Are neuroanatomical phenotypes for psychiatric disorders robust? An assessment of the reproducibility of grey matter differences in mental illness

BackgroundDespite thousands of magnetic resonance imaging (MRI) studies reporting grey matter alterations in psychiatric disorders, the field has failed to converge on robust neuroanatomical phenotypes for any specific diagnosis. Here, we examine whether current practices will ever converge on such a phenotype, which is essential for tracking illness risk, progression, and treatment. MethodsWe evaluated the consistency of brain-wide maps of grey matter volume and cortical thickness alterations obtained for each of 59 study sites of five psychiatric disorders (schizophrenia, schizoaffective disorder, autism spectrum disorder, major depressive disorder, and bipolar disorder), totalling 2437 patients and 2065 controls. We calculated cross-site consistency using spatial correlations between pairs of site-specific difference maps and benchmarked the findings against 7 study sites of Alzheimers disease (654 patients, 937 controls). FindingsDisorder-specific volume and thickness alterations showed low consistency, with a median pair-wise cross-site correlation of [≤] 0.16 for psychiatric disorders compared to = 0.54 in Alzheimers disease. Consistency estimates were not strongly associated with site-specific variations in 19 different demographic, clinical, and scanner characteristics of the study participants and were robust to data processing and analysis. Bootstrapping analyses indicated that consistent results ( > 0.5) could be obtained for schizophrenia if study-specific sample sizes exceed approximately 200 (for cases and controls), but consistent findings for other disorders may require much larger samples. InterpretationOur findings indicate that current widespread practices, involving case-control comparisons of convenience samples numbering between 30 and 100 patients, will not converge on robust neuroanatomical phenotypes for psychiatric disorders. Increasing sample size will facilitate this goal in schizophrenia, but much larger samples, or refined ascertainment strategies aiming to recruit phenotypically homogeneous patient subgroups, may be required for other disorders.