Traumatic Brain Injury and Bipolar Psychosis in the Genomic Psychiatry Cohort

Abstract

Approximately three million individuals in the United States sustain traumatic brain injury (TBI) every year, with documented impact on a range of neurological and psychiatric disturbances including mania, depression, and psychosis. Identification of subsets of individuals that may demonstrate increased propensity for posttraumatic symptoms and who may share genetic vulnerabilities for gene-environment interactions can enhance efforts to understand, predict, and prevent these phenomena. A sample of 11,489 cases from the Genomic Psychiatry Cohort (GPC), a NIMH-managed data repository for the investigation of schizophrenia and bipolar disorder, was used for this study. Cases were excluded if TBI was deemed causal to their mental illness. A k-means clustering algorithm was used to probe differences between schizophrenia and bipolar disorder associated with variables including onset age, hallucinations, delusions, head injury, and TBI. Cases were separated into an optimum number of seven clusters, with two clusters including all cases with brain injury. Bipolar disorder with psychosis and TBI were significantly correlated in one cluster in which 72% of cases were male and 99.2% sustained head injury. This cluster also carried the longest average period of unconsciousness. This study demonstrates an association of TBI with psychosis in a subset of bipolar cases, suggesting that traumatic stressors may have the ability to impact gene expression in a vulnerable population, and/or there is a heightened occurrence of TBI in individuals with underlying psychosis. Further studies should more closely examine the interplay between genetic variation in bipolar disorder and suseptibility to psychosis following TBI.

INTRODUCTION

An estimated three million individuals in the United States sustain traumatic brain injury (TBI) every year [Silver et al., 2011], with studies demonstrating as high as 38% of males and 24% of females experiencing at least one TBI before the age of 25 [McKinlay et al., 2008]. TBI may encompass a loss, decrease, or alteration of consciousness, loss of retrograde or anterograde memory, neurological deficits, and/or an intracranial lesion, following the application ofan outside force [Group, 2009]. The adverse effects of TBI are far-reaching and numerous; associations have been established between TBI and development of unprovoked seizures, premature mortality, endocrine dysfunction, Parkinsonism, Alzheimer’s disease, social dysfunction, long-term unemployment, and mental illness [War et al., 2009; Whelan-Goodinson et al., 2009].

A relationship between TBI and psychosis dates to the writings of Kraepelin, who in 1919 proposed that brain injury may cause dementia praecox [Kraepelin, 1919]. Estimations of psychosis risk following TBI range from 0.7% to 26.5% [Lishman, 1968; Kornilov, 1980; Davison, 1983; Thomsen, 1984], however interpretation of previous studies remains complicated by a lack of standardized classification for both TBI and psychosis, non-defined severity of injury, and recall bias. An approximate threefold increase in the prevalence of psychosis in the post-TBI population as compared to those without injury is consistently reported [Batty et al., 2013]. Retrospective studies of patients with schizophrenia also find increased rates of premorbid TBI as compared to the general population [Wilcox and Nasrallah, 1987; Burg et al., 1996; Abdel Malik et al., 2003]. Delusional disorders have been generally reported to develop within a year post-TBI, with schizophrenia-like persecutory delusions and auditory hallucinations on average having a more delayed onset of three to four years, often complicating determination of the contribution of TBI to disease onset [Fujii et al., 2014].

TBI is associated with a broad range of psychiatric disturbances, even when neurotrauma is not judged to be the primary etiology. Fann et al. found that 49% of individuals with moderate to severe TBI developed a form of psychiatric illness within one year, and individuals with mild TBI were more likely to have a persistent psychiatric illness than those with psychiatric illness without TBI [Fann et al., 2004]. Major depression most commonly follows TBI [Hibbard et al., 1998; Kreutzer et al., 2001; Seel et al., 2003], but depending on the nature and context of the injury, posttraumatic stress disorder (PTSD) is also prevalent [Shalev et al., 1998; Hiott and Labbate, 2002; O’donnell et al., 2004].

Manic episodes also succeed TBI [Shukla et al., 1987; Bracken, 1987], and a recent study of 113,906 Danish individuals found that head injury was significantly related to a higher risk of bipolar disorder (incidence rate ratio (IRR) = 1.28), as was depression (IRR = 1.59) and schizophrenia (IRR = 1.65) [Orlovska et al., 2014]. Particular populations, namely those with a genetic vulnerability to psychosis, or prior psychological style or neurological pathology, are at an increased risk for psychosis following TBI [Fujii and Ahmed, 2001; Sachdev et al., 2001], and individuals with a pre-injury history of depression or PTSD are more likely to develop a major mood disorder following TBI [Bombardier et al., 2010].

