Get started
Work

Differentiating patients with bipolar disorders and depressive disorders using self-report screening questionnaires and anger coping scale

Year: 2022
Type: article
Study design: Observational Study
Abstract: specific disease signatures and improve diagnostic processes [2].This project aims to implement a novel data-driven clustering of subtypes of depressed patients using structural neuroimaging data.Another concomitant objective is to investigate whether the identified clusters are linked to specific clinical profiles.Methods: T1-weighted and diffusion tensor images were acquired from a sample of 112 major depressive (MDD) and 205 bipolar disorder (BD) patients.In a subsample of 197 patients (69 MDD, 128 BD), depressive symptomatology was evaluated using the Beck Depression Inventory-Short Form (BDI-SF) [3] and Hamilton Depression Rating Scale (HDRS-21) [4].Regional measures of cortical and subcortical surface area and volumes (CAT12, FreeSurfer), and extracted tract-based fractional anisotropy (FA) white matter values (TBSS, FSL) were used as input features for clustering.Specifically, three different combination of features were tested: 1) volume-based; 2) surface-based; 3) volume-based, surfacebased and FA measures combined.We compared the clustering solutions obtained by kmeans with Euclidean distance and HDBSCAN with uniform manifold approximation and projection (UMAP) for dimensionality reduction.For each clustering algorithm, clusters' labels were used within Support vector machine (SVM) and elastic net logistic regression to learn how to best predict the labels.A parallel grid search was implemented for hyperparameter tuning.A stabilitybased clustering algorithm with 10x2 repeated cross-validation was run with reval python package [5], with a number of clusters varying from 2 to 5 for kmeans.Normalized stability (i.e., prediction error), silhouette and Davis-Bouldin scores were computed to assess the best clustering-classifier combination.GM measures were adjusted for total intracranial volume.The clinical homogeneity of the best clustering solution was assessed through MANOVA, entering total scores and domains of BDI-SF (cognitive, behavioral, affective) and HDRS-21 (anxiety, depression, insomnia, somatic) as dependent variables and clusters' labels as fixed factors.The proportion of BD and MDD patients among clusters was explored though Chi-Square test.Results: kmeans achieved more stable clustering solutions compared to HDBSCAN (normalized stability: 0.36-0.51for HDSBCAN; 0.10-0.17for kmeans).Kmeans and elastic net based on GM and FA measures resulted in the best clustering-classifier combination, identifying 2 clusters with 0.10 normalized stability (silhouette score¼0.16,Davis-Bouldin score¼1.99).The features that best discriminated the clusters were the right inferior frontal gyrus, temporal-occipital surface, posterior cerebellum, insula, and left anterior cingulate cortex.A significant between-clusters difference in BDI-SF and HDRS-21 scores was found (F(9,187)¼1.75,p¼0.026), with a significant effect of clusters' labels on BDI-SF cognitive dimension (F(1,195)¼54.84,p¼0.029).No differences in MDD and BD proportions were found between clusters (χ2(1, N¼197)¼2.17,p¼0.178). Conclusions:The stability-based cluster validation efficiently identified stable and reproducible subgroups of depressed patients based on neuroimaging data.Structural abnormalities in cortico-cerebellar connectivity were found to define patients with cognitive symptoms of depression.This approach has the potential to discover varying etiologies in heterogeneous psychiatric disorders in the field of precision psychiatry.
DOI: 10.1016/j.nsa.2022.100175
Language: English
Field: Medicine
Open Access status: gold
Estimated APC paid: $2,000
OpenAlex ID: W4313200430
FWCI:
Cites:
Cited by:
Related to: 10