Mostly normal, but also different: Individualised predictions from normative modelling in depression
Poster No:
1980
Submission Type:
Abstract Submission
Authors:
Johanna Bayer1,2,3, Laura van Velzen2, Elena Pozzi4, Christopher Davey5, Laura Han3, Paul Thompson6, Dick Veltman7, Andre Marquand8, Lianne Schmaal9
Institutions:
1Donders Institute for Brain, Cognition and Behaviour, Nijmegen, Gelderland, 2Orygen, Melbourne, Australia, 3The University of Melbourne, Melbourne, Australia, 4Orygen, Parkville, Australia, 5University of Melbourne, Carlton, Vic, 6USC, Marina Del Rey, CA, 7Amsterdam UMC location Vrije Universiteit Amsterdam, Department of Psychiatry, Amsterdam, North Netherlands, 8Radboud University Nijmegen, Nijmegen, Gelderland, 9Centre for Youth Mental Health, The University of Melbourne, Parkville, VIC, Australia., Melbourne, Australia
First Author:
Johanna Bayer
Donders Institute for Brain, Cognition and Behaviour|Orygen|The University of Melbourne
Nijmegen, Gelderland|Melbourne, Australia|Melbourne, Australia
Donders Institute for Brain, Cognition and Behaviour|Orygen|The University of Melbourne
Nijmegen, Gelderland|Melbourne, Australia|Melbourne, Australia
Co-Author(s):
Dick Veltman
Amsterdam UMC location Vrije Universiteit Amsterdam, Department of Psychiatry
Amsterdam, North Netherlands
Amsterdam UMC location Vrije Universiteit Amsterdam, Department of Psychiatry
Amsterdam, North Netherlands
Lianne Schmaal
Centre for Youth Mental Health, The University of Melbourne, Parkville, VIC, Australia.
Melbourne, Australia
Centre for Youth Mental Health, The University of Melbourne, Parkville, VIC, Australia.
Melbourne, Australia
Introduction:
Major Depressive Disorder (MDD) is a complex, multi-factorial mental illness, impacting more than 320 mio. people world-wide1. The search for neuroimaging biomarkers in MDD has been impacted by unreliability, induced by small sample sizes2,3, site effects in pooled neuroimaging data4-6, complex interactions due to brain development, and by the heterogeneity7-9 of the disorder, paired with a group average comparison approach10. We use a normative modelling algorithm that allows to deal with site effects11 to calculate z-scores of deviations of cortical thickness from the norm and to make individualized predictions in MDD. Our analysis is driven by two main aims: 1.) To uncover the heterogeneity in cortical thickness measures in MDD along a spectrum of deviations from the norm, stepping beyond group average comparisons. 2.) to investigate what makes individuals with extreme deviations different from individuals falling within the norm.
Methods:
We trained a normative model on 35 bilateral cortical thickness (CT) measures derived from Freesurfer parcellation (Desikan-Killiany atlas) of 3181 healthy individuals and tested the model on 3645 individuals with MDD from the ENIGMA MDD consortium12. This allowed us to get an estimate of region wise cortical thickness deviations of individuals with MDD under a healthy control model. To test out-of-sample generalizability, we also made predictions from a test set of 2119 healthy individuals. Beyond calculating group z-score distributions per region, we investigated into individuals with extreme deviations from the norm: Individuals with a z < -1.96 were marked to have an infra-normal score, those with z > 1.96 were marked to have a supra-normal score. We summarized those extreme deviations by two summary scores that target extreme deviations: A load score was calculated to summarize the number of regions with an infra- or supra-normal scores across all regions per individual, a extremity score summarized the most extreme deviation across all regions per individual13.
Results:
At group level, we found an overall trend towards decreased CT in z-score distributions of individuals with MDD for most brain regions. Those differences were in location and magnitude of previously reported effect sizes (Cohen's d: 0.01 - 0.1, overlap of distribution of z-scores: 95% 14,15, Fig 1). Similarly, we found that ~70% of individuals showed z-scores within the norm for all brain regions.
For those 30% of individuals with depression that showed an extreme deviation in at least one brain region, we found widespread spatial heterogeneity of that extreme deviation (see Fig. 2). The maximum percentage of individuals that showed an infra-normal z-score in the same region was 11.8% (fusiform gyrus). For supra-normal z-scores, this number amounted to 12.2 % (both pericalcarine gyrus and cuneus).
