Can brain scans divide depression into distinct biological types?

Why do current psychiatric diagnoses fail to consistently align with underlying biological causes?

They are diagnosed exclusively based on clinical symptom presentation rather than biological mechanisms, leading to high heterogeneity and frequent comorbidities.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

What is the main objective of the Research Domain Criteria (RDoC) initiative?

To address psychiatric heterogeneity by evaluating multiple levels of analysis (such as biology, behavior, and symptoms) across domains of functioning, rather than relying strictly on diagnostic categories.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

Why might clustering psychiatric patients based on clinical symptoms alone fail to yield biological subtypes?

Clinical symptom groupings do not necessarily reflect or map directly onto the underlying biological differences causing those symptoms.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

What is a major pitfall of clustering psychiatric populations based on biological data alone?

It can capture dominant nuisance variances (like head size, ancestry, or body type) that are completely unrelated to psychiatric pathology.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

How did Drysdale et al. attempt to constrain biological clustering to clinically relevant dimensions?

They used Canonical Correlation Analysis (CCA) to map resting-state fMRI functional connectivity measures specifically to major depressive disorder (MDD) symptoms, using the resulting components for clustering.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

What mathematical relationship does Canonical Correlation Analysis (CCA) identify between two multidimensional datasets?

It finds linear combinations of variables from each dataset (canonical variates) that are maximally correlated with each other.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

What clinically impactful claim did Drysdale et al. make regarding their identified four biotypes of depression?

They claimed that these biotypes could predict patient responsiveness to transcranial magnetic stimulation (TMS) treatment.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

What was the main aim of the replication study conducted by Dinga et al.?

To apply Drysdale et al.'s analytical pipeline to an independent, more heterogeneous clinical cohort of depression and anxiety patients to test the external validity of the biotypes.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

Which two clinical cohorts were combined to form the sample for this replication study?

The Netherlands Study of Depression and Anxiety (NESDA) and the Mood Treatment with Antidepressants or Running (MOTAR) study.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

How did the replication sample's clinical profile differ from the original sample used by Drysdale et al.?

The original study used hospitalized, currently active treatment-resistant MDD patients, while the replication sample included a naturalistic, heterogeneous mix of active, remitted, and mild MDD and anxiety cases.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

Which clinical rating scale did the replication study use instead of the original study's HAMD?

The self-rated Inventory of Depressive Symptomatology (IDS), using a subset of 17 items matching the HAMD.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

What major difference in fMRI motion correction was introduced in the replication study's preprocessing pipeline?

The replication used ICA-AROMA instead of motion scrubbing (volume censoring) to preserve the temporal degrees of freedom in the fMRI time series.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

How did the replication study maintain the same subject-to-feature ratio during feature selection as the original study?

By retaining the top 150 RS-fMRI connectivity features (approx. 80% of the 187 participants) with the highest correlation to clinical symptoms.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

Why is CCA highly prone to overfitting when applied to typical neuroimaging datasets?

Because of a very high number of features compared to a relatively small sample size, leading to unstable models.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

Why were the analytical p-values (Wilk's lambda) for the CCA in Drysdale et al. overly optimistic?

The significance test did not account for the feature pre-selection step (selecting top correlated brain features in the same dataset), which artificially inflated the apparent correlation.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

How did the replication authors correct for feature pre-selection when testing the significance of CCA?

They used a permutation test of the entire procedure, repeating both feature selection and CCA 1,999 times with shuffled clinical data.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

What happened to the replication study's canonical correlations when evaluated out-of-sample via 10-fold cross-validation?

They dropped from very high within-sample values (0.99 and 0.97) to near zero (~0.07 and -0.07) in the test set.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

What does an out-of-sample canonical correlation of approximately 0 indicate about a CCA model?

It indicates that the high within-sample association was driven by overfitting and does not generalize to independent data.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

What resampling method did the authors use to test the stability of canonical loadings?

A delete-one jack-knife procedure (leaving one subject out of the entire pipeline and recalculating loadings).

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

How stable were the canonical loadings when removing just a single subject from the pipeline?

They were highly unstable, changing dramatically with the removal of only one subject.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

Which clustering parameters were used to identify subtypes from the canonical variates?

Hierarchical clustering using Euclidean distance and Ward's D linkage.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

What does the Calinski-Harabasz (CH) index measure in clustering analysis?

It is a variance ratio criterion that measures the ratio of between-cluster variance to within-cluster variance.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

What is measured by the Silhouette index in the context of clustering?

It compares average within-cluster distances to average distances between points in different clusters.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI

What is a fundamental pitfall of many standard clustering algorithms, such as hierarchical clustering?

They will always partition data and output clusters, even if the underlying data distribution is completely continuous and cluster-free.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

How did the authors statistically test whether the clustering in their data was real or artificial?

They simulated an empirical null distribution by repeatedly clustering data sampled from a single, cluster-free 2D Gaussian distribution.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

Did the replication study's 3-cluster solution show statistically significant clustering when compared to a Gaussian null?

No, the CH and silhouette indices were not statistically significant, meaning they are easily observed in data with no true underlying clusters.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

How stable was individual subject cluster assignment under the leave-one-out resampling evaluation?

The clusters and subject assignments changed substantially just by changing a single subject in the pipeline.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

Why does correcting for scanner-site effects potentially reduce statistical power in clinical fMRI studies?

Because if different sites also differ in their average patient symptom severity, regressing out site effects will inadvertently remove real disease-related connectivity variance.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI

Why is predicting treatment response (e.g., to TMS) using continuous biological axes preferred over using discrete biotypes?

Because treatment response probability likely varies smoothly across continuous biological dimensions, and arbitrary subtyping reduces statistical power and predictive accuracy.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

How does the 'borderline jump' issue highlight the clinical danger of arbitrary biotype classification?

Highly similar patients on either side of an arbitrary boundary can receive wildly different treatment predictions (e.g., 80% vs 20% chance of response).

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

Why can continuous biological axes be valuable in neuropsychiatry even if they do not form distinct clusters?

They provide a parsimonious dimensional representation of disease spectrums, capturing comorbidities and sub-threshold symptoms without arbitrary dichotomization.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

What is the recommended subject-to-variable ratio in classical literature to ensure a stable CCA model?

Between 10 and 42 subjects per variable.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI

How should future neuroimaging studies handle feature selection to avoid inflated significance in multivariate analyses?

By performing statistical tests strictly in an independent test set, or by fully incorporating the feature selection step into the permutation testing procedure.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified

Why are non-exact methodological replications highly informative in clinical neuroscience?

They test the generalizability and robustness of original findings to different scanner types, patient populations, and clinical measures.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI

What was the ultimate conclusion regarding the existence of the four biotypes of depression proposed by Drysdale et al.?

There is insufficient evidence to claim distinct biotypes exist, as the original clustering metrics do not rule out a single continuous distribution.

Source: Evaluating the evidence for biotypes of depression: Methodological replication and extension of · DOI · quote verified