Risk factors for violent crime in patients with schizophrenia: a retrospective study

Study design and participants

The methodology employed in this study is based on the approach detailed in our preprint (Lin, Li & Liu, 2023). From January 1, 2013, to December 31, 2016, we retrospectively analyzed archival data and medical records of patients with schizophrenia who received forensic psychiatric assessments in the Judicial Psychiatric Appraisal Unit of the Second Xiangya Hospital of Central South University. We planned to conduct multiple regression analysis to examine the factors related to violent crimes by patients with schizophrenia, and the required number of subjects is ten times the number of independent variables (Okamura et al., 2021). Since there were approximately 25 independent variables in this study, the sample size needed to be 250, and considering a 10% refusal rate and potential data loss, we expanded the final sample size to 275 people. Based on the presence or absence of violent crimes, patients were categorized into the case group and the control group. Within the case group, further subgroups were formed based on the severity of violence, including homicide and intentional injury categories. The inclusion criteria were (1) aged 18–60 years, (2) diagnosed with schizophrenia according to the International Classification of Diseases criteria, 10th edition (ICD-10) (Geneva), which was the specific criteria used for the diagnosis, based on a clinical diagnostic interview conducted by at least two qualified forensic psychiatrists, (3) with normal intelligence. The exclusion criteria were (1) patients with incomplete socio-demographic data in the case file, (2) patients who were uncooperative in the mental status assessment, (3) patients with hearing loss or verbal communication difficulties that precluded them from completing the assessment, and (4) patients with comorbidities of other severe mental disorders (such as mental retardation, dementia, epilepsy and severe traumatic brain injury, etc.). Data collected from each patient’s file data and medical records were recoded to remove any identifiable information (e.g., name or date of birth). All methods were performed in accordance with relevant guidelines and regulations. All participants provided informed consent. The study was approved by the Ethics Committee of the Second Xiangya Hospital (2013068).

Definition of violence

In the present study, violent crimes refer to crimes involving serious acts of violence, including homicide, aggravated assault, robbery, sexual coercion, and rape. Nonviolent crimes involve other relatively benign offenses such as smuggling, drug trafficking, theft, fraud, and racketeering. To explore the potential differences in influencing factors between violence of different severities, the violence group was further divided into the homicide group and the intentional injury group. The homicide group included those who had committed murder, manslaughter and infanticide.

Tools and assessments

A self-designed standardized form was used to record the basic demographic and clinical information of all subjects. Demographic data included age, gender, level of education, marital status, household registration, ethnicity, and work status; the following clinical data were recorded: duration of illness, duration of untreated psychosis, previous treatment, regular treatment in the previous four weeks, family history of mental disorders, history of drug use, alcohol abuse, history of violence, and history of self-harm. Based on a prior study (Witt, Van Dorn & Fazel, 2013), we assigned codes to specific symptoms that have been found to be associated with violence in psychosis. These symptoms include command hallucinations, persecutory delusions, delusions of jealousy, depression, excitement, apathy, and insight (Witt, Van Dorn & Fazel, 2013). By reviewing forensic files or assessment records, the researchers identified whether patients exhibited any of the above psychotic symptoms within days to a week before the index offense. Each symptom was rated 0 (not present) or 1 (present).

Statistical analysis

All statistical analyses were performed using the IBM SPSS 25 software. The Shapiro–Wilk test was used to check the distribution of continuous variables. Characteristics of all participants were summarized as frequency (percentage) for categorical variables and median and interquartile range (IQR) for continuous variables. For the comparison of categorical variables, the chi-square test or Fisher’s exact test was used. Given the non-normal distribution of continuous variables, the comparison between two groups was conducted using the Mann–Whitney U test, and the comparison among three groups was performed using the Kruskal-Wallis test. Post hoc comparisons were made using the Bonferroni correction for multiple comparisons. We assessed the missing data and excluded variables with missing data exceeding 30% from the analysis, such as grandiose delusions, irritability, and affective ambivalence. The tolerance and the variance inflation factor (VIF) were calculated for each independent variable of the final model to detect multicollinearity. VIFs over 10 indicated significant multicollinearity that needed to be corrected (Menard, 2002). Binary logistic regressions (forward: LR) were conducted, variables with P < 0.10 in the univariate analysis used as independent variables and violent crime used as the dependent variable, to examine the independent correlates of violent crime. The level of statistical significance was set at 0.05 (two-tailed).

