Pediatric traumatic brain injury (TBI) is an important public health issue, as approximately 280 per 100,000 children and adolescents incur a TBI annually worldwide (
1). More than 80% of these injuries are considered to be mild TBI (
1). In the United States, TBI is a leading cause of pediatric morbidity and mortality (
2); post-TBI sequelae can include new-onset psychiatric disorders, neuropsychological deficits, poor school performance, and deficits in social competence and adaptive function (
3–
6). Long-term psychiatric outcomes for adults with a history of pediatric TBI are less well known.
Several studies have focused on psychiatric outcomes among youths with TBIs of all severities; new-onset psychiatric disorders (referred to here as “novel psychiatric disorders” [NPDs]) occur in 50%−60% of children and adolescents who have experienced severe TBI (
7), 5%−9% in control subjects with orthopedic injuries, and 10%−36% with mild to moderate TBI (
3,
8). Because mild TBI constitutes the vast majority of all pediatric TBIs, it is worth noting the findings of a systematic review of psychiatric complications (
9). In the review by Emery et al. (
9), the investigators found that psychiatric problems were more common when mild TBI was associated with hospitalization, when assessment occurs earlier postinjury (suggesting that problems resolve over time), when there are recurrent previous mild TBIs, among individuals with preinjury psychiatric disorder, when outcomes are based on retrospective recall in contrast to prospective studies, and when the comparison group is uninjured children rather than injured children (e.g., with orthopedic fractures). In our prospective studies of hospitalized children with mild to severe TBI, depending on when participants were assessed, NPDs were significantly associated with the following injury and preinjury variables: lower fractional anisotropy in bilateral frontal lobe, temporal lobe, uncinate fasciculi, and centrum semiovale tracts; lifetime psychiatric disorders; family function; adaptive function; family psychiatric history; and socioeconomic status (
8,
10–
13).
To date and to our knowledge, there are no long-term psychiatric interview prospective studies of pediatric TBI, and the longest follow-up assessment of prospective psychiatric interview studies was 2–2.25 years (
3,
11,
14). There are three published prospective investigations of long-term outcomes (at 16–23 years old) in adulthood of pediatric TBI, which examined behavioral domains of function in unselected cohorts (
15–
21). These studies have identified associations between severity of TBI and long-term social functioning, personality, educational, vocational, and mood-related outcomes. Besides injury severity variables, lower socioeconomic status and a less intimate family environment were associated with poorer emotional perception at the follow-up (
19), and preinjury adaptive function was related to long-term internalizing problems (
18). Furthermore, significant associations were found between domains of function, including social communication, externalizing behaviors, and emotion perception (
20). The aforementioned long-term studies were limited by high attrition rates (34%−69%), and other long-term outcome studies are limited due to their referral bias (
22–
25).
The goal of the present long-term extension of our prospective longitudinal study from 2 years (
10–
13) to 24 years postinjury was to fill a scientific void by examining the natural history, occurrence, phenomenology, and biological and psychosocial predictive variables for long-term psychiatric outcomes following pediatric mild to severe TBI that required acute hospital treatment. The age group that we studied (ages 29–39) is important because while follow-up to this point does not constitute an entire life course, prospective birth cohort studies suggest that individuals in this age range have reached the highest risk period for onset of many psychiatric disorders (
26–
29). In this study, we hypothesized that NPDs current at long-term follow-up would be significantly related to some of the following six domains examined at the baseline assessment: severity of injury and the following preinjury variables, lifetime psychiatric disorder, behavioral and adaptive function, family psychiatric history, family function, and socioeconomic class and intellectual function.
METHODS
Study Recruitment (1992–1994)
Children and adolescents ages 6–14, consecutively hospitalized between 1992 and 1994 for mild to severe TBI, were eligible for recruitment in this prospective longitudinal study, which was approved by the institutional review board at the University of Iowa. Additional eligibility criteria included having completed a head computerized tomography (CT) scan during initial hospitalization and having English as a primary language. Youths were excluded if they had a penetrating TBI, loss of consciousness greater than 3 months, prior TBI requiring hospitalization, history of child abuse, history of intellectual disability, or history of another neurologic or serious medical illness. Written informed consent was obtained from parents, and youths provided assent after they were able to demonstrate decision-making capacity to do so.
