR anx* OR externalizing OR internalizing) NOT (adult*)”.
Inclusion Criteria
Peer-reviewed articles published through March 2017 were considered for inclusion. A lower limit was not set on the date of publication, but all studies that met inclusion criteria were published between 1998 and 2016. Studies had to be published in a peer-reviewed journal and be written in English. Studies had to include children from ages 2-17 years and contain a control group of non-head injured peers to be included in the analyses. All studies had to contain (1) means and standard deviations (or standard errors), (2) correlation coefficients,(3) t or Z values, (4) F rations, or (5) odds ratios in order to compare psychiatric status between groups. While we considered studies that did not report means and standard deviations but that included (2-5) such that effect sizes could be calculated, all of the studies that met inclusion criteria reported means and standard deviations or standard errors. Thus, t, Z, F values nor odds ratios were used to compute effect sizes.
Data Extraction
After identifying studies that meet the inclusion criteria described above, two trained members of the research group independently extracted relevant data including author names and publication year, sample size, and statistical results related to measures of psychiatric status. Glasgow coma scale (GCS) scores, age at testing, and age at injury were also be extracted from source studies. The extractors discussed any discrepancies in the extracted data to correct discrepancies. In studies that included multiple time points, on the most recent data (i.e., those at the latest time point) were extracted.
Group Categorization by Injury Severity
Groups were categorized using the average GCS score for the group provided in the article. An average GCS score of 3-8 was considered severe, scores between 9-12 were considered moderate, and scores ranging from 13-15 were considered mild. These groupings of injury severity are consistent with a majority of articles included in the peer-reviewed literature.
Statistical Analysis and Data Synthesis
We used Comprehensive Meta-Analysis version 3.0 (Biostat, Englewood, NJ) to calculate effect sizes, homogeneity statistics, and meta-regressions. This software was also used to address publication bias using a fail-safe N and funnel plots. Rosenthal's Fail-safe N estimates the number of studies that would be required to bring the p-value for any statistically significant effect size above .05. A funnel plot shows the relationship between study size or precision and effect size. We plotted effect sizes of the source studies on the x-axis and the standard errors on the y-axis. There should be a symmetrical distribution around the mean effect size if publication bias is not present. Asymmetry demonstrated by "missing" studies with large standard errors but small effect sizes in the context of small studies with large effect sizes suggests publication bias.
Aim 1. In order to address aim one of this meta-analysis, which is to determine the overall effect size of psychiatric status in children with TBI compared to non-head injured peers, a summary Hedges g effect size for psychiatric status was calculated using a random-effects model from each individual source study. Effect sizes were obtained for each severity group (mild, moderate, and severe) and Q and I2 tests were conducted to determine whether there were significant differences in effect sizes between severity groups.
Aims 2 and 3. The second and third aims of this meta-analysis were to determine whether age at injury and time since injury predict psychiatric status following TBI. In order to address these aims, a meta-regression analysis will be completed for each injury severity group (mild, moderate, and severe). The effect size for psychiatric status served as the dependent variable in the meta-regressions. For each of the regressions, age at injury and time since injury acted as the independent variables. GCS was used as a covariate in these regression analyses. These regression analyses will enable us to determine whether or not age at injury and time since injury vary in a linear manner with the effect size for psychiatric status.
Results
Search Results
We reviewed the titles and abstracts of articles potentially meeting inclusion criteria based on the search terms resulting in 9725 full articles for further review of abstracts (Web of Science = 1412, PubMed = 4426, and PsycInfo = 3887). We retrieved full reports from 63 articles (Web of Science = 29, PubMed = 21, and PsycInfo = 13) for critical analysis. Of these, 15 met inclusion criteria (Table 1).
Meta-Analysis
Internalizing Problems Across Group Severity
Overall internalizing problems based on parent-report measures had an effect size of -.560 (95% CI [-.874, -.246]; p<.001). The confidence interval for the effect size is small, thus providing a tight range for the actual effect size. The classic fail-safe N test showed that an additional 443 studies with non-significant results would be needed to bring the p values for the parent-reported measures to above .05. A Q-test analysis revealed that there were significant effect-size differences between severity groups Q(12)=140.025, p<.001.
Internalizing Problems by Group Severity
Analysis of the mild TBI condition included ten studies that reported internalizing symptoms. The overall effect size for this group was -.624 (95% CI [-1.090, -.158]; p=.009). Only four studies included measures for the moderate severity group. The effect size for this group was -.238 (95% CI [-.452, -.024]; p=.029). Ten studies provided information for the severe TBI group. This effect size was -.923 (95% CI [-1.341, -.504]; p<.001).
Externalizing Problems Across Group Severity
Overall externalizing problems based on parent-report measures had an effect size of -.520 (95% CI [-.729, -.311]; p<.001). The confidence interval for the effect size is small, thus providing a tight range for the actual effect size. The classic fail-safe N test showed that an additional 351 studies with non-significant results would be needed to bring the p values for the parent-reported measures to above .05. A Q-test analysis revealed that there were significant effect-size differences between severity groups Q(11)=51.769, p<.001.
Externalizing Problems by Group Severity