Neurological disorders and violence: a systematic review and meta-analysis with a focus on epilepsy and traumatic brain injury
- First Online:
- Cite this article as:
- Fazel, S., Philipson, J., Gardiner, L. et al. J Neurol (2009) 256: 1591. doi:10.1007/s00415-009-5134-2
- 467 Views
The objectives of this study were to systematically review and meta-analyze the research literature on the association of common neurological disorders and violence. Keywords relating to neurological disorders and violence were searched between 1966 and August 2008. Case–control and cohort studies were selected. Odds ratios of violence risk in particular disorders compared with controls were combined using fixed-effects meta-analysis with the data presented in forest plots. Sensitivity analyses were conducted to identify possible differences in risk estimates across surveys. Information on risk factors for violence was extracted if replicated in more than one study. Nine studies were identified that compared the risk of violence in epilepsy or traumatic brain injury compared with unaffected controls. For the epilepsy studies, the overall pooled odds ratio for violent outcomes was 0.67 [95% confidence interval (CI) 0.46–0.96]. For traumatic brain injury, the odds ratio was 1.66 (95% CI 1.12–2.31). An additional 11 case–control studies investigated factors associated with violence in epilepsy and traumatic brain injury. It was not possible to meta-analyze these data. Comorbid psychopathology was associated with violence. Data on other neurological conditions was limited and unreplicated. In conclusion, although the evidence was limited and methodological quality varied, epilepsy and traumatic brain injury appeared to differ in their risk of violence compared with control populations. Longitudinal studies are required to replicate this review’s provisional findings that epilepsy is inversely associated with violence and that brain injury modestly increases the risk, and further research is needed to provide information on a broader range of risk factors.
KeywordsForensicEpilepsyEpidemiologyBrain injuryBehavioral neurology
There is uncertainty whether neurological disorders are associated with violence. Most experienced neurologists will have come across cases of postictal violence or violent automatisms that are clearly linked to the effects of the disease [13, 47]. Consequently, there is a long-standing belief that epilepsy predisposes individuals to violent behavior [37, 46]. The contribution of head injury to violence is clearly exemplified in the case of Phineas Gage , and clinical experience in neurosurgery and neurology suggests that violent or at least high-risk behavior both precedes and follows brain injury . A review of the topic written about 15 years ago, however, was not able to confirm a strong association between specific neurological disorders and violence, although it stated that epilepsy was overrepresented in individuals prone to recurrent violence .
A consensus statement from the Aspen Neurobehavioral Conference  concluded that epidemiological studies of risk factors should be a research priority. In particular, the biomedical perspective on violent behavior has been relatively neglected compared with social science contributions. We have therefore undertaken a systematic review of studies examining the association of neurological disorders with violence. We examined ten common neurological conditions based on population-based estimates of frequency . We aimed to investigate two questions. First, is there an association of specific neurological illnesses with violence compared with the general population? Second, what risk factors are associated with violence in these disorders?
Studies of the association of neurological illnesses and violence reported between January 1966 and August 2008 were sought by searches of electronic bibliographic databases (MEDLINE, EMBASE, PsycINFO, CINAHL) using combinations of keywords relating to the neurological illnesses (headache/migraine, stroke, CNS infection, Parkinson’s disease, essential tremor, epilepsy, head injury, multiple sclerosis, chronic fatigue syndrome, brain tumours, Huntington’s disease) and to violence (e.g., viole*, aggress*, crim*, prison*, sexual). This was supplemented with scanning of article reference lists. Studies were included if they: (1) presented primary data on the association between violence or aggression or violent crime and the neurological conditions being studied, (2) used clear criteria for diagnoses and violent outcomes, and (3) presented quantitative findings including numerators and denominators. Studies were excluded if selected samples of violent individuals were used , only the results of statistical tests were presented (as distinct from rates of violence) , they used comparison groups of psychiatric patients  or from other time periods , or they did not meet the criteria above. Studies with anger and impulsivity as outcomes (or where verbal aggression was the predominant outcome ) were excluded as the focus was interpersonal violence and physical aggression in line with other reviews in forensic medicine and psychiatry [10, 50]. We excluded dementia as there have been previous systematic reviews of factors associated with violence .
Two authors (R.M. and S.F.) independently rated the abstracts. Those papers selected for further review were then independently rated by another two authors (J.P. and L.G.). Any discrepancies were resolved by further review and discussion with the senior author (S.F.).
Studies were rated according to a method used by Sackett that assigns the order of methodological quality as: prospective cohort, retrospective cohort, nested case–control, case–control (with controls similar to the patient group) or case–control (with unclear or different controls) . Predetermined study characteristics were extracted, including information on study design, recruitment, location, numbers of cases and controls, diagnostic criteria, outcome measures, mean age, and any adjustment.
