Introduction

Background

The association between severe mental illnesses and violent behavior is complex and represents a research topic of longstanding interest for its clinical, social, and public health repercussions [1, 2].

People living with severe mental illnesses are frequently victims of violence [3, 4]. In fact, violent victimization of people with severe psychiatric disorders is a frequent phenomenon, which carries a high risk of serious physical injuries; it also has an important negative impact on the quality of life and on the trajectory of the illness of victims, and is linked to stigma [5,6,7,8].

Some individual psychiatric disorders, however, are also associated with an increased relative risk of violent behavior [2, 9,10,11]. Individuals diagnosed with schizophrenia spectrum disorders (SSD) [12, 13] show a 6–10% prevalence of violent crime perpetration leading to arrest or conviction, which rises to more than 10% in people with substance use disorders [14, 15]. In fact, substance use represents the strongest single risk factor for violent behavior among people diagnosed with mental disorders [16, 17].

As violent crimes are events that could be preventable in some cases, identifying risk factors and clinical correlates in individuals living with mental disorders is of great relevance to recognize high-risk situation and implement preventive interventions and strategies. Beside substance use, male gender, low socio-economic status, homelessness, criminal history, parental violence, parental substance use, victimization, inpatient admission, previous self-harm and previous suicide attempts, all represent recognized risk factors for violence in subjects diagnosed with mental disorders [18,19,20,21]. Dynamic and modifiable factors include hostility, impulsivity and, particularly in psychotic disorders, lack of insight and treatment adherence and severe positive symptoms such as persecutory beliefs [22].

Alterations of cognitive performance are frequently observed in many psychiatric disorders and, in particular, those with higher violence risk such as SSD. Cognitive deficits represent one of the core features of SSD, with significant impairments in several neurocognitive domains as well as in social cognition performance [23,24,25,26], and cognitive deficits have been hypothesized as a risk factor for violent behavior in mental disorders. A meta-analysis conducted in 2014 and including 29 studies for a total of 4764 participants diagnosed with SSD [27] found that impaired global cognitive performance represents a significant predictor of violent aggression; however, a high degree of heterogeneity was observed, with single studies showing conflicting results. Moreover, important discrepancies were observed regarding individual cognitive domains: most analyses on single domains showed excessive heterogeneity or a non-significant effect of impaired cognitive performance, while, contrary to the expected, better motor functioning was associated with increased risk of violence.

Conflicting results can also be found in more recent literature, with some studies reporting that cognitive impairment does indeed represent a predictor of violent behavior in SSD [28, 29] and others reporting a better cognitive performance in violent patients [30].

These discrepancies suggest that a better characterization of the relationship between cognitive functioning and violence represents an issue that requires further investigation. Moreover, greater insight on this topic could provide further understanding of violence risk conditions, with potential research and real-world clinical implications. It could also provide opportunities to optimize treatment of people living with SSD and enhance violent acts prevention.

Conviction represents the most official and robust measure of violent outcomes as it overcomes recall-bias, inter-rater variability, and the risk of missing outcomes measurements [2]: in this context, offender SSD patients with better character in forensic contexts could provide valuable clinical insight.

Aims

The primary aim of the present study was to assess the cognitive profile, as well as clinical, impulsivity, aggression, and psychopathy correlates of offenders diagnosed with SSD (offender participants, OP), compared to non-offenders participants (non-OP) matched for age, gender, education, and diagnosis.

Secondary aim was to identify individual cognitive and clinical predictors belonging to the OP group, and to assess the clinical correlates of cognitive characteristics that emerged as significantly different in OP and non-OP. In fact, the primary hypothesis of this study is that OP might show a different cognitive profile, as well as higher levels of aggression risk and psychopathy.

Materials and methods

Sample

For the present study, 50 inmates convicted for violent crimes and enrolled in the Residences for the Execution of Security Measures in Castiglione delle Stiviere, Mantova, Italy were recruited from June 2018 to June 2020.

Inclusion criteria were: age between 18 and 65 years; main diagnosis of SSD according to the DSM- 5 criteria.

Exclusion criteria were: diagnosis of substance use disorder; main diagnosis of intellectual disability; presence of severe or neurodegenerative organic pathologies.

Non-OP were recruited in the same period and with the same inclusion and exclusion criteria among the patients of the Mental Health centers of the Spedali Civili Hospital in Brescia, Italy. They were matched on a 1:1 basis with OP for age, gender, education, and diagnosis.