The Genomic Psychiatry Cohort (GPC) [Pato et al., 2013] is a population-based sample comprised of individuals with schizophrenia and bipolar spectrum disorders, family members, and controls ascertained for genetic studies. It allows a unique opportunity to understand how TBI influences the features of a psychiatric condition, as individuals were excluded from ascertainment if a TBI was considered to be the cause of their psychiatric diagnosis. The study did not exclude individuals ever having a TBI however, because as noted above, TBI is a common exposure.

Though the evidence for an association between TBI, schizophrenia, and mood disorders has been frequently reported, further characterization of individuals who have developed psychiatric disorders prior to experiencing TBI is of great interest. We propose the possibility that a shared genetic vulnerability and a propensity for TBI may represent a gene-environment interaction, including increased propensity for posttraumatic symptoms. These probes, through genetic studies, could reveal patterns with important implications for prevention, diagnosis, and prognosis of mental illness.

Thus, the purpose of this study was not to assess the role of TBI in development of mental illness, but rather, its potential role in predicting illness features within a population with established psychiatric diagnoses. As groups of individuals with different diagnoses are included in the study and these conditions are themselves heterogeneous, and current thinking in the field favors the view of substantial biological overlap between the different conditions [Cuthbert and Insel, 2013], we used the statistical approach of k-means cluster analysis to sort the cases based on statistical inferences about the latent structure of the different subgroups of individuals recruited into the project with respect to clinical features and TBI.

MATERIALS AND METHODS

Sample

The sample used for this analysis consisted of 11,489 cases in the GPC, a National Institute of Mental Health-managed data repository for the investigation of schizophrenia and bipolar disorder. Recruitment sites included the University of Southern California, Cedars Sinai Medical Center, Emory University, Georgia Regents University, New York University, State University of New York Upstate, State University of New York Stony Brook, Wright State University, Texas Tech University, University of California at Los Angeles, and University of North Carolina. Institutional review board approval was received for all sites, and informed consent was obtained from all participants.

Participants in this study carried diagnoses on the schizophrenia and bipolar spectrums, and were recruited as inpatients in acute care facilities, chronic care facilities, and outpatient settings. All individuals completed a screening questionnaire eliciting demographic data, personal and family psychiatric history, and medical history. Psychiatric diagnoses were confirmed through administration of the Diagnostic Interview for Psychosis and Affective Disorder (DIPAD), which assesses phenomenological symptoms and utilizes the Operational Criteria Checklist algorithm [Farmer et al., 1992] to arrive at a DSM [American Psychiatric Association. Task Force on DSM-IV, 1995] diagnosis. Trained clinicians confirmed diagnoses for schizophrenia, schizoaffective disorder, or bipolar disorder with psychosis based on DSM-IV criteria. Categories of diagnosis and modifiers in this study included bipolar I disorder (+/− psychosis), manic episode with bipolar I (+/− psychosis), bipolar II disorder (+/− psychosis), schizoaffective disorder (bipolar type and depressive type), and schizophrenia (hebephrenic, paranoid, and undifferentiated).

At recruitment, a history of head and brain injury was acquired through self-report, and individuals were excluded from participation in this study if head or brain injury appeared the most prominent cause of their condition. Despite excluding such individuals, a portion of the subjects did report a history of head injury on interview. To qualify as TBI, an individual must have experienced alteration of consciousness, loss of memory, neurological deficit, and/or an intracranial lesion. Number of head injuries and duration of unconsciousness were also assessed via self-report in categorical fashion. For duration of unconsciousness, a value of 0 was assigned to no unconsciousness, 1: less than 60 min, 2: one to 24 hr, 3: one day to one week, 4: more than one week.

Cluster Analysis

In k-means methodology [MacQueen, 1967], each item in an algorithm is assigned to a cluster having the nearest centroid (mean). This nonhierarchical method initially takes the number of components of the population equal to the final required number of clusters. The final required number of clusters is chosen such that the points are mutually farthest apart. Items are partitioned into initial k clusters and sequentially assigned to a cluster with the nearest centroid by Euclidean distance. The centroid is then recalculated for both the cluster losing and gaining the item, and this process is completed for each item on the list until all items have been grouped.

The above described k-means clustering algorithm was used to probe differences between schizophrenia and bipolar disorder associated with several factors, particularly with TBI. The analysis was performed with SAS version 9.2 (SAS Institute, Cary, NC) proc fastclus. The number of clusters considered, “k,” ranged from 2 to 10, and the optimal k was determined to be 7 using the Likelihood Ratio χ² test statistics.

