AUTHOR=Beaudoin Mélissa , Potvin Stéphane , Dellazizzo Laura , Luigi Mimosa , Giguère Charles-Edouard , Dumais Alexandre 

TITLE=Trajectories of Dynamic Risk Factors as Predictors of Violence and Criminality in Patients Discharged From Mental Health Services: A Longitudinal Study Using Growth Mixture Modeling

JOURNAL=Frontiers in Psychiatry

VOLUME=Volume 10 - 2019

YEAR=2019

URL=https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2019.00301

DOI=10.3389/fpsyt.2019.00301

ISSN=1664-0640

ABSTRACT=Background: Individuals with severe mental illnesses are at greater risk of offences and violence, though the relationship remains unclear due to the interplay of static and dynamic risk factors. Static factors have generally been emphasized, leaving little room for temporal changes in risk. Hence, this longitudinal study aims to identify subgroups of psychiatric populations at risk of violence and criminality to reduce these behaviors, considering the dynamic changes of symptomatology and substance use.
Method: A total of 825 patients from the MacArthur Violence Risk Assessment Study having completed 5 post-discharge follow-ups were analyzed. Individuals were classified into outcome trajectories (violence and criminality). Trajectories were computed for each substance (cannabis, alcohol and cocaine, alone or combined) and for symptomatology and inputted as dynamic factors, along with other demographic and psychiatric static factors, into binary logistic regressions for predicting violence and criminality. Best predictors were then identified using backward elimination and ROC curves were calculated for both models. 
Results: Two trajectories were found for violence (Low vs. High violence). Best predictors for belonging in the High-violence group were low verbal intelligence (baseline), higher psychopathy (baseline) and anger (mean) scores, persistant cannabis use (alone) and persistent moderate affective symptoms. The model’s AUC was 0.773. Two trajectories were also chosen as being optimal for criminality. The final model to predict High-criminality yielded to an AUC of 0.788, retaining as predictors male sex, lower educational level, higher psychopathy score (baseline), persistent polysubstance use (cannabis, cocaine and alcohol) and persistent cannabis use (alone). Both models were moderately predictive of outcomes. 
Conclusion: Static factors identified as predictors are consistent with previously published literature. Concerning dynamic factors, unexpectedly, cannabis alone was an independent co-occurring variable in the violence model, as well as affective symptoms. For criminality, our results are novel because a very there are very few studies on criminal behaviors in a psychiatric non-forensic population. In conclusion, these results emphasize the need to study more the impact of longitudinal patterns of specific substance use and high affective symptoms, and to evaluate more profoundly the predictors of crime, separately from violence.