Patients were informed about the study and were invited to participate through a written and signed consent form. The study was carried out in accordance with the Code of Ethics of the World Medical Association and the Declaration of Helsinki. The protocol was approved by the Ethical Committee of Brescia (Project Identification Code NP3060). All precautions were taken for the management of sensitive data, and participants were not given monetary compensation for their involvement in the study.

Measures

Socio-demographic information (age, gender, years of education, number of school failures, marriage status, work situation) was collected through a specific data collection form.

To provide a comprehensive assessment of the cognitive profile, participants were evaluated through the Brief Assessment of Cognition in Schizophrenia (BACS) [31], the Trail Making Test (TMT) [32] and the Stroop Color–Word Test (STROOP) [33]. Test performances were attributed to different cognitive domains according to previous meta-analytic literature [34, 35]: TMT-A: processing speed; TMT-B: working memory; TMT-B–A: executive functions; STROOP errors: attention; STROOP time: processing speed; BACS Word Recall: verbal memory; BACS Digit Sequencing: working memory; BACS Token Motor Task: processing speed; BACS Category Instances: fluency; BACS Symbol Coding: processing speed; BACS Tower of London: executive functions. Raw scores were corrected for age and education and equivalent scores were included in the analyses. Social cognition was assessed with the Facial Emotion Identification Test (FEIT) [36], and the percentage of correct answers was included in the analyses.

The clinical assessment included an evaluation of the severity of psychopathological condition with the Clinical Global Impression Severity Scale (CGI-S) [37] and the Excited Component of the Positive and Negative Syndrome Scale (PANSS-EC) [38, 39].

Impulsivity was assessed through the administration of the Barratt Impulsiveness Scale (BIS-11) [40] and the Iowa Gambling Task (IGT) [41]; aggressiveness was assessed with the Modified Overt Aggression Scale (MOAS) [42], the risk of violent relapse with the Historical Clinical and Risk Management scale 20 (HCR-20) [43] and the level of psychopathy with the Hare Psychopathy Checklist (PCL-R) [44].

All subscales of the HCR-20 were included in the analyses. The PCL-R items were included in the analyses divided in two factors as defined by factor analysis conducted by the original scale authors: a factor describing the selfish, callous, and remorseless use of others (Factor 1 or “Callous”) and a factor describing chronically unstable and antisocial lifestyle or social deviance (Factor 2 or “Unstable”) [45]. Given their association with aggressive behavior and violent outcomes, these items were considered as more violence related [2].

All the assessments were carried out by trained physicians who were not involved in the usual care process of participants.

Statistical analyses

OP and non-OP groups were compared regarding all socio-demographic, clinical, cognitive, aggression, impulsivity, and risk of violent relapse parameters using Mann–Whitney U tests and χ2-tests for continuous and categorical variables, respectively. Only non-parametric analyses were used as all continuous variables had a non-normal distribution (all Shapiro–Wilk tests had p < 0.001). These univariate exploratory analyses were conducted to select potential predictors for the regression analyses [46].

Variables that emerged as significantly different in the univariate between-groups comparisons were introduced as potential predictors in multivariate binary logistic regression models conducted to separately identify socio-demographic and clinical, cognitive, and violence-related individual predictors of violent offence, allowing to assess which variables could explain the largest proportion of between-group variance. This approach was adopted considering the exploratory nature of the study to avoid using a single regression model in which strongly violence-related parameters would mask other significant variables, leading to type II errors. All regression analyses were conducted using a backward approach. Multicollinearity between individual predictors was assessed and was considered significant if the variance inflation factor (VIF) exceeded 4.0 [47]. As the number of potential predictors introduced in each regression model was lower than one for every ten observed subjects, which is recommended for logistic regressions according to conservative estimates [48, 49], the number of the included predictors was considered appropriate.

Post-hoc analyses were conducted to explore potential correlations between potential predictors’ variables in the whole sample using Spearman’s rho correlation test to better understand the relationships between different explored factors.

All analyses were conducted using the statistical software SPSS version 15.0 (SPSS Inc., Chicago, IL, USA, 2005); p values < 0.05 (two tailed) were considered significant.

Results

Sample characteristics and between-groups comparisons

The total sample, composed by 100 participants (50 OP and 50 non-OP) diagnosed with SSD was characterized by a mean age of 37.81 (± 11.06) years, the presence of 14 (14%) female subjects, and a mean education of 12.16 (± 2.88) years. As OP and non-OP were matched at recruitment regarding age, gender, and education, as well as diagnosis, no difference regarding these parameters could be observed.

A higher number of school failures, (p = 0.002), a higher prevalence of lifetime substance use (p = 0.024), and a more overall severe clinical condition (p = 0.002) and more severe excitatory symptoms (p < 0.001) were observed in OP (see Table 1).

