Poster No:
568
Submission Type:
Abstract Submission
Authors:
Wenhao jiang1, Javier Romero2, Sandeep Panta2, Jay Fournier3, Jing Sui4, Vince Calhoun5, Jessica Turner6
Institutions:
1ZhongDa Hospital; School of Medicine, Southeast University, Nanjing, Jiangsu, 2Georgia State University, Atlanta, GA, 3The Ohio State University Wexner Medical Center, Columbus, OH, 4Beijing Normal University, Beijing, China, 5GSU/GATech/Emory, Decatur, GA, 6Ohio State University Wexner Medical Center, Columbus, OH
First Author:
Wenhao jiang
ZhongDa Hospital; School of Medicine, Southeast University
Nanjing, Jiangsu
Co-Author(s):
Jing Sui
Beijing Normal University
Beijing, China
Jessica Turner
Ohio State University Wexner Medical Center
Columbus, OH
Introduction:
Hippocampal volumes tend to be decreased in major depressive disorder (MDD); the ENIGMA MDD meta-analysis of subcortical volumes in 1728 individuals with MDD and over 7000 healthy volunteers found the mean hippocampal volume was reduced in recurrent rather than first episode MDD, but did not show a relationship with depression severity. They did not find effects of severity using the subset which had that data, and they examined the average volumes only. We aimed to replicate those findings in an international, independent sample, examining both left and right hippocampal volumes and with symptom scores.
In these analyses, we tested the abillty of a federated analysis platform, COINSTAC, do perform a decentralized regression across the individual sites. The goal of the COINSTAC platform is to bring the analysis to the data, rather than the data to the analysis. In this case, the volumes could be shared but the images could not; we tested the COINSTAC single-shot regression algorithm against a combined analysis. The use of the federated platform allows the potential for more complex analyses on these datasets in the future, to do decentralized machine learning and individual predictions.
Methods:
Data from twelve independent sites or studies were included in this analysis. The datasets were pre-existing and collected under IRB approval at each institution. All sites used 3T scanners and collected T1-weighted images with a fixed protocol for that site. The original images could not be aggregated at one location; each site processed their own data through Freesurfer 7.2, and performed visual QC on the segmentations. Left, right, and total hippocampal volume, as well as intracranial volume (ICV) were provided for each participant. Each site included age, sex, diagnosis (MDD or none), and Hamilton's Depression Scale (HAM-D) scores.
The total dataset comprised 931 subjects, of whom 602 with MDD, 324 were men, 607 were women. The age range was limited to 18 to 55 in all datasets, with an overall mean age of 31.8 years. The HAM-D scores ranged from 0 to 39 with a median value of 15.
The first combined analysis modeled each of left, right, and total hippocampal volumes against diagnosis status, age, sex, and ICV, with site as a random effect. The second analysis regressed HAMD scores for all subjects without considering diagnosis. Analyses that were not done in COINSTAC were done in R using the lme4 package.
The COINSTAC analysis did a "single-shot" regression of each of these analyses, which is an automatic, separate analysis on each site's data and a weighted average of the results.
Results:
The effects of MDD, sex, and ICV on the total hippocampal volume were significant. The effects of sex and ICV were as expected: women had smaller volumes, and a larger ICV correlated with a larger hippocampal volume. The effect of age in this sample was not significant. Those with MDD on average had a smaller volume (t(770.2) = -1.98, p< .048).
A similar pattern was found in the right hippocampus: the MDD group had a smaller volume (t(831)=-2.4, p=.017), as well as sex and ICV effects. The same analysis in the left hippocampus, however, found significant effects of sex and ICV only.
The HAM-D effect was significant only for the right hippocampus (t(856) =-2.28, p= 0.023). The federated analysis found comparable single-site results, though the single-shot combination of all sites was not as powerful as the combined analysis.
Conclusions:
We extended the original ENIGMA MDD analysis and found that the effects are primarily in the right hippocampal volume. We also found a relationship with symptom severity scores in the same structure. The federated analysis serves as a proof of principle, a foundation for future analyses including cross-site harmonization using ComBat or other techniques, as well as iterative approaches such as machine-learning analyses on the original, unshared images.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 1
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Anatomy and Functional Systems 2
Keywords:
Affective Disorders
Sub-Cortical
Other - Hippocampus
1|2Indicates the priority used for review
Provide references using author date format
None.