Predicting Violence in Schizophrenia: an unfulfilled prophecy

Introduction #

The need to predict #

While the ability to predict the future with certainty would be a superpower for anyone, the justice system would probably deliver the most value out of this superpower. This trope ultimately shows in Spielberg’s movie Minority Report (Spielberg, 2002), where crime predictors allowed the state to virtually eliminate crime.

Even if a thing like certain prediction is not possible, probabilistic predictions could also be of value. Poldrack et al. state the reasons this can be the case, as such predictions could guide judgements regarding bail, probation or parole, as well as lower state economic losses caused by violent crime (2018).

Prediction tools and their categories #

Science has provided, to the justice system, tools for exactly this need. These tools claim to predict the possibility of an individual committing a violent crime. They are split into three main categories, depending on how they work. The clinical tools that rely on a clinicians experience to provide a prediction, the structured (or actuarial) instruments that work like inventories - using scales and statistically combining their result to create a prediction, and finally the structured clinical/professional judgment (SCJ or SPJ) ones that are a combination of the previous two (Singh et al., 2011; Poldrack et al, 2018).

Additionally, for these instruments to provide their results, they measure factors that could predict violence, named risk factors. Douglas & Skeem have argued for a distinction between static factors, such as precedence of violence, and proposed dynamic factors, such as the status of interpersonal relationships of an individual (2005). These two types of factors create the landscape of Risk State (Douglas & Skeem, 2005).

Prediction on mentally disordered offenders #

As can be deduced by the other dynamic factors proposed by Douglas & Skeem, it becomes evident that they are suitable to target mentally disordered offenders. Specifically, psychosis is one of these dynamic factors (2005). Indeed, as psychosis is a positive symptom of schizophrenia, it does account for a positive correlation between violence and schizophrenics (Bo et al., 2011).

Schizophrenia as a cause of violence #

Only 10-15% of patients with schizophrenia have demonstrated violent behaviour, as described in the systematic review by Singh et al. (2011). This could mean, that, as Large et al. put it: “in order to prevent one homicide of a stranger, 35,000 high-risk patients with schizophrenia would require completely successful individual risk management”, with successful individual risk management often meaning detainment. (2011, p 28). This is also backed by studies such as the one by Michel et al (2013). Additionally, there is also a need to define violence, often pinpointed by the literature on that subject (Bo et al., 2011; Michel et al., 2013), as not all violence falls in the category of criminal offense, hence predicting it does not provide direct value to the justice system, which is the main client of this research (Poldrack et al, 2018).

As Bo et al. pointed out, psychosis is only one of the factors responsible for violent behaviour in schizophrenics (2011). Other factors exist and also sometimes are not directly tied to the disorder at all. For example, demographic factors, such as economic deprivation, social living status, age, even gender are predictors of violence in general psychiatric population (Bo et al., 2011).

The work by Michel et al. on evaluating HCR-20 on patients with schizophrenia also mentions that predictability on a patient can differ on the setting where the evaluation is conducted (2013). Specifically, the predicted probability for violence is higher when the patient is hospitalized, or when the psychotic symptoms are more intense, and lower when the patient lives in the community (Michel et al., 2013). Finally, this study has some positive outcomes, showing that HCR-20 could successfully predict aggressive behaviour in a 24-month window, yet it could not do so in the first 6 months after discharge.

Finally, the systematic review of Singh et al. (2011) evaluating ten structured violence risk assessment tools (including HCR-20), concluded that “there was little direct evidence to support the use of these risk assessment tools in schizophrenia, specifically” (2011, p 904).

Deploying Neuroscience #

As the “traditional” approach seems so inconclusive, the focus has also been switched to neuroscience. As Poldrack et al. beautifully put it “The promise of opening the black box of the human mind […] generated substantial excitement amongst the general public. […] and nowhere is the level of anticipation higher than in the courts, because law - more than almost any other profession - faces daily the challenge of rendering judgements based on the contents of that black box” (2018, p 5).

The field of predicting violent behaviours (or any behaviour) through neuro-imaging, referred to as “neuroprediction” does not seem very promising. Specifically, an example of why this field’s endeavors can be difficult is this approach of profiling violence. To profile violence, psychopathy has been studied and profiled, as psychopathy is often directly linked to violence. Yet, there is no evidence that this link is a link of causation, meaning that there also is non-violent psychopathy following the psychopathy profile (Poldrack et al, 2018).

While some brain region profiling of violence in patients with schizophrenia research has been done (Athanassiou et al., 2022), the study states that “the following data should not be used directly in forensic psychiatry settings to establish criminal non-responsibility” (Athanassiou et al., 2022; p 187).

Criticism #

Judging the future of an individual depending on calculations can have serious ethical problems, even if these calculations are scientific. In the case of the structured and SCJ instruments and schizophrenia patients, this seems to not be safe, as evidence remains inconclusive (Singh et al., 2011). But this is not the only qualm one can have about violence prediction.

De-generalization #

As the science of psychology and psychiatry, where these instruments are drawn from, are committed to revealing trends among populations, the justice system’s goal is the opposite: to judge each case individually. This can lead to what is referred to as the G2i problem, analyzed in the work of Poldrack et al. (2018). The gist of G2i (general to individual) problem is that while a population can reveal a trend, this does not mean that each individual of the population does follow the trend. If the established trend is the link between psychosis and violence, this cannot mean that everyone in the group, that reveals this association, is part of it.

The Static Factors #

As Douglas & Skeem differentiated between the Dynamic and Static factors (2005), it is possible to realize that the static factors often are demographic, such as gender, age, socioeconomic background, even race. Creating predictions using this kind of information can either mean that we believe a priori that biologically they play a role in violence (or other ill behaviours), which means that races and genders can be ordered from good to bad, or that we fall into the fallacy of explaining instead of predicting violence (Shmueli, 2010).

Conclusion #

Prediction is always prone to false-positives and false-negatives. Care is needed to treat results of all predictive tools for what they are: numbers that have been somehow calculated, always carrying a risk of error. But, as such, they have their use.

Also, using explanations to predict the future of an individual can be vastly problematic by itself. Probably the best example about it in the literature is the below extract, by the work of Poldrack et al. (2018, p 5):

The United States Supreme Court recently overturned a death penalty because an expert witness had testified that race was “know[n] to predict future dangerousness.” The Court stated that the case “is a disturbing departure from a basic premise of our criminal justice system: Our law punishes people for what they do, not who they are” (Buck v. Davis, 137 S. Ct. 759, 2017).

References #