Table 1 Comparison for socio-demographic and clinical characteristics
Full size table

OP showed a significantly worse performance in working memory domain as measured by the TMT-B test (p = 0.037) and in processing speed domain as measured by the Stroop test time (p = 0.026) and BACS Token Motor Task (p < 0.001); they also showed a better performance regarding attention domain as measured by Stroop Test errors (p < 0.001; for more details, see Table 2).

Table 2 Comparison for cognitive characteristics
Full size table

OP also showed higher scores in all HCL-20 subscales and PCL-R factor scores (all p < 0.001; for more details, see Table 3).

Table 3 Comparison for aggression, impulsivity, and risk of violent relapse
Full size table

Predictors of violent offence

As regards the multivariate regression analyses, the number of school failures (p = 0.007) and the severity of excitatory symptoms (p < 0.001) emerged as socio-demographic and clinical individual predictors of belonging to the OP group (for more details, see Table 4).

Table 4 Socio-demographic predictors of violent offence
Full size table

Better attentive performance as measured by the Stroop Test errors (p = 0.003) and worse performance in speed of processing as measured by the Stroop Test time (p = 0.025) and BACS Token Motor test (p = 0.024) emerged as individual cognitive predictors (for more information, see Table 5).

Table 5 Cognitive predictors of violent offence
Full size table

Finally, the HCR-20 Risk Management subscale (p = 0.002) and the PCL-R “Callous” factor emerged as individual predictors in the last multivariate regression model (p = 0.031). This model explained the largest portion of between-group variance (Cox–Snell R2 = 0.692, Nagelkerke R2 = 0.923; for more details, see Table 6).

Table 6 Aggression, impulsivity, and risk of violent relapse predictors of violent offence
Full size table

Post-hoc analyses

Post-hoc correlation analyses highlighted several significant correlations between cognitive and violence-related variables in the whole sample: of particular interest, better attentive performance in the Stroop Test errors is correlated with higher scores in all HCR subscales, in both PCL-R factors and negatively, as expected, with the BIS-11 Attention subscale, and worse performance in several cognitive domains is correlated with higher score in PCL-R “Unstable” factor (more details in Table 7).

Table 7 Post-hoc correlation analyses
Full size table

Discussion

In the present study, 50 patients diagnosed with SSD and convicted for violent crimes and 50 non-offender participants matched for age, gender, education, and diagnosis were compared on a wide array of cognitive, clinical, impulsivity, aggression, and psychopathy characteristics. Predictors of belonging to the violent offenders participants subgroup were also assessed for each category of characteristics.

OP showed a larger prevalence of lifetime substance use: this finding was expected as substance use represents the most prominent predictor of violent behavior in people living with mental disorders [14, 16]. They also showed a higher number of school failures: this can, on one hand, be considered an indirect proxy of worse cognitive functioning, a factor that is better explained in the dedicated analyses; however, it can also be considered a marker of greater general adversity during childhood, including social and economic difficulties, a less supportive home environment, or the presence of undiagnosed neurodevelopmental disorders, such as attention deficit hyperactivity disorder (ADHD) or oppositional defiant disorder (ODD), which are correlated with both a higher number of school failures in childhood and to a higher risk of committing offences in in adulthood [50, 51]. Regarding clinical variables, offender participants showed higher CGI-S and PANSS-EC scores, corresponding to a more severe clinical condition; this finding is also in line with previous literature [2, 22].

As regards cognitive abilities, OP were characterized by an overall worse cognitive functioning, highlighted by a significantly worse performance in the working memory domain, as measured by the TMT-B, and a significantly worse performance in the processing speed domain, as measured by the BACS Token Motor Task and the Stroop Test time.

However, OP also showed a better attentive performance, as measured by the Stroop Test errors.

These results could be due to the high level of cognitive heterogeneity showed by people with SSD and violent behavior, as reported in the most recent and comprehensive systematic review and meta-analysis on this topic that highlighted a high level of heterogeneity both within and between studies [27]. Moreover, previous studies report that patients convicted in forensic settings could participate in cognitive tests with greater engagement, resulting in better scores when compared to controls [52].

However, it can also be hypothesized that patients with SSD convicted for violent crimes could show a peculiar cognitive profile, and that some patients with specific clinical correlates could show an overall better cognitive performance or a better performance in some specific domains. Concerning factors that are more closely correlated with violent outcomes, OP showed significantly higher scores in all HCL-20 subscales and all PCL-R factor, which represents an expected outcome.

