Suicidal Offenders and Non-Offenders with Schizophrenia Spectrum Disorders: A Retrospective Evaluation of Distinguishing Factors Using Machine Learning

Abstract

Patients with schizophrenia spectrum disorders (SSD) have an elevated risk of suicidality. The same has been found for people within the penitentiary system, suggesting a cumulative effect for offender patients suffering from SSD. While there appear to be overlapping characteristics, there is little research on factors distinguishing between offenders and non-offenders with SSD regarding suicidality. Our study therefore aimed at evaluating distinguishing such factors through the application of supervised machine learning (ML) algorithms on a dataset of 232 offenders and 167 non-offender patients with SSD and history of suicidality. With an AUC of 0.81, Naïve Bayes outperformed all other ML algorithms. The following factors emerged as most powerful in their interplay in distinguishing between offender and non-offender patients with a history of suicidality: Prior outpatient psychiatric treatment, regular intake of antipsychotic medication, global cognitive deficit, a prescription of antidepressants during the referenced hospitalisation and higher levels of anxiety and a lack of spontaneity and flow of conversation measured by an adapted positive and negative syndrome scale (PANSS). Interestingly, neither aggression nor overall psychopathology emerged as distinguishers between the two groups. The present findings contribute to a better understanding of suicidality in offender and non-offender patients with SSD and their differing characteristics.

Machine Learning for prediction of violent behaviors in schizophrenia spectrum disorders: a systematic review

“…Among these studies, four employed ML models to predict VB during the current admission ( 35 , 41 , 47 , 51 ). Additionally, nine studies categorized patients based on the occurrence of VB prior to their current admission ( 38 – 40 , 43 , 44 , 46 , 48 – 50 ), while another four classified patients into violent and non-violent groups by retrospectively reviewing their medical records since their disease onset ( 36 , 37 , 42 , 45 ). Moreover, eight studies were part of a larger project investigating the relationship between SSD and offending and used the same dataset of offender patients as their sample population ( 39 , 41 , 42 , 45 , 46 , 48 – 50 ).…”

“…Most of the included studies utilized only sociodemographic and clinical features of patients to predict VB. Of these studies, five evaluated a large number of features (over 100 features) as predictors ( 39 , 41 , 45 , 49 , 50 ). Tzeng et al.…”

“…(2022, 2023) in two separate studies evaluated the differences between offender and non-offender SSD patients regarding psychiatric pr e scription patterns and illness-related factors ( 46 , 49 ). Also, ten studies analyzed the relationship between different rating tools scores and VB in patients with SSD ( 36 , 38 , 39 , 41 , 43 , 45 , 46 , 48 – 50 ), including the Brief Psychiatric Rating Scale (BPRS) ( 38 , 43 , 52 ), the Psychopathy Checklist: Screening Version (PCL-SV), the Historical, Clinical and Risk management (HCR-20) scale ( 38 , 53 ), The Barratt Impulsiveness Scale version 11 (BIS-11) ( 38 , 54 ), the Positive And Negative Symptom Scale (PANSS) ( 36 , 39 , 41 , 43 , 45 , 46 , 48 – 50 , 55 ), the Social Disability Screening Schedule (SDSS) ( 43 ), Insight and Treatment Attitude Questionnaire (ITAQ) ( 47 , 56 ), Family Adaptation, Partnership, Growth, Affection and Resolve (APGAR) ( 47 , 57 ), Social Support Rating Scale (SSRS) ( 47 , 58 ), and Family Burden Scale of Disease (FBS) ( 47 , 59 ). Furthermore, two studies evaluated neuroimaging data of patients as VB predictors, along with sociodemographic features.…”

“…(2022) included physical VB regardless of the aim ( 41 ). On the other hand, six studies used a shared database to distinguish between violent and non-violent offenses ( 39 , 42 , 46 , 48 – 50 ). In a seventh study, they attempted to predict the risk of homicide among other offenses ( 45 ).…”

“…Seven studies employed data-driven feature selection by ML before model training to control overfitting: one utilized LASSO ( 38 ), three used RF ( 39 , 41 , 45 ), one applied boosted tree ( 42 ), one utilized both LASSO and LR ( 43 ), and one selected features after calculation of variable importance for each employed model separately ( 51 ). Sixteen studies used cross-validation, with two using 10-fold cross-validation ( 43 , 44 ), one using 7-fold ( 36 ), nine using 5-fold ( 37 , 39 , 41 , 42 , 45 , 46 , 48 – 50 ), one using 4-fold ( 47 ), and one using 3-fold ( 34 ). Two studies did not use cross-validation ( 35 , 40 ).…”