The data contain 11,489 observations including diagnosis categories (schizophrenia (SZ) and bipolar I disorder (BPD)), modifier categories (bipolar I disorder (BPD-I); bipolar I disorder, manic episode (BPD-M); bipolar I disorder with psychosis (BPD-P); bipolar I disorder with psychosis, manic episode (BPD-PM); bipolar II disorder (BPD-II); bipolar II disorder with psychosis (BPD-II-P); schizoaffective disorder bipolar type (SAD-BP); schizoaffective disorder depressive type (SAD-D); schizophrenia, hebephrenic (SZ-H); schizophrenia, paranoid (SZ-PA); schizophrenia, undifferentiated (SZ-U)) and variables (Table I), of which 760 observations with missing values or possible miss-labels were excluded.

TABLE I.

Clustering Variables

Variable	Mean	Std dev
Screen_age	43.516	12.450
Onset_age	21.386	8.770
Sex (M=0, F=1)	0.385	0.487
Halluc	1.563	0.768
Delusion	0.863	0.344
Head_injured	0.249	0.432
Head_injury_num	0.464	1.026
Unconsc_length	0.360	0.787
Brain_injured	0.023	0.150

RESULTS

Despite screening out individuals in which TBI was deemed causal to development of mental illness, 24.88% of the sample had head injury, 2.29% had brain injury, and 1.11% had both head and brain injury. Among males (n = 6598), 27.34% had head injury, 2.44% had brain injury, and 1.29% had both head and brain injury. 20.94% of females (n = 4131) reported head injury, 2.08% had brain injury, and 0.82% had both head and brain injury. The association between sex and occurrence of TBI did not reach significance (P = 0.276).

The optimal number of clusters of individuals in the sample was seven. As shown in Figure 1, the likelihood ratio chi-square value reaches its maximum at seven clusters. Each row of Table II gives the percentages of the occurrences of a given category of diagnoses in the seven different clusters. Table III provides a summary of each cluster.

FIG. 1.

Likelihood ratio chi-square vs. number of clusters.

TABLE II.

The Proportions (%) of a Given Modifier Category in Different Clusters are Shown in Each Row

	Cluster
Category	1	2	3	4	5	6	7
BPD-I	1.57	1.75	2.83	0.47	71.43	20.57	1.41
BPD-P	1.18	18.54	19.29	1.68	33.29	11.70	14.31
BPD-II	0.00	1.30	0.00	1.30	76.62	19.48	1.30
BPD-II-P	0.00	6.67	26.67	0.00	46.67	13.33	6.67
BPD-M	1.61	1.61	0.00	1.61	90.32	4.84	0.00
BPD-MP	0.94	18.87	20.75	0.00	46.23	5.06	7.55
SAD-BP	1.37	24.30	35.01	0.85	6.92	2.16	29.39
SAD-D	1.09	32.25	35.19	1.09	8.37	2.17	19.84
SZ-H	1.10	33.86	43.48	1.10	5.18	1.22	14.07
SZ-P	0.70	36.27	28.87	1.27	6.90	2.32	23.66
SZ-U	1.31	33.78	33.71	1.07	7.00	1.98	21.16

TABLE III.

Mean (Standard Deviation) of Variables in Each Cluster

Cluster	N	Screen age	Onset age	Sex (F = 1)	Hallucination	Delusion	Head injury	Head injury number	Unconsc length	Brain injury
1	126	43.032 (11.461)	19.778 (8.636)	0.413 (0.494)	1.349 (0.842)	0.881 (0.325)	0.000 (0.000)	0.000 (0.000)	0.008 (0.089)	1.000 (0.000)
2	2992	53.334 (7.799)	27.252 (8.908)	0.474 (0.499)	1.830 (0.472)	0.988 (0.108)	0.005 (0.071)	0.005 (0.077)	0.012 (0.122)	0.000 (0.000)
3	3248	34.843 (9.496)	17.105 (5.258)	0.280 (0.449)	1.887 (0.336)	0.998 (0.050)	0.003 (0.053)	0.003 (0.053)	0.010 (0.126)	0.000 (0.000)
4	120	43.542 (10.890)	22.242 (9.873)	0.283 (0.453)	1.583 (0.762)	0.833 (0.374)	0.992 (0.091)	2.083 (1.441)	1.633 (1.061)	1.000 (0.000)
5	1713	41.409 (12.575)	20.867 (8.387)	0.527 (0.499)	0.407 (0.753)	0.371 (0.483)	0.000 (0.000)	0.000 (0.000)	0.010 (0.115)	0.000 (0.000)
6	504	43.516 (12.577)	21.442 (9.974)	0.421 (0.494)	0.480 (0.792)	0.369 (0.483)	0.998 (0.045)	1.857 (1.235)	1.433 (0.924)	0.000 (0.000)
7	2026	44.729 (11.418)	20.060 (8.308)	0.299 (0.458)	1.906 (0.292)	1.000 (0.000)	0.999 (0.038)	1.857 (1.278)	1.412 (0.983)	0.000 (0.000)