Odds ratios (ORs) of the risk of violence in particular disorders compared with controls were combined using fixed-effects meta-analysis with the data presented in forest plots. Ninety-five percent confidence intervals (95% CI) were reported. Heterogeneity among studies was estimated using Cochran’s Q (reported with a χ2-value and p value) and the I2 statistic, the latter describing the percentage of variation across studies that is due to heterogeneity rather than chance [28, 29]. I2, unlike Q, does not inherently depend upon the number of studies considered with values of 25%, 50%, and 75% taken to indicate low, moderate, and high levels of heterogeneity, respectively. As the heterogeneity in the studies was low, fixed-effects models were used. We also tested for publication bias by funnel-plot asymmetry using the rank correlation method as proposed by Begg  and the weighted regression approach of Egger . All analyses were undertaken with STATA statistical software package (Statacorp 2008), version 10.0.
In an attempt to identify likely risk factors, only those factors that were significantly associated with violence (using a p value of <0.05 in unadjusted models) in two separate studies for each condition were reported.
As no individual participant data was obtained, ethics approval was not required. The study was conducted according to the ethical standards of the Declaration of Helsinki.
Nine investigations compared the risk of violence in epilepsy or brain injury compared with nonaffected controls, and the other 11 studies examined factors associated with violence in these conditions.
Risk of violence compared with nonaffected controls
The most common outcome measure was criminal convictions on routinely collected crime registers, in six studies [7, 26, 40, 43, 49, 51]. There were some discrepancies in how violent crime was defined, and two did not include sex offences [26, 40]. The three remaining studies used self-report questionnaires, including the Overt aggression scale , and two instruments developed for the study in question [22, 36].
For the epilepsy studies, the overall fixed-estimate OR was 0.67 (95% CI 0.46–0.96), with negligible heterogeneity between the studies (Q = 0.60, p = 0.80, I2 = 0). For traumatic brain injury, the overall fixed-estimate OR was 1.66 (1.12–2.31), with little heterogeneity (Q = 3.8, p = 0.59, I2=0%). In a separate analysis limited to male-only data in brain injury, the risk estimates did not materially change: the OR was 1.46 (95% CI 1.00–2.13) with no heterogeneity (Q = 1.34, p = 0.72, I2 = 0%).
In a subgroup analysis comparing prison samples with other samples, there was little difference in risk estimates. If the prison studies were excluded [7, 26, 43, 51], the overall OR for epilepsy was 0.88 (0.35–2.17), and in brain injury the OR was 1.64 (1.07–2.52). A subgroup analysis examining the effect of the outcome measure in the head injury studies found no significant difference in risk estimates when criminal convictions compared with self-report was used [OR = 1.42 (0.98–2.05) versus OR = 3.03 (1.36–6.78), respectively].
Overall, there was evidence of publication bias. Asymmetry was found in the funnel plot using weighted regression (p = 0.02) with a trend towards publication bias using Begg’s test (p = 0.06). This was specifically related to the brain injury studies (weighted regression: p = 0.04).
There was little evidence for other neurological disorders—we identified one study that found no association between risk of violent crime compared with nonviolent crime in female prisoners with migraine or other headaches, central nervous system (CNS) tumors, and CNS infections . Another study compared men and women with Huntington’s disease with controls from the general population, and reported a nonsignificant trend for an increased risk for violent crime (OR = 3.6; 95% CI 0.7–17.4) .
Factors associated with violence
We identified 13 case–control studies that investigated factors associated with violence within diagnostic groups, 11 of which were in epilepsy and traumatic brain injury. There were six studies in epilepsy [6, 21, 27, 38, 41, 52], five case–control investigations in head injury [4, 24, 34, 45, 54], one in stroke , and one in Parkinson’s disease  (Table 2). Due to the heterogeneity of outcome measures and populations sampled, a meta-analysis of these data was not feasible. Violence was defined by self-report in eight studies (supplemented by carer information in some) [4, 21, 24, 34, 38, 39, 45, 54] and from medical records in three investigations [6, 27, 41]. In order to assess psychiatric comorbidity, four studies used standardized clinical interviews [6, 38, 39, 45] and two examined medical records [27, 41].
We examined those factors that were significantly associated with violent outcomes in at least two studies per condition, which limited the analysis to epilepsy and traumatic brain injury. In epilepsy, these were a history of childhood behavioral difficulties [27, 41] and comorbid psychopathology. Comorbid diagnosis of schizophrenia and learning disability/mental retardation  and a trend towards a history of psychosis  were associated with violence. Comorbid psychosis has also been reported in postictal aggression .
In traumatic brain injury, comorbid depressive symptoms were associated with violence [4, 45]. Frontal lobe lesions on imaging  and neuropsychological changes associated with impairment to the prefrontal cortex  were reported in two studies. A history of lower educational attainment was also demonstrated in two investigations [24, 45].
In this systematic review, we identified nine studies that reported risk of violence in epilepsy and traumatic brain injury compared with unaffected controls, and another 11 studies that examined factors associated with violence in these conditions. There were three main findings. First, there was a paucity of studies addressing the topic, and what little there was had methodological shortcomings. Of the included investigations, only two used cohort designs, both of which were retrospective. A second finding was that epilepsy and brain injury appeared to differ in their risk of violence compared with the general population—epilepsy was inversely associated with violence, while brain injury increased the risk. Finally, comorbid psychopathology appeared to be associated with violence in both of these disorders.