During the course of the recruitment period, 87 patients met eligibility criteria; 50 participants enrolled in the study and completed at least the psychiatric assessment component of the baseline evaluation. The most common reason provided for not enrolling in the study was that children with mild TBI and their parents believed that participation was unwarranted due to observation that the child had returned to baseline following the injury. Participating and nonparticipating youths did not differ in terms of age, sex, race, socioeconomic status, preinjury psychiatric disorder, or treatment, but they did differ in terms of TBI severity: participants were more likely to have severe TBI compared with nonparticipants. Overrepresentation of severe TBI in study participants may also be related to the main study site being a tertiary care hospital with a large catchment area (
13).
We updated the severity classification of the participants to conform with current definitions for descriptive purposes. Severe injury was defined by a lowest postresuscitation Glasgow Coma Scale (GCS) score <8, moderate injury by a lowest postresuscitation GCS score of 9–12, complicated mild injury by a score of 13–15 with an intracranial lesion or depressed skull fracture on initial CT scan, and uncomplicated mild injury was defined by a lowest postresuscitation GCS score of 13–15 unaccompanied by abnormality detected on the initial CT scan. The initial cohort of participating children (N=50) was categorized by injury severity as having experienced uncomplicated mild (N=26), complicated mild (N=5), moderate (N=4), and severe (N=15) TBI. Coma duration in subjects was as follows: no coma (N=22), <15 minutes (N=10), 1–24 hours (N=6), 1–2 days (N=2), 2–7 days (N=6), and >7 days (N=4). Subjects’ posttraumatic amnesia (PTA) durations were no PTA (N=6), <15 minutes (N=1), 15–59 minutes (N=8), 1–24 hours (N=18), 1–2 days (N=3), 2–7 days (N=5), and >7 days (N=9). One subject remained in a vegetative state at all follow-up points and was therefore dropped from analyses. Causes of injury included motor vehicle accidents (N=10; 20%); bicycle or car accidents (N=6; 12%); falls from bicycles (N=9; 18%); other falls (N=10; 20%); sports and recreation (N=7; 14%); pedestrian-motor vehicle accidents (N=2; 4%); motorcycle/all-terrain vehicle accidents (N=2; 4%); and other (N=4; 8%).
Predictive Domain Models Derived From Baseline Assessments (1992–1994)
Comprehensive neurologic, psychiatric, family, and adaptive functioning assessments were conducted at baseline (mean=14 days [SD=13] postinjury) to assess severity of injury and degree of preinjury functioning. The assessments measuring distinct aspects and risk factors of NPD following pediatric TBI were grouped in six domains (severity of injury, lifetime psychiatric disorder, behavior/adaptive function, family psychiatric history, family function, and socioeconomic class and preinjury intellectual function).
The severity of injury domain consisted of three items: lowest postresuscitation GCS score (
30), Traumatic Coma Data Bank (TCDB) categorization (
31), and normal/abnormal initial day-of-injury CT scan. The GCS is a standard measure of acute brain injury severity; scores range from 3 (unresponsive) to 15 (normal) (
30). Lowest postresuscitation score for each participant was obtained from the medical record. A pediatric radiologist and a pediatric neuroradiologist independently classified the initial day-of-injury CT scans as either showing an intracranial traumatic lesion or not. The radiologists additionally classified the CT scans according to the TCDB categorization, which incorporates the degree of brain edema and focal lesions into a single severity rating on a scale from 1 to 6 (
31). Interrater reliability was excellent (
13).
The lifetime psychiatric disorder domain was composed of the single variable, lifetime psychiatric disorder, and was based on baseline psychiatric assessments, which in all cases were conducted by a board-certified adult and child and adolescent psychiatrist (J.E.M.). Standardized, semistructured psychiatric interview assessments were used, including the Schedule for Affective Disorders and Schizophrenia for School-Age Children (
32,
33), and psychiatric diagnoses were based on DSM-III-R criteria. The Neuropsychiatric Rating Schedule (NPRS), which is designed specifically to identify symptoms and subtypes of personality change due to TBI (
34), was also administered at baseline.