Patients in Cluster 2 are oldest on average, while patients in Cluster 3 are youngest. Clusters 3, 4 and 7 have higher frequencies of males, with Cluster 5 having the largest proportion of females. For the whole sample, the ratio between females and males is about 2:3 (4131 vs. 6598). Hallucinations occurred more frequently than average in Clusters 2, 3 and 7 but less frequently in Clusters 5 and 6. Delusions were less frequent in Clusters 5 and 6 than the others.

Table II shows that BPD, BPD-M, and BPD-II are highly concentrated in Cluster 5 (71.43%, 90.32% and 76.62% respectively), and have no delusions or brain injuries. BPD-P, BPD-MP, and BPD-II-P are also highly populated in Cluster 5, though less so than BPD, BPD-M, and BPD-II; these three categories do not have hallucinations or brain injuries, but all have delusions. SAD-BP, SAD-D, SZ-H, SZ-PA, and SZ-U are evenly split into Clusters 2 and 3. Most of the patients in clusters 2 and 3 had delusions and hallucinations present most of the time, however none had brain injury.

Table III shows that Clusters 1 and 4 include all cases with brain injury. No individuals in Cluster 1 had head injury (n = 126), compared to 99.2% of patients in Cluster 4 (n = 120). The highest concentrations of head injury are found in Clusters 4, 6, and 7. Cluster 6 (n = 504) and Cluster 7 (n = 2026) have 99.8% and 99.9% of the head injury cases, respectively, but no patients in these clusters had brain injury. Head injury and length of unconsciousness are highly correlated.

BPD-P cases represent approximately 15% of the entre sample, with a high frequency in Cluster 4 (22.5%), and account for 27 of 120 brain injuries in Cluster 4 (Table IV). The frequency of BPD-P in Cluster 1 is approximately 15%, similar to the frequency in the entire sample. According to the χ² test, the correlation between the occurrence of BPD-P and brain injury (246 brain injuries out of 1,607 BPD-P cases) is not significant for the whole sample (χ² value 2.74, 1 DF, P = 0.10). However, in Cluster 4, the occurrence of BPD-P and brain injury cases is significantly correlated (χ² value 5.39, 1 DF, P = 0.026). Frequencies of other modifier categories were not correlated with brain injury.

TABLE IV.

Distribution of Modifier Categories in the Whole Sample and Each Cluster

	Modifier category
Cluster	BPD-I	BPD-M	BPD-MP	BPD-P	SAD-BP	SAD-D	SZ-H	SZ-P	SZ-U	BPD-II	BPD-II-P	Total
Whole Sample	637	62	106	1607	1531	645	1642	1420	2987	77	15	10729
1	10	1	1	19	21	7	18	10	39	0	0	126
2	11	1	20	298	372	208	556	515	1009	1	1	2992
3	18	0	22	310	536	227	714	410	1007	0	4	3248
4	3	1	0	27	13	7	18	18	32	1	0	120
5	455	56	49	535	106	54	85	98	209	59	7	1713
6	131	3	6	188	33	14	20	33	59	15	2	504
7	9	0	8	230	450	128	231	336	632	1	1	2026

In Cluster 4, 72% of patients are males (61.5% overall), frequency f hallucinations is 1.583 (0: not present, 1: present less than one month, 2; present most of the time in one month period), 63% of patients have delusions, and 99.2% of the subjects had head injury (2.083 occurrences of head injury on average) with the longest average period of unconsciousness among the clusters.

DISCUSSION

This study demonstrated an association between TBI and psychosis in a subset of bipolar cases. Though temporal data was not available to determine if TBI preceded or followed onset of psychosis, and at what intervals, our findings nonetheless point towards a potential gene-environment interaction, and/or an increased propensity for TBI among individuals with psychosis. This utilization of k-means cluster analysis allowed for a hypothesis-independent examination of potential associations between the schizophrenia and bipolar spectrums and salient variables.