Our systematic review and meta-analysis suggests that the risk of violence in those with traumatic brain injury—between one to threefold that of the general population—is of a similar magnitude to the risk of violence in those with major mental disorders after adjustments are made for socioeconomic status and comorbid substance abuse . However, this finding is provisional as the included studies were drawn from prisoners, veterans or patient samples—no population-based surveys were identified. Although there was no statistical heterogeneity in the association with violence, there was clinical heterogeneity in the samples included and therefore the causal pathways to violence will differ. Nevertheless, our review provides a crude estimate of the risk that appears to be similar irrespective of the sample. One of the clinical implications suggested by this finding is that structured violence risk assessment should be considered in some patients with head injury. Such assessments are now routinely done in severe mental illness, with increasing emphasis to systematize and further incorporate dynamic, or changeable, factors .
Although based on three studies, no clear association of epilepsy with violence was found. This is contrary to authors of some modern textbooks who have stated that epilepsy is associated with an increased risk for violence [35, 44, 46], although this view has been undermined by lack of evidence for an increased prevalence of epilepsy in prisoners . It is possible that the impact of high-profile cases of epileptic violence is unrepresentative, or that the long-term sedative  or mood-stabilizing effects of medication prescribed to most patients with epilepsy counterbalances this effect. The acquittal of a very few individuals with epilepsy on the basis of criminal insanity may impact on homicide rates . In addition, these articles in textbooks have cited a population-based study from the 1960s in Iceland, which reported a three times increased rate of criminality compared with the general population, but this was excluded from our review as the controls were drawn from a different study and time period, and included the cases . Two of the three epilepsy studies included in the review were in prisoner samples, and the remaining study was based on a diagnosis of epilepsy before age 15 years. In total, these investigations included 369 individuals with epilepsy. Thus, the evidence for epilepsy is quite limited and needs replication in a true population-based study of crime and violence. Any such study would be important, as epilepsy remains heavily stigmatized [12, 31]. It is also not known reliably how the risk profiles compare with different subtypes of epilepsy. Finally, it is important to clarify that our findings do not mean that epilepsy cannot be the basis of violent behavior in some individuals, as attested in many case reports .
Our systematic review of factors associated with violence provides a basic outline of risk assessment for patients with head injury or others who are thought to be at increased risk. In patients with traumatic brain injury, those with comorbid depression or prefrontal/frontal lobe lesions could benefit from risk assessment. Other patients where risk assessment could be considered include those with other comorbid psychopathology, and those with premorbid difficulties. In those with frontal lobe impairment, clinical assessment together with neuropsychological testing of executive function could be informative . However, the need for risk assessment and management needs to be balanced by the risk of further stigmatization for some patients, and further research could examine the positive predictive value of risk assessment for different combinations of risk factors. The relevance of other possible risk factors, such as comorbid drug and alcohol abuse, previous violence, and young age (all of which are risk factors for violence in severe mental illness ), would be an area of possible future research. The putative risk factors identified in this study find some congruence with factors associated with aggression in dementia. A recent review identified four studies that demonstrated associations with psychosis, two investigations with depression, and a number that reported higher rates of assault in frontotemporal dementia .
The relevant research base is limited, and in particular, studies of individuals with neurological disorders in the general population (as distinct from selected samples, such as prisoners) are needed. Most of the studies included used case–control and cross-sectional designs, and no prospective investigations were identified. There was evidence of publication bias in the head injury studies. Therefore, our findings warrant caution in their interpretation and serve as hypothesis-generating. Meta-analyses of observational studies can be problematic as they are prone to confounding and selection biases . Although using individual participant data would be an alternative approach to deal with this, in view of the lack of available evidence, new longitudinal research would be more promising. The studies reviewed here included little or no adjustment for potential confounders, and it is therefore possible that the risk estimates overstate the association of these disorders with violence. Furthermore, a variety of outcome measures were used, including violent crime and various self-report measures of aggression. Investigators in the field would benefit from agreed outcome measures, a challenge shared by research into violent outcomes in psychiatry , as it would enable more appropriate comparisons. The review of studies examining factors associated with violence was limited in that all but one of the included investigations were retrospective in design, and used different methods to assess violence (including self-report and medical records) and potential risk factors from neuroimaging to clinical interview. Four of the reports used standardized clinical interviews to assess comorbidity, which should be incorporated in future research. A final issue is whether this review should have concentrated on head injury alone, where the evidence is strongest. We think that the contrast with epilepsy in relation to risk of violence compared with controls strengthens the provisional findings. Furthermore, the fact that risk factors were shared across these two conditions suggests research into common causal mechanisms may be worth exploring.
In summary, in this systematic review of neurological conditions and violence, we found some evidence to suggest that traumatic brain injury is modestly associated with violent outcomes and that epilepsy is not. Furthermore, violence risk assessment of patients could consider the assessment and treatment of comorbid psychopathology, the presence of which appears to increase the risk. Improved understanding of the relationship between neurological disorders and violence should inform neuropsychiatric, public health, and public policy interventions to reduce violence.
There was no funding for this study.
Conflict of interest statement
The authors report no conflicts of interest.