Two years ago, a study used fMRI technology to come to a startling conclusion—there might be four biological subtypes of depression, called depression biotypes. This would herald a new age of depression research—one where mental health concerns could finally be linked to neurobiology. Now, a new study has attempted to replicate those findings, but their results cast doubt on whether such “biotypes” exist. According to the researchers, no statistically significant differences were found.
“Extending the original methods with additional more rigorous statistical procedures,” the researchers write, “we did not ﬁnd stable or statistically signiﬁcant biotypes of depression and anxiety in an independent sample from a diﬀerent clinical population.”
The research was led by Richard Dinga at the Department of Psychiatry, University Medical Center, Amsterdam. The researchers attempted to adhere to the exact methodology of the previous study and even contacted the former research team to discuss how to follow their methods precisely.
The 187 participants in this study had a diagnosis of either major depressive disorder or an anxiety disorder or both. They had no other psychiatric diagnosis and were not currently taking any antidepressant medications. Resting-state fMRI brain scans were conducted and then correlated.
Dinga and the other researchers write that the specific type of correlations done by the previous research team could have led to false positives—that is, those statistics were overly likely to lead to positive results. The researchers for the current study used a slightly more conservative method to compensate for this.
The researchers found that although they replicated the correlations of the original paper, none of the findings were statistically significant. These findings were just as likely to be due to chance. Dinga and the others note that the original study did not bother to test the statistical significance of their correlations: “It remains to be confirmed whether the canonical variates identified in their original study were significant.”
Also, Dinga and the others tested whether the clustering of biological data might be due to chance. They found that it very well might have been. According to the researchers, clustering analyses “always yield clusters, regardless of the structure of the data, even if there are no clusters at all.”
The researchers write, though, that the previous study did not analyze this possibility: “In the original study, this was not tested. Therefore, we cannot say if the data in the original study really formed clusters, instead of just random ﬂuctuation of the data.”
Dinga and the others note many problems with clustering based on biology—even if the clusters do exist. For instance, supposed biological subtypes might not be based on mental health symptoms, but instead, be based on other physical traits. “The variability of biological data is more often than not unrelated to any speciﬁc psychiatric disorder or symptom class.” It might instead be based on random variations such as “groups of people with similar brain size or body type or common ancestry in the case of genetics.”
Interestingly, in the original study, the researchers admit that there was no way to clinically differentiate the subtypes they found with fMRI data.
“Due to the clinical heterogeneity of many psychiatric disorders and the quest for personalized medicine, there is a tendency towards subtyping and expanding psychiatric nosology,” Dinga and the other researchers write.
“However, the presumption of distinct and homogeneous subtypes might not be clinically useful and might not represent the underlying biology. Many clustering approaches will always produce some clusters and would so even for uniformly distributed data. It is, therefore, crucial to distinguish real biologically or clinically meaningful subtypes from random ﬂuctuation of the data.”
Dinga, R., Schmaal, L., Penninx, B. W. J. H., van Tol, M. J., Veltman, D. J., van Velzen, L. . . . & Marquand, A. F. (2019). Evaluating the evidence for biotypes of depression: Methodological replication and extension of Drysdale et al. (2017). NeuroImage: Clinical. https://doi.org/10.1016/j.nicl.2019.101796 (Link)