Importance of statistical methods for assessing performance and moderator effects in neuroimaging-based classification models

In our recent publication, we report a meta-analysis of the diagnostic performance of neuroimaging-based classification models for the differentiation of patients with a depressive disorder from healthy control individuals https://paperpile.com/c/AGY9NR/nGdv(1). In summary, our results indicate that across studies patients can be identified with an estimated accuracy of 77 %. Moderator analysis provided some evidence for effects of moderating factors including neuroimaging modality. However, despite theoretical arguments and similar findings in comparable analyses in schizophrenia https://paperpile.com/c/AGY9NR/C57a+6hON(2, 3), we did not find evidence for an effect of sample size on classification accuracy.

Source: Biological Psychiatry - February 7, 2018 Category: Psychiatry Authors: Joseph Kambeitz, Carlos Cabral, Matthew D. Sacchet, Ian H. Gotlib, Roland Zahn, Mauricio H. Serpa, Martin Walter, Peter Falkai, Nikolaos Koutsouleris Tags: Correspondence Source Type: research