The behavior/adaptive function domain included the following baseline variables: the adaptive behavior composite standard score, derived from the Vineland Adaptive Behavior Scale (VABS) interview conducted by a trained research assistant (
35); and the total raw score on the parent-completed Pediatric Behavior Scale (PBS), a behavioral rating scale designed specifically for use with pediatric neurological and other medical disorders (
36). The VABS and PBS scores constituted the behavior/adaptive function variable.
The fourth domain, family psychiatric history, was based on the Family History Research Diagnostic Criteria Interview (
37,
38). The interview was conducted at baseline, with parents acting as the informants. We summarized family ratings for first-degree relatives only and for a combined grouping of first- and second-degree relatives on a 4-point scale of increasing morbidity (
39); both ratings comprised the family psychiatric history variable.
The fifth domain, family function, included baseline family assessment measures collected by both interview and questionnaire methods. The McMaster Structured Interview of Family Functioning is a research interview of the family and is based on the McMaster model of family functioning (
40). The interviewer used the Clinical Rating Scale to rate each of six domains and global family functioning on a 7-point Likert scale, where higher scores indicated better function. In addition, the Family Assessment Device questionnaire was completed by family members at least 12 years of age (
40), and the scores were used to calculate a mean global functioning dimension score for each family, where higher scores indicated worse function. The family function variable consisted of two ratings: the global family functioning score from the interview and from the questionnaire.
The socioeconomic class and preinjury intellectual function domain included four predictors. Socioeconomic class was assessed using the Four Factor Index, where higher scores indicated lower socioeconomic class (
41). Measures used to assess intellectual function included teacher-report of preinjury intellectual ability and academic achievement on the PBS, along with preinjury national percentile rank for vocabulary on the Iowa Tests of Basic Skills, the latter of which is highly correlated with verbal IQ (
42). Scores on the above four items made up the socioeconomic class and preinjury intellectual function variable.
Follow-Up at 24 Years (288 Months) Postinjury (2016–2018)
With approval from the institutional review boards of the University of Iowa and University of California, San Diego, participants who completed at least the baseline psychiatric assessment from 1992 to 1994 (N=50) were recontacted from 2016 to 2018 and invited to participate in a 24-year postinjury assessment that included psychiatric evaluation. At this long-term follow-up, we studied 43 (86%) participants with TBI in person, as well as the sibling of an original participant who died approximately 3 years prior to this wave of the study, and the parent of another participant who did not respond to our invitation to participate (N=45). The sibling and parent of these two original participants displayed in-depth knowledge of the probands. The study design allowed for the recruitment of a significant other (e.g., parent, partner, friend) for participants with TBI because of the possibility that awareness of deficits may be compromised after TBI; a total of 31 significant others participated. The mean interval from injury to long-term assessment was 23.92 years (SD=2.17).
Excluded participants included one original enrollee, who was still alive but who remained in a vegetative state and was therefore not eligible to participate. Four originally enrolled participants (males, N=2; females, N=2; severe TBI, N=2; mild TBI, N=2) declined participation in the 24-year follow-up.
Psychiatric assessments were repeated at long-term follow-up by J.E.M. using the Mini-International Neuropsychiatric Interview (MINI) (
43), and psychiatric diagnoses were made based on DSM-5 criteria. We used the NPRS again to diagnose personality change due to TBI (
34). In cases where individuals with TBI and a significant other participated, best-estimate diagnoses were assigned based on integration of self-report and significant other report (
44). An interrater reliability study was conducted by a board-certified psychiatrist (E.T.), which relied on ratings of videotaped interviews of TBI subjects, significant others when applicable, and healthy control subjects. Control subjects (N=45) were recruited and completed a single cross-sectional assessment for additional analyses (
45), which are outside the scope of the current article. The board-certified psychiatrist (E.T.) was blind to group status (TBI versus control), and rated the videotaped interviews of every seventh TBI (N=7) and control participant (N=7). Interrater reliability for diagnoses was excellent (κ=0.962). There was perfect agreement on diagnoses in 12 out of 14 (86%) cases, as well as agreement on 54 out of 56 (96%) specific diagnoses that were recorded.
