The HARM models: Predicting longitudinal physical aggression in patients with schizophrenia at an individual level

Area of Science:

• Forensic Psychiatry
• Clinical Psychology
• Psychiatric Epidemiology

Background:

• Aggression prediction and prevention are critical in forensic psychiatric settings for individuals with schizophrenia.
• Existing risk assessment tools lack the ability to predict longitudinal physical aggression at an individual level in this population.

Purpose of the Study:

• To develop and validate machine learning models (HARM models) for predicting longitudinal physical aggression in forensic psychiatric patients with schizophrenia.
• To identify modifiable risk and protective factors associated with future aggression.

Main Methods:

• Utilized baseline data on risk/protective factors and treatment course for 151 schizophrenia patients in forensic mental healthcare.
• Employed machine learning models (HARM models) to predict physical aggression at 4, 12, and 18-month follow-ups.
• Compared HARM model performance against clinician-rated risk assessments.

Main Results:

• HARM models demonstrated balanced accuracy ranging from 59.73% to 87.33% for predicting physical aggression across follow-up periods.
• Identified key modifiable predictors: impulse control, substance abuse, impulsivity, treatment non-adherence, mood/psychotic symptoms, and family support.
• Models showed promise but require larger cohorts for broader generalizability due to low aggression base rates.

Conclusions:

• The HARM models represent a novel, individual-level approach to predicting longitudinal physical aggression in forensic schizophrenia patients.
• Identified modifiable factors offer targets for intervention to reduce aggression.
• Findings highlight the need for further validation in larger, diverse cohorts.