The issue explored by the study by Hodgins and collaborators (Hodgins et al. Reference Hodgins, Cree, Alderton and Mak2007) is whether conduct disorder (CD) may be a modifiable risk factor for violent crime in persons with severe mental illness (SMI). Let us forget the methodological drawbacks of the study and consider that valid findings have been obtained, showing an association between CD and history of violent behaviour in persons with SMI. Let us also forget that this study does not provide direct evidence that CD is more frequent in persons with SMI compared to persons from the general population living in the same neighbourhood and with similar demographic characteristics. Notwithstanding these possible biases, several questions are still raised by the interpretation of these findings.

Persons with SMI are more likely to present with CD as well as with (almost) all other diagnoses listed in the current diagnostic classifications. Amazingly, the public, the mass media and elected representatives show little interest in the findings of studies exploring the public health implications of the co-morbidity between schizophrenia and obsessive–compulsive disorder, while they are prone to pay a disproportionate attention to any finding supporting the link between mental health and violence (Appelbaum, Reference Appelbaum2006). In other words, irrespective of the academic interest of studies exploring the causal pathway between CD, SMI and violence, such a research question cannot be fully disconnected from its societal impact, particularly regarding stigma. It is well established that a perceived dangerousness of persons with SMI is a key factor for increasing stigma (Penn et al. Reference Penn, Kommana, Mansfield and Link1999; Appelbaum, Reference Appelbaum2006), and that psychiatrists may contribute to stigmatizing attitudes towards persons with SMI; indeed, they may hold more negative stereotypical views about persons with SMI compared to individuals from the general population (Nordt et al. Reference Nordt, Rossler and Lauber2006). Even if it is legitimate to identify risk factors for violent behaviour in persons with SMI, a respectful and destigmatizing attitude vis-à-vis these persons would be to systematically include, in all communications regarding this issue, information about the fact that a very low proportion of violent behaviour is attributable to persons with SMI (Walsh et al. Reference Walsh, Buchanan and Fahy2002). For example, a population-based study in Sweden showed that persons with SMI were implicated in only 5% of all violent crimes (including homicide and attempted homicide, assault, robbery, threatening behaviour, harassment, arson, and any sexual crime) (Fazel & Grann, Reference Fazel and Grann2006). This proportion was 2% for persons with schizophrenia. Fazel and Grann estimated that one violent crime every 5 years per 1000 inhabitants could be attributed to patients with SMI. These findings are concordant with those reported in a literature review showing that the population-attributable risk fraction of violent crimes due to schizophrenia was always lower than 10% (Walsh et al. Reference Walsh, Buchanan and Fahy2002), and that this figure was probably overestimated since other risk factors for violence, such as substance use, were not taken into account. Even if the prevalence of CD is higher in persons with SMI compared to persons from the general population, the proportion of violent behaviour attributable to this co-morbid condition is hence probably much lower than 1%. Calculation of the relative risk of presenting with violent behaviour in persons with SMI (or CD, or both disorders) compared to persons from the general population does not shed any light on the population impact of this association. Even in deprived areas of inner big cities, recommendations regarding mental health services' organization and public policy decisions should be based upon population-attributable risk, and not upon relative risks.

Another closely related issue is the necessary distinction between risk factors increasing the likelihood to present with a given characteristic, and prediction for preventive purposes of occurrence of this characteristic at the individual level (Verdoux & Cougnard, Reference Verdoux and Cougnard2006). The confusion between aetiological research on risk factors and prediction is not merely a semantic issue, since it may favour the erroneous application of aetiological research findings for recommendations regarding implementation of preventive programmes. A well-known study of Swedish conscripts illustrates this point (Malmerg et al. Reference Malmerg, Lewis, David and Allebeck1998). Individuals presenting four premorbid characteristics (fewer than two close friends, preferring small groups, more sensitive than others, no steady girlfriend) were 30 times more likely to develop schizophrenia after several years of follow-up. Although these characteristics dramatically increased the risk of schizophrenia, they could not be used to predict the occurrence of the disorder at an individual level. Indeed, only 3% of the persons with these characteristics had actually developed schizophrenia. In the same perspective, knowledge that CD is a risk factor increasing the likelihood of violent behaviour in persons with SMI cannot be actually considered as useful in predicting violence at the individual level. Although the objective of preventing violent behaviour in persons with SMI is of course laudable, the feasibility and usefulness of a programme focused on identification and ‘treatment’ of CD in persons with SMI is questionable, and this argument should cautiously be used to justify the interest of studies on CD and schizophrenia. A further comment regarding justification of this programme: there should be no room in a research paper for over-emotional comments predicting a gloomy future for neglected children's patients if such programmes are not implemented.

From a public health point of view, current priorities seem to be elsewhere. It is first necessary to better assess the benefits of programmes aimed at preventing the occurrence of CD in children and adolescents from the general population. Second, for persons with SMI with (or without) co-morbid CD, a sensible strategy might be to allocate (limited) health resources to programmes with established efficiency, such as those aimed at improving treatment adherence or targeting co-morbid substance use disorder. As underlined by Hodgins and collaborators, it has been shown that the frequency of violent behaviour is decreased if the persons with SMI benefit from adequate treatment.

In conclusion, the lack of distinction between relative risk and population-attributable risk, or between risk factors and predictive factors, decreases the scientific credibility of the interpretation of research findings, but it is particularly worrying when it concerns burning topics with the potential for a strong public impact.