Publication
Article
Psychiatric Times
Author(s):
The use of validated and scalable risk prediction tools can potentially improve mental health outcomes.
Oleksandr/AdobeStock
SPECIAL REPORT: FORENSIC PSYCHIATRY
Clinical psychiatry is shifting toward precision medicine. Until now, one-size-fits-all strategies have had a patchy success rate, leaving many patients with delayed diagnoses, uncertainty about their care, and suboptimal treatment. By harnessing the opportunities provided by big data, new research is investigating how diagnosis, prognosis, and treatment can be personalized.
In terms of prognosis, clinical prediction models can provide individual-level risk estimates for important future outcomes.1 These models statistically combine data on risk factors from large data sets to assess the risks of serious adverse outcomes for mental health, such as suicidal behaviors, violence toward others, and severe relapses in mental state. These can be translated into risk assessment tools, which allow clinicians to input information on individual
Leveraging and Balancing Skills in a Big Data Era
James L. Knoll IV, MD
Issues in Civil Commitment: The Cases of Incarcerated Patients
Esther Schoenfeld, MD; Raina Aggarwal, MD; and Danielle B. Kushner, MD
Understanding and Evaluating Conspiracy Theories: A Primer for the General and Forensic Psychiatrist
Brian Holoyda, MD
factors, which can be used to calculate a risk for that individual based on the underlying statistical model. Such tools based on high-quality prognostic models can complement clinical assessment and assist clinicians in making more evidence-based decisions, particularly early on in the patient pathway. This is especially important for estimating the risk of perpetration of violence and criminality, which is increased in those with severe mental illness and affects a minority of patients.
For example, individuals with schizophrenia spectrum disorders have a risk that is 2 to 5 times higher than the general population (including in studies with careful adjustment for confounders, in individuals without diagnosed comorbidities, and in sibling comparisons that inherently account for shared background familial factors such as early environment).2 This risk increases further with comorbid substance misuse,2 and absolute rates vary according to illness stage and setting.3 Moreover, a review of the prevalence of violent outcomes in psychiatric patients found some patient groups with elevated rates. This included around a fifth of individuals who present in psychiatric emergency settings in the next 12 months and more than one-third in involuntarily committed patients and patients with first-episode psychosis, also over the subsequent 12 months (Figure 1).3 More recent work in first-episode psychosis has also found that around 1 in 10 individuals have violent outcomes that lead to police contact in the year after referral to mental health services.4
FIGURE 1. Estimated Mean Prevalence of Violence Perpetration in Psychiatric Patients by Study Setting Over 6 to 12 Months: Pooled Estimate3
These observational studies’ findings highlight the need for accurate and validated violence prediction tools in this population, which, if linked to effective interventions, can assist in reducing future risks. Although some risk assessment approaches and instruments exist, such as the Historical Clinical Risk Management-20, they are mainly used in forensic mental health and may have limitations if used in general psychiatry. One significant limitation is that they are often overly time-consuming and costly. For example, some of these older risk assessment approaches, particularly those described as structured clinical judgment tools, take up to 16 hours to complete,5 making them impractical for clinical use. Another shortcoming is that they do not provide probability scores for individuals but only categorize them into broad groups of low, medium, and high risk. However, classification (using cutoffs) is not a good aim when models are relatively accurate—here, you will want to know probabilities. This is exactly how the Framingham Risk Score (or, for that matter, the weather) is communicated—not in a categorical way of yes/no but in relation to probabilities, which allows individuals to decide how to act depending on the decision to be made and its implications (eg, changing diet or adding a statin or not). Additionally, structured clinical judgment tools typically show poor predictive accuracy in real-world clinical settings compared with research settings, raising concerns about their everyday clinical use. This is likely because these tools were developed using small sample sizes, focused on a specific patient population (eg, forensic patients), and were not externally tested in real-world settings different from the one in which they were developed.6,7
Therefore, novel tools are required to assist clinicians in making well-informed decisions, facilitating early interventions, and ensuring consistency in risk assessment within and across clinical teams. A new generation of tools that provide probability scores have been developed using population-based registers that cover a wide range of patient populations and provide extensive information about them over time.