A higher number of school failures and a higher score on the PANSS-EC emerged as socio-demographic and clinical predictors of belonging to the OP group. Worse performance at the Stroop Test time and worse performance at the BACS Token Motor Task, both measuring processing speed, and better Stroop error performance emerged as individual cognitive predictors. This result highlights the importance of this factor, pointing out that specific cognitive correlates might also predict important and distal outcomes such as violent behavior and suggesting that a thorough assessment of cognitive abilities is warranted in people with SSD, not only in the perspective of implementing specific treatments and to better personalize the rehabilitation program [53,54,55,56] but also, potentially, to increase the prevention of violent acts.

Higher scores on the HCR-20 Risk Management subscale and the PCL-R “Callous” factor also emerged as individual predictors of violent behavior, confirming the importance of using both instruments in the assessment of forensic patients.

Moreover, this regression model explained the largest portion of the variance. This was an expected result: on one hand, this finding remarks the importance of psychopathic and aggressive traits in determining violent behavior, but on the other, it confirms the need to design separate regression models when taking into account different categories of variables, including cognitive and socio-demographic characteristics, as potential predictors of violence.

Lastly, correlation analyses conducted in the whole sample show that a worse cognitive performance in several domains appears to be associated with a higher score in the PCL-R “Unstable” factor, whereas a better attentive performance is correlated, alongside a lower BIS-11 attentional impulsivity sub-scale score, with a higher score in PCL-R “Callous” factor. This represents a very interesting finding, suggesting that a better cognitive performance could identify patients who are more prone to callousness, manipulation and, perhaps, premeditating violent acts. On the contrary, cognitive deficits appear to identify those patients with more marked emotional instability, which could in turn be more prone to violent outbursts.

These results are particularly interesting because they could at least partly explain the heterogeneity reported in previous literature regarding cognitive characteristics of offenders diagnosed with SSD, highlighting a cognitive profile that is globally worse compared to non-offender patients but showing peculiarities that are worthy of further research.

Future studies should investigate whether it could be possible to identify distinct categories of patients on the basis of these characteristics: one characterized by callous aspects, with a globally less impaired cognitive profile and a better performance in some specific domains, and one characterized by aspects of impulsivity and instability with more prominent cognitive deficits.

Identifying characteristic profiles could have a positive impact not only to better characterize individual patients, but also in the development and implementation of personalized rehabilitation interventions. In fact, patients with more severe cognitive impairment could have consistent benefits from integrating pharmacological therapy and cognition-oriented treatments, including cognitive remediation and physical exercise [26, 35, 57,58,59], while other behavioral and pharmacological approaches could be more effective in other patients. Moreover, the feasibility and efficacy of a cognitive remediation intervention in a forensic setting have been already investigated in previous studies, showing promising results [60,61,62]. This could produce substantial benefits for the clinical condition as well as on functional outcomes of patients but also, at least partially, help in avoiding and preventing acts of violence.

Another interesting perspective for future studies could be to assess the impact of cognitive characteristics on different type of offenders diagnosed with SSD: better defining the cognitive correlates of early and late starters, which show several behavioral, clinical, and personality traits differences [63], could further improve the possibility to personalize treatment and prevention strategies.

This study presents a series of strengths. Patients were recruited in real-world setting and the sample can be considered quite representative of the population of individuals diagnosed with SSD and convicted in forensic settings. The assessment comprised a wide array of well-validated tools, considering also different test for several cognitive domain and including an evaluation of social cognition performance. Conducting separate analyses for different categories of variables and including sub-scales and factors of several assessment tools allowed to carefully evaluate a series of nuances in the participants’ cognitive, clinical, and behavioral profile.

The main limitation of the present study is represented by the small sample size, which did not allow to better identify and define categories of patients. Moreover, comparisons between hypothetical categories could not be performed, again for the small size of the sample: the fact that this could, at least in part, explain the observed heterogeneity and the between-variables correlations could represent an interesting perspective that has to be further explored in future studies. Another limitation is that overall SSD symptoms severity was only assessed with the PANSS-EC; a more comprehensive assessment using the full PANSS or, even better, using more recent and accurate tools for positive and negative symptoms severity assessment [64] could provide useful insight and should be considered in future studies.

In conclusion, the present study highlights the importance of carefully assessing people living with SSD with violent behavior in all clinical, cognitive, and behavioral aspects. In particular, assessing cognitive performance could provide interesting insights both in a research and in a clinical perspective. In fact, a careful assessment of all these variables could considerably help in the definition and implementation of personalized rehabilitation programs, which could be more effective both in improving patients’ clinical conditions and help in avoiding and preventing violent behaviors.