Research over the last few decades has reported high rates of victimization in people with severe mental health problems1, and this is increasingly viewed as a key adverse outcome to prevent. Consequences can arise directly: more commonly, worsening of psychiatric conditions through the effects of trauma, but also physical health morbidities and even death. Indirect consequences may be disruptions to care, breakdowns in social support and networks, and the harmful use of drugs and alcohol to manage the physical and psychological effects of victimization.

However, despite the importance of the issue, research designs have had until now significant limitations. Many studies have used cross-sectional designs, asking people with vs. without psychiatric conditions to report on their victimization histories. This approach can be informative, but is likely to overestimate the association with psychiatric conditions, as people who are unwell are more likely to attribute their current problems to external causes. More importantly, these studies cannot deal with reverse causality – that the victimization has led to severe mental health problems rather than the reverse. This information remains useful to estimate needs, but not in terms of understanding causal links, which is necessary for prevention.

These designs are particularly problematic when rates of victimization are compared with other adverse outcomes, such as violence perpetration, as thresholds and time scales for these outcomes may be different. The commonly repeated statement that psychiatric patients are ten times more likely to be victims of crime than the general population, and that this rate is higher than the perpetration rate, is based on research with these suboptimal designs.

More informative are cohort studies, which can account for the timing of victimization and mental health conditions. Birth cohorts in the UK2 and New Zealand3 have reported that the following factors increase victimization risk: male gender, self-reported financial difficulties (but not other more objective markers of socioeconomic status), and comorbid alcohol and cannabis dependence. Confounds can be accounted for, but only those that are measured, and measured accurately. Residual confounding is, therefore, a threat to the validity of these studies.

One way of addressing such residual confounds is to use genetically informed controls, such as siblings. With biological full-sibling controls, half the co-segregating genes and much of the early environment are accounted for, which most observational studies do not capture. Siblings with and without mental health conditions can be followed up for victimization outcomes and, after adjusting for age and using same-sex sibling controls, studies can rule out several alternative hypotheses and provide stronger evidence for the associations to be consistent with a causal inference.

One such study using Swedish registers examined more than 250,000 patients diagnosed with psychiatric disorders and compared them with nearly 195,000 of their full siblings without psychiatric disorders4. Those with psychiatric diagnoses were found to be about three times as likely as their siblings to be violently victimized, and there was a four-fold increase in perpetration of violence in psychiatric patients.

Another genetically informative cohort is the E-Risk twin study, which found that measures of victimization up to age 18 were at least moderately heritable (>30%) and correlated with other heritable traits, including lower self-control and cognitive abilities, childhood conduct disorder, substance misuse, and family history of mental illness and antisocial behaviours5. These findings underline the importance of accounting for unmea­sured genetic confounding in studies of victimization risk.

In the above-mentioned Swedish study4, the risk of victimization was increased three-fold in siblings with bipolar disorder and doubled in those with depression compared to siblings without mental health problems. Unexpectedly, the risk was not increased in siblings with schizophrenia-spectrum disorders compared to their unaffected siblings, which may be explained by the fact that people with such disorders are more socially isolated, with less opportunities to be victimized than others.

Another national investigation that used a novel design, in which individuals acted as their own controls (“within individual”), found that violent victimization was the strongest trigger for violent perpetration in psychotic disorders6. Consideration, therefore, should be given to providing psychosocial support for at least one week following any victimization, to minimize the risk of a cycle of violence.

What do these findings mean for psychiatrists, other mental health professionals, and services? First, there is a considerable overlap between violence perpetration and victimization. Any improvements are likely to lead to reductions across these outcomes, and may also reduce suicide and premature mortality. Second, research design is critically important in this area, since small study effects have been magnified by poor measurement in previous work. Third, prevention will require two components: better risk stratification and effective interventions.

Risk stratification is required to determine who can benefit from additional interventions aimed at prevention, which will likely be resource intensive and complex. Criticisms of risk assessment rarely consider real world implications: psychiatric services need to stratify in order to allocate resources effectively, transparently and consistently, and cannot provide gold standard interventions to all people with mental health problems.

Most clinicians are unable to weigh up more than a few risk factors simultaneously, and very unlikely to make sense of their interactions. Once you reach more than five or so risk factors, assessment will benefit from simple algorithms to support, rather than replace, clinical decision-making. Simple scalable online tools with high negative predictive values can usefully screen out low-risk persons to preserve resources7.

But evidence-based risk assessment will only improve outcomes if linked to interventions, and effective ones. A key uncertainty is whether treating symptoms of mental illness will prevent victimization outcomes. There is some evidence suggesting that depressive symptoms may be predictive of victimization8, but this work needs replication.

Research on specific interventions aiming to reduce victimization risk in persons with mental disorders remains rare, because victimization has traditionally been viewed as a risk factor rather than a consequence of mental illness. One significant change would be to consider including victimization as an outcome in mental health treatment trials, particularly those that follow up people beyond a few weeks. Improving access to treatment for comorbid substance misuse is an important policy consideration, as research has clearly demonstrated that this comorbidity explains a large share of the elevated victimization risk in persons with mental illness4.

More contact with friends and family members may act as a protective factor against victimization risk, and supporting mea­sures to promote this can be enhanced across all mental health services. However, it is important to make sure that such interactions do not actually lead to increased exposure to criminogenic environments9. Finally, large-scale clinical and genetically informed studies, preferably linked with registry data and electronic health records, may clarify specific etiological mechanisms involved, leading to trials of interventions targeting these mechanisms.