One of these tools, using data from more than 75,000 individuals with severe mental illness in Sweden, is the Oxford Mental Illness and Violence (OxMIV) tool, which aims to assist clinicians in assessing the risk of violence within 12 months of assessment among individuals with schizophrenia spectrum and bipolar disorders. The tool was built with 16 risk factors, including demographic, criminal history, and clinical variables, such as age, sex at birth, previous violent crime, past drug or alcohol misuse, and recent antipsychotic treatment.8 The OxMIV tool has been validated and updated across diverse patient populations in England, Germany, and the Netherlands,9-11 demonstrating applicability in different clinical settings. Clinicians can easily access its user-friendly interface online, making it simple to integrate into practice (Figure 2).
FIGURE 2. OxMIV’s Web-Based Interface
For example, when a clinician assesses the risk of violence in an individual with schizophrenia spectrum disorders or bipolar disorder, they can quickly enter all necessary data into the OxMIV calculator and estimate the individual’s risk level in percentages. Risk estimates become less accurate and clinically useful at very high levels; therefore, they are presented at a maximum of 20%. Importantly, OxMIV allows unknown values for some risk factors (providing a range of predicted risks).
Integrating OxMIV and similar novel tools into clinical practice can complement clinical assessment and improve clinician confidence in making early personalized decisions to improve patient care, such as:
As psychiatry embraces precision medicine, the use of validated and scalable risk prediction tools can potentially improve mental health outcomes. Integrating these tools into clinical practice can assist violence risk assessment and allow for more personalized decisions early on in the treatment process, benefiting patients and their families while improving safety and well-being.
Dr Fazel is a professor of forensic psychiatry at the University of Oxford in England. Mr Scola is a postdoctoral research associate (pending PhD viva) in psychiatric epidemiology at the University of Oxford in England.
References
1. Oliver D, Arribas M, Perry BI, et al. Using electronic health records to facilitate precision psychiatry. Biol Psychiatry. 2024;96(7):532-542.
2. Whiting D, Gulati G, Geddes JR, Fazel S. Association of schizophrenia spectrum disorders and violence perpetration in adults and adolescents from 15 countries: a systematic review and meta-analysis. JAMA Psychiatry. 2022;79(2):120-132.
3. Swanson JW, McGinty EE, Fazel S, Mays VM. Mental illness and reduction of gun violence and suicide: bringing epidemiologic research to policy. Ann Epidemiol. 2015;25(5):366-376.
4. Whiting D, Lennox BR, Fazel S. Violent outcomes in first-episode psychosis: a clinical cohort study. Early Interv Psychiatry. 2020;14(3):379-382.
5. Viljoen JL, McLachlan K, Vincent GM. Assessing violence risk and psychopathy in juvenile and adult offenders: a survey of clinical practices. Assessment. 2010;17(3):377-395.
6. Fazel S, Wolf A, Vazquez-Montes MDLA, Fanshawe TR. Prediction of violent reoffending in prisoners and individuals on probation: a Dutch validation study (OxRec). Sci Rep. 2019;9(1):841.
7. Ogonah MGT, Seyedsalehi A, Whiting D, Fazel S. Violence risk assessment instruments in forensic psychiatric populations: a systematic review and meta-analysis. Lancet Psychiatry. 2023;10(10):780-789.
8. Fazel S, Wolf A, Larsson H, et al. Identification of low risk of violent crime in severe mental illness with a clinical prediction tool (Oxford Mental Illness and Violence tool [OxMIV]): a derivation and validation study. Lancet Psychiatry. 2017;4(6):461-468.
9. Whiting D, Mallett S, Lennox B, Fazel S. Assessing violence risk in first-episode psychosis: external validation, updating and net benefit of a prediction tool (OxMIV). BMJ Ment Health. 2023;26(1):e300634.
10. Negatsch V, Voulgaris A, Seidel P, et al. Identifying violent behavior using the Oxford mental illness and violence tool in a psychiatric ward of a German prison hospital. Front Psychiatry. 2019;10:264.
11. Lamsma J, Yu R, Fazel S; Genetic Risk and Outcome of Psychosis (GROUP) investigators. Validation and recalibration of OxMIV in predicting violent behaviour in patients with schizophrenia spectrum disorders. Sci Rep. 2022;12(1):461.
12. Whiting D. Improving Violence Risk Assessment and Intervention in First Episode Psychosis. Doctoral thesis. University of Oxford; 2021.
