Schizophrenia, variability, and the Anna Karenina principle

Abstract

In neuroimaging studies of people with schizophrenia there is often higher within group variance in the patient group compared to the control group. This is counterintuitive - why would a subset of people selected because they all have the same disease be more varied than the general population? We used simulated data and real neuroimaging data to identify a potential cause of elevated variance in populations of patients with schizophrenia. We demonstrated that elevated variance can arise within variables that are unrelated to disease status simply because people with a set of neurological perturbations that cause schizophrenia are more likely to have higher numbers of perturbations overall. Additionally, we showed that observed elevated variances in people with schizophrenia can be reproduced by models that only rely on perturbation count. These results highlight an important barrier in our attempts to understand the pathophysiology of schizophrenia. Standard statistical practices in schizophrenia research do not account for the fact that schizophrenia is, at every level of analysis that has been studied, highly heterogeneous. This heterogeneity by itself is sufficient to produce elevated variances. Our work suggests that the most effective way to prevent schizophrenia may not be to identify and mitigate specific pathologies but rather to reduce the impact of broadly damaging factors such as those associated with poverty.