Cluster 4, which demonstrated a significant association between bipolar I disorder with psychosis and TBI, contained the second-highest proportion of male to female cases of the seven clusters, with a 72% male predominance. Though reported rates of bipolar disorder in men and women are roughly equal [Diflorio et al., 2010], males are at a two-to-threefold increased risk for TBI, particularly during adolescence and young adulthood [Bruns and Hauser, 2003]. Thus, future studies should pay particular attention to this at-risk population. Additionally, individuals in cluster 4 had the highest average length of unconsciousness among the clusters. Length of unconsciousness has previously been described as a primary determinant of long-term outcome, more-so than the presence of focal lesions [Ross et al., 1994].

Given the association between psychosis and TBI found among individuals with bipolar disorder, a similar association may have been expected among individuals on the schizophrenia spectrum, but was not found. Importantly, this study excluded TBI as causal to any syndrome since subjects with TBI prior schizophrenia or bipolar diagnoses were excluded. An association was thus only seen between psychosis and TBI among a subset of individuals with bipolar disorder, as psychosis is a modifier of this disease, not a core feature required for diagnosis, and thus can develop later in the illness course.

Including only childhood TBI prior to onset of schizophrenia, Abdel Malik et al. found schizophrenia subjects more likely than their unaffected siblings to have a history of childhood head injury, with severity correlated with a younger onset age [Abdel Malik et al., 2003]. Similarly, a recent study by Orlovska et al. found head injury to be associated with a higher risk of schizophrenia, with head injury between ages 11 and 15 as the strongest predictor [Orlovska et al., 2014]. An association between genetic loading for schizophrenia and increased risk of subsequent TBI has also been reported [Malaspina et al., 2001], underscoring that not only may head trauma increase the risk for psychiatric illness, psychiatric illness may increase the risk for head trauma. Additional features of schizophrenia associated with subsequent TBI following development of the illness may not have been assessed by this current study, and warrant further examination.

Current literature highlights the increasing contribution of environmental factors to development of mental illness, as such elements have the ability to interact with and influence gene expression [Jaffee and Price, 2007]. For example, studies suggest that adverse exposures including childhood maltreatment interact with a serotonin transporter variant implicated in depression [Karg et al., 2011], and that numerous similar gene-environment interactions are mediated through epigenetic mechanisms including DNA methylation [Uher, 2014]. Given the recent expanded identification of associated genetic loci in schizophrenia [Consortium, 2014], the call to examine potential genetic interactions with environmental factors becomes critically imperative. Future studies should focus on certain genotypes that may interact with known and quantifiable environmental stressors such as TBI.

The inflammatory response to TBI is also important to consider in association with development of psychosis. Following TBI, an initial inflammatory response is mediated in reaction to direct damage to microglia and astrocyte homeostasis, followed by subsequent influx of extracerebral inflammatory cells due to endothelial damage [Cederberg and Siesjö, 2010]. Abnormal levels of pro-inflammatory cytokines have been well-documented in schizophrenia [Miller et al., 2011], and inflammation has been proposed as a potential marker and therapeutic target for first-episode psychosis [Zajkowska and Mondelli, 2014]. The association reported here between bipolar psychosis and TBI lends further weight to growing evidence for the relationship of inflammation with psychosis.

Strengths of this study include the use of k-means cluster analysis, which has proven very useful in disentangling the largely heterogeneous disease processes of the schizophrenia and bipolar disorder spectrums; inherent to this method is the grouping of individuals with minimal variation within, but maximum variation between, the set clusters. Additionally, with the ability to draw from the GPC for cases included in this study, a large sample size of 11,489 individuals was utilized. Limitations of this study include a lack of temporal data; though TBI was excluded as a primary cause of development of mental illness in these individuals, information is not available on the relationship between occurrence of TBI and length of time preceding development of psychosis. Additionally, though frequency of TBI was quantified, information on severity of TBI was not included in this study, and thus may impact the presence and extent of further sequelae. Self-report was used in assessment of history of head injury and TBI; official records to confirm these reports were not available, and thus this information is subject to recall bias and potential inaccuracy.

The presence of TBI in this subset of individuals demonstrates that traumatic stressors have the ability to impact gene expression in a vulnerable population, and/or a heightened occurrence of TBI in individuals with underlying psychosis. This study lends further weight to a growing literature stressing the importance of noting a history of brain injury, and the potential dual association between brain injury and mental illness. To further the work presented here, future studies should be aimed at more closely examining the interplay between genetic variation in individuals with bipolar disorder and susceptibility to psychosis following TBI.