Current NPD (NPD-C) was the outcome variable of interest at the long-term follow-up; it was defined as a psychiatric disorder not present prior to TBI that was present at the assessment 24 years postinjury. Examples of NPD-C among participants included an individual with no preinjury psychiatric disorder who had attention deficit hyperactivity disorder (ADHD) at the 24-year assessment and an individual with only ADHD preinjury who had generalized anxiety disorder at the 24-year assessment.
Statistical Methods
To investigate the association between each of the six prespecified injury and preinjury domains (severity of injury, lifetime psychiatric disorder, behavior/adaptive function, family psychiatric history, family function, and socioeconomic class and preinjury intellectual function) and NPD-C at the 24-year follow-up, for each domain the
k variables (
k=1–4) were reduced to a single composite variable. The domain variable was the first principal component (PC1) from the principal component analysis (PCA) applied to the
k variables in the domain, after centering and rescaling to unit variance, so that all domain variables receive equal weight in the PCA. PC1 is the projection of the rescaled
k-dimensional response vectors onto the direction that captures the most variation in the data, and thus provides the best one-number summary of the
k variables. The proportion of variance in the predictors explained by PC1 was reported. This dimension reduction is dictated by the low power due to the small sample size combined with a binary outcome. Logistic regression analyses were then conducted separately for each domain, with the PC1 as the predictor and NPD-C as the outcome. The association between the PC1 predictor and the outcome was evaluated using the likelihood ratio test. The strength of association was measured using R
2, the proportion of variability explained for logistic regression based on the average improvement in deviance per observation (
46), and an analogous measure for variable-specific partial R
2. As a last step, a multipredictor logistic regression examined the domains associated with NPD-C at 24 years in an adjusted analysis. The final multipredictor model was determined via stepwise backward model selection, with a threshold of a p value <0.15 of inclusion in the model. As a result of the small number of events, the starting multipredictor model only included the PC1 composite variable from domains that were associated with NPD-C at the 24-year outcome at a p value <0.15 in single-predictor analyses (
47). All statistical analyses used the R statistical program, version 4.0.2 (
48).
DISCUSSION
The primary finding from this prospective longitudinal 24-year follow-up study of individuals who experienced a TBI requiring hospitalization at ages 6–14 years was that NPD-Cs in adulthood following pediatric TBI are independently predicted by preinjury lifetime psychiatric disorder and by greater severity of injury.
Preinjury lifetime psychiatric disorder has previously been shown to influence NPDs in the first 2 years after pediatric TBI (
3), including earlier time points in the current prospective longitudinal study, particularly during the first 3 months, and in the second postinjury year (
11,
13). The current finding suggests that this variable exerts its influence not only in the short term but also over more than two decades. The durability of this finding is striking, given that the participants were only aged 6–14 years when their vulnerability trait of preinjury psychiatric disorder was documented. It may be useful to think of this phenomenon as limited “behavioral reserve” increasing risk of psychiatric complications akin to the concept of limited “cognitive reserve” increasing the risk of cognitive problems after pediatric TBI (
50).
The finding that severity of TBI independently significantly predicts long-term psychiatric outcome after pediatric TBI extends existing findings of a dose-response relationship from the first 2 years postinjury (
3,
11,
51). This, too, is striking given that there are many intervening influences present over a span of 24 years of growth and development of children and adolescents that could mitigate the effect of injury severity.
Tying the independent significant effects of preinjury lifetime psychiatric disorder and severity of injury on NPDs may be additionally appreciated by inspection of the data. The predictive effect of preinjury lifetime psychiatric disorder is most evident in the mild to moderate TBI group, where individuals with a preinjury lifetime psychiatric disorder have significantly increased frequency of NPDs, compared with those without premorbid psychiatric history. In the severe TBI group, most individuals experienced onset of NPDs following their injury, regardless of premorbid psychiatric history. These findings suggest that severe TBI may overwhelm the protective effect of lack of preinjury lifetime psychiatric history, while mild to moderate TBI may not.
