Violence and Mental Illness: Simplifying Complex Data Relationships
The blog depression introspection has an entry about this unfortunate post by the Treatment Advocacy Center, an advocacy organization that basically wants to paint some serious mental disorders as medical illnesses and demonize them (in order to increase access to medical treatments for them, yeah, that makes a lot of sense to us too).
When people make outrageous statements like this one,
The CATIE violence study found that patients with schizophrenia were 10 times more likely to engage in violent behavior than the general public (19.1% vs. 2% in the general population).
it really makes me mad. It’s clear the author has never actually read the study he’s commenting on (as few people actually bother to read the research, instead preferring to read other people’s summaries of the research, or an abstract). If you’re going to make statements about research, one of the requirements is that you actually read the study you’re commenting on.
As researchers in psychological disorders know, the connections between violence and mental disorders is complex. It’s not a simple, “Well, if you have X diagnosis, you’re Y times more likely to commit violence.” It’s not that at all.
Lots of people like to point to the Swanson (2006) study in the Archives of General Psychiatry as some sort of gold standard in answering the question, “Are people with schizophrenia more violent than others?” It is not. It has specific sampling issues that suggest the sample they had was not a representative sample at all. For instance, here are the exclusion criteria (e.g., these folks were excluded from the study):
Patients were excluded if they were in their first episode of schizophrenia; had a diagnosis of schizoaffective disorder, mental retardation, or other cognitive disorder; had past serious adverse reactions to any of the proposed treatments; had a history of treatment resistance, defined by persistence of severe symptoms despite adequate trials of one of the proposed treatments or prior treatment with clozapine for treatment resistance; were pregnant or breastfeeding; had had a myocardial infarction in the previous 6 months; had a history of or a current QTc prolongation; had uncompensated congestive heart failure; had sustained cardiac arrhythmia, a first-degree heart block, or complete left bundle branch block; or had another serious and unstable medical condition.
That’s a lot of people. Anybody newly diagnosed? Nope, don’t want you. Don’t respond to past treatments? Nope, don’t want you either. Have other mental disorder diagnoses? Generally, we don’t want you. This is not a representative sample. It is what we call a “biased sample.” What influence the bias has in the final results, nobody can say. Except to say that if you start with dirty data, you’re results are going to also be dirty (e.g., biased). The researchers’ sampling methods resulted in 17% of the people screened not being a part of the study for these reasons. That 17% could’ve completely changed the findings of the study (which the researchers acknowledge, “The third limitation is that participants in the CATIE project may not be representative of all persons with schizophrenia”).
A whopping 36% of the study participants had a substance abuse issue. More about this in a moment.
The researchers had baseline violence data on just 42% of their subjects.
Swanson and his colleagues conveniently “redefined” how the assessment measure they used, the MacArthur Community Violence Interview, describes violence. The Interview uses two categories — “violence” and “other aggressive acts.” This is an important distinction, because the researchers who developed the Interview had a clear theoretical construct they pursued and put into an objective interview format. Swanson and his colleagues redefined these two categories to reflect “severe violence” and “minor violence.”
Simple Semantics? Not so, because “other aggressive acts” was a category specifically designed to exclude the concept of “violence” by the original MacArthur researchers. By changing this wording, Swanson and his colleagues could arrive at the astounding 19.2% figure they find (and emphasize) by the end of the study. But this is smoke and mirrors — the 19.2% is an inaccurate representation of the researchers’ own data.
Now, keep in mind that the vast majority of Swanson’s violence indicators come from the patients themselves, without family corroboration — in other words, self-report. How reliable is self-report amongst people with schizophrenia?
It’s also not clear that the study differentiated domestic or family violence from other violence (something most other studies into violence prevalence and incidence generally do). This is an important differentiator, since it has significant policy and public health implications. Should we be increasing domestic and family violence monitoring, treatment and support for this population of people, or should we be cutting off all access to a Constitutional right for all people in this population?
Coming back to the substance abuse question and the sampling issues, the researchers write:
Additional analyses revealed that the sex effect in the final model was influenced by a subgroup of younger women with substance abuse problems and history of arrest. Women in the sample were also more likely to live with family, thereby presumably having more opportunities for physical fights with social network members.
As the researchers themselves note, a small group of people in a sample can create significant results in the data. We’ve long known that people who substance abuse and alcohol abuse issues tend to be more violent than the general population, but this has little to do with traditional mental disorders. (Although classified as such, they are often treated in different facilities by different professionals with specific training.) And the fact they attribute these findings to basically access to family members in order to perpetrate their violence is telling.
At the end of the study, what they did find is that 3.6% of their biased sample self-reported violence (not simply “aggressive acts”). Some have then compared this number to the 30-year-old data gathered from 1980-1985 from the NIMH’s Epidemiologic Catchment Area study, suggesting that data demonstrated a 2% incidence of violence in people without mental disorders. A lot could change in 30 years in terms of incidence of anything in the general population, but we don’t know if that’s still a valid number 30 years later (I’d suggest it’s not). More importantly, the criteria for measuring what is “violence” is different between the two studies — it’s like comparing apples to oranges. You can do it, but it’s not a valid comparison. Just because an anonymous person from the NIMH’s press office did it doesn’t make it valid.
So there you have it. A difference of 1.6% between a biased-sample study and that of 30-year-old data. Significant? Hard to say. I think Swanson and colleague’s words sum it up quite nicely:
Nonclinical variables, such as family coresidence, may affect violence risk in complex ways, either preventing or provoking violent behavior, depending on whether the family environment serves as a protective matrix or an opportunity for aggressive interactions. Consistent with some previous reports, our study presents a complex picture of the linkage between violence, social contact, and social support.
The words of Paul S. Appelbaum (2006) are also appropriate to consider:
The relationship between mental disorders and violence is complex. Among the variables that have been identified as increasing the risk of violence, in addition to psychotic symptoms and substance abuse, are socioeconomic status and even the neighborhoods in which persons with mental disorders reside. No single approach to reducing the risk is likely to be completely effective. And given the relatively modest contribution to the overall risk of violence by persons with mental disorders, the likelihood and magnitude of adverse effects from any intervention must be carefully considered before it is embodied in law.
Appelbaum, P.S. (2006). Violence and mental disorders: Data and public policy. Am J Psychiatry 163:1319-1321.
National Institute of Mental Health (1985). Epidemiologic Catchment Area Study, 1980-1985.
Swanson, J.W.; Swartz, M.S.; Van Dorn, R.A.; Elbogen, E.B; Wagner, H.R.; Rosenheck, R.A.; Stroup, T.S.; McEvoy, J.P. & Lieberman, J.A. (2006). A National Study of Violent Behavior in Persons With Schizophrenia. Arch Gen Psychiatry, 63:490-499.
Grohol, J. (2007). Violence and Mental Illness: Simplifying Complex Data Relationships. Psych Central. Retrieved on September 2, 2015, from http://psychcentral.com/blog/archives/2007/05/02/violence-and-mental-illness-simplifying-complex-data-relationships/