Sociodemographic characteristics

The patient recruitment process is presented in Fig. 1. A total of 447 (94.5%) patients completed the interview. In the violent group, six were excluded for hearing difficulties, 10 were excluded for uncooperativeness and five were excluded for incomplete information. In the non-violent control group, three were excluded due to hearing difficulties and two were excluded for uncooperativeness. Finally, 308 patients with schizophrenia who engaged in violent criminal cases (155 in the homicide group and 153 in the intentional injury group) and 139 non-violent patients with schizophrenia were included in this study. The median and IQR for both violent and non-violent groups were 34.0 (28.0–42.0). There were no significant differences between the two groups in age, ethnicity, marital status, and family history of mental disorders. Compared to the non-violent group, the violent group was more likely to be males (P = 0.020), have low educational attainment (P = 0.042), be unemployed (P = 0.024), and live in rural areas (P = 0.002). Furthermore, the homicide group was more likely to live in rural areas compared to the non-violent group (see Table 1).

Clinical characteristics

The violent group did not differ from the non-violent group in terms of duration of illness, alcohol abuse, history of drug use, and history of self-harm. However, there were significantly fewer patients having received previous treatment (P < 0.001), regular treatment in the previous four weeks (P < 0.001), but more patients having a history of violence (P < 0.001) in the violent group.

Compared to the intentional injury group, we found that the homicide group was less likely to have received a regular treatment in the previous four weeks. Additionally, compared to the non-violent group, the homicide group exhibited lower probabilities of previous treatment, regular treatment in the previous four weeks. They were more likely to have a history of violence (Table 2).

Psychopathology and violence

There was no difference between the violent group and the non-violent group with regard to command hallucinations, depression, and excitement. Compared to the non-violent group, the violent group were more likely to have persecutory delusions (P < 0.001), delusions of jealousy (P = 0.004), and apathy (P = 0.011). Additionally, they exhibit a longer duration of untreated psychosis (P = 0.028) and poorer insight (P < 0.001). Furthermore, in comparison to the non-violent group, the homicide group were more likely to have persecutory delusions and apathy, along with poorer insight (see Table 2).

Risk model for violence

The variables that demonstrated significance in the univariate analysis, including gender, education, work status, household registration, previous treatment, duration of untreated psychosis, regular treatment in the previous four weeks, history of violence, persecutory delusions, delusions of jealousy, apathy, and insight, incorporated into a binary logistic regression model. This model aimed to ascertain the predictive factors associated with violent behaviors. After gradually adjusting for all potential confounders, persecutory delusions, history of violence, regular treatment in the previous four weeks, and insight were independent predictors of violent behaviors. The results of the multicollinearity test indicate that the tolerance values for the variables included in the final model are all greater than 0.2, and all variance inflation factors (VIFs) are less than 2 (see Table 3). This indicates that there is no covariance. Later, these independent predictors were used as independent variables and violence as dependent variables to construct the risk model. Generally, factors with OR values greater than 1 were considered as risk factors for violence, whereas those with OR values less than 1 were regarded as protective factors for violence. Specifically, the risk model of violence can be presented as the following equation:

${\displaystyle \text{Logit}P=0.30-1.25\ast \text{regular treatment in the previous four weeks}-1.22\ast \text{insight}+1.06}$ ${\displaystyle \ast \text{violence history}+0.95\ast \text{delusions of persecutory}.}$

Hosmer-Lemeshow test showed the fitness of this risk model was acceptable (χ² = 1.044, P = 0.959); the total degree of accuracy of this model was 75.8%, and the degree of accuracy for the absence and presence of violence was 35.3% and 94.2%, respectively, indicating that this model was sensitive for the identification of the risk of violence. After controlling for sociodemographic variables (e.g., gender, age). The multivariate analysis revealed that persecutory delusions (OR = 2.57, 95% CI [1.63–4.06]) and history of violence (OR = 2.88, 95% CI [1.79–4.64]) were independently associated with increased risk of violent behaviors, whereas regular treatment in the previous four weeks (OR = 0.29, 95% CI [0.16–0.51]) and insight (OR = 0.30, 95% CI [0.14–0.62]) were independently associated with reduced risk of violent behaviors (see Table 4).

Limitations

Despite the strengths of this study, there are also some limitations. Firstly, as this study adopted a cross-sectional design, the temporal relationships between violent behaviors and its correlates could not be investigated in this study. As the relationship between psychopathological factors and violence might be dynamic and fluctuating over time (Coid et al., 2013; Dorn et al., 2017), longitudinal studies are warranted to examine such a relationship. Secondly, because the scope of this study is limited to Hunan Province, our results may not be universally applicable. Therefore, these findings should be interpreted and generalized with caution. Thirdly, this study relied on medical records and assessment notes to identify the presence of psychotic symptoms in patients before committing a crime, which may lead to the omission of symptoms and introduce certain biases. Additionally, because this study only assessed the presence or absence of symptoms, it is difficult to compare these symptoms with other studies or groups, thereby limiting the comparability of the research. We recommend that future studies conduct prospective studies, use standardized tools to assess psychotic symptoms, and establish more comprehensive recording systems to reduce these biases.