The rate of NPDs in this sample is high (53%) but consistent with other studies of shorter follow-up duration (
3,
11,
52). The only disorder that is specific to TBI is personality change due to a general medical condition, which in this instance is TBI. Such heterogeneity in NPDs is typical in pediatric TBI studies (
3,
11,
52). In the absence of a prospectively studied non-TBI control group, it is not possible to say conclusively which disorders occur at a significantly higher rate than expected. However, the rates of novel ADHD (30%) and novel generalized anxiety disorder (16%) appear to be higher than what might be expected in a general population sample (3.5% and 5.0%, respectively) (
53,
54).
The current findings should be interpreted in light of the following limitations. First, the sample was relatively small (N=49 to N=50), and findings require replication in larger samples. However, attrition after 24 years was only 8%. Second, the longitudinal design did not include control subjects; we therefore cannot know whether the obtained model predictive of NPD-C is unique to TBI patients. However, control subjects were recruited for the long-term assessment only and will be included in additional analyses (
45). Third, most prospective longitudinal TBI studies have the limitation of requiring a retrospective assessment of preinjury variables. However, baseline assessments were completed as soon as possible after the initial injury, within a mean of 14 days (SD=13). Fourth, the psychiatrist (J.E.M.) assessing the participants was not blind to group affiliation. Fifth, the psychiatric diagnoses during the first 2 years of follow-up applied DSM-III-R criteria, while the long-term follow-up diagnoses applied DSM-5 criteria. It is unclear how this may have changed the analyses, because preinjury lifetime psychiatric disorders were examined according to DSM-III-R and NPDs at long-term follow-up according to DSM-5. Sixth, in this exclusively hospitalized cohort, approximately half the participants had an uncomplicated mild TBI, a patient group that may be expected to have a low rate of NPD. However, a key finding in a systematic review on pediatric mild TBI is that a history of hospitalization for the mild TBI is associated with increased risk of psychiatric sequelae (
9). Seventh, notwithstanding the in-depth knowledge displayed by the sibling of the deceased subject and the parent of the subject who did not respond to the invitation to participate, data generated directly from the probands would be preferable. Finally, consistent with the distribution of race in Iowa, all but one participant was White, which potentially limits generalizability to more diverse populations.
There were several notable strengths of the study. First, this is, to our knowledge, the only long-term prospective longitudinal psychiatric interview study of pediatric TBI. Second, the psychiatric assessment itself was a strong aspect of the study methodology. All psychiatric assessments from baseline through 24-year follow-up were completed by the same psychiatrist, who is board-certified in general psychiatry and in child and adolescent psychiatry. Diagnoses were made only in the face of true impairment, which requires clinical judgment. Assessment of most participants with TBI benefited from the input of significant others and served to address potential under-reporting due to lack of awareness of impairment in this group (
55). Excellent interrater reliability was documented for psychiatric diagnoses with another board-certified psychiatrist who viewed videotapes of 16% of all interviews and who was blind to TBI versus control group affiliation. Third, attrition was only 8% at 24-year follow-up. The attrition rate was much lower than the three other long-term studies (16 to 23 years) that have examined behavioral domains of function in unselected cohorts, which had attrition rates of 34% to 69% (
15–
21). Fourth, the predictive models used for NPDs in the TBI cohort were based on comprehensive and clinically relevant biopsychosocial data derived from multiple sources, including the participants, parents, and teachers.
In conclusion, both preinjury lifetime psychiatric disorder and severity of injury are significant and independent predictors for long-term psychiatric complications following pediatric TBI. The results of the current study suggest that surveillance for the development of problems can be targeted in order to allow for early intervention to be delivered to affected individuals. For such intervention to be clinically meaningful, it will have to account not only for the novel disorders but also for the pre-existing disorders in this high-risk subpopulation of children with TBI constituting up to half of cases (
7). Future studies involving longer term follow-up could reveal whether psychiatric disorders persist into early and late middle age and whether the injury and preinjury risk factors change. Future analyses of data from the present study will compare adults exposed to pediatric TBI with healthy control subjects matched to age, socioeconomic status, and sex, and will also analyze cognitive function, occupational function, adaptive function, and neuroimaging correlates (
6,
56) and their relationships with psychiatric disorders at long-term follow-up.