Quantifying the degree of extreme deviations in MDD by calculating load and severity scores, we found that extreme negative deviations from the norm of CT were a risk factor for remission status, number of depressed episodes, anti-depressant use patterns and age off onset of depression. Extreme positive deviations, in turn, were a protective factor and moreover, negatively associated with symptom severity and positively associated with a more favourable outcome regarding remission status and anti-depressive use patterns.
For those 30% of individuals with depression that showed an extreme deviation in at least one brain region, we found widespread spatial heterogeneity of that extreme deviation (see Fig. 2). The maximum percentage of individuals that showed an infra-normal z-score in the same region was 11.8% (fusiform gyrus). For supra-normal z-scores, this number amounted to 12.2 % (both pericalcarine gyrus and cuneus).
Quantifying the degree of extreme deviations in MDD by calculating load and severity scores, we found that extreme negative deviations from the norm of CT were a risk factor for remission status, number of depressed episodes, anti-depressant use patterns and age off onset of depression. Extreme positive deviations, in turn, were a protective factor and moreover, negatively associated with symptom severity and positively associated with a more favourable outcome regarding remission status and anti-depressive use patterns.
Conclusions:
This study shows the widespread heterogeneity of extreme z-score deviations from the norm of CT in MDD that lies below previously reported thinner cortices in MDD at group level. While 70% of individuals diagnosed with MDD no extreme z-scores deviation across all brain region, the number and overall degree of extreme deviations could be used to make clinically useful predictions. This study also shows the potential of individualized z-scores from normative modelling to inform clinical predictions beyond group averages and how normative modelling may lead to additional insights in heterogeneous disorders such as MDD.
Lifespan Development:
Lifespan Development Other 2
Modeling and Analysis Methods:
Bayesian Modeling
Other Methods 1
Novel Imaging Acquisition Methods:
Anatomical MRI
Keywords:
Cortex
DISORDERS
Modeling
NORMAL HUMAN
Statistical Methods
STRUCTURAL MRI
Other - normative modelling
1|2Indicates the priority used for review
Provide references using author date format
2.Schnack, H. G. & Kahn, R. S. Detecting neuroimaging biomarkers for psychiatric disorders: Sample size matters. Front. Psychiatry 7, 50 (2016-3).
3.Young, J. J. et al. Is there Progress? An Overview of Selecting Biomarker Candidates for Major Depressive Disorder. Front. Psychiatry 7, 1–15 (2016).
4.Chen, J. et al. Exploration of scanning effects in multi-site structural MRI studies. J. Neurosci. Methods 23, 37–50 (2014).
5.Glocker, B., Robinson, R., Castro, D. C., Dou, Q. & Konukoglu, E. Machine Learning with Multi-Site Imaging Data: An Empirical Study on the Impact of Scanner Effects. arXiv [eess.IV] 1–5 (2019).
6.Gronenschild, E. H. B. M. et al. The Effects of FreeSurfer Version, Workstation Type, and Macintosh Operating System Version on Anatomical Volume and Cortical Thickness Measurements. PLoS One 7, e38234 (2012).
7.Goldberg, D. The heterogeneity of “major depression”. World Psychiatry 10, 226–228 (2011).
8.Fried, E. I. & Nesse, R. M. Depression is not a consistent syndrome: An investigation of unique symptom patterns in the STAR*D study. J. Affect. Disord. 172, 96–102 (2015-2).
9.Fried, E. I. Moving forward: how depression heterogeneity hinders progress in treatment and research. Expert Rev. Neurother. 17, 423–425 (2017).
10.Marquand, A. F., et al. Beyond Lumping and Splitting: A Review of Computational Approaches for Stratifying Psychiatric Disorders. Biological Psychiatry: Cognitive Neuroscience and Neuroimaging 1, 433–447 (2016).
11.Bayer, J. M. M. et al. Accommodating site variation in neuroimaging data using normative and hierarchical Bayesian models. Neuroimage 264, 119699 (2022).
12.Thompson, P. M. et al. ENIGMA and global neuroscience: A decade of large-scale studies of the brain in health and disease across more than 40 countries. Transl. Psychiatry 10, 100 (2020).
13.Elad, D. et al. Improving the predictive potential of diffusion MRI in schizophrenia using normative models—Towards subject-level classification. Hum. Brain Mapp. 42, 4658–4670 (2021).
14.Schmaal, L. et al. Cortical abnormalities in adults and adolescents with major depression based on brain scans from 20 cohorts worldwide in the ENIGMA MDD Working Group. Mol. Psychiatry 22, 900–909 (2017).
15.Winter, N. R. et al. Quantifying Deviations of Brain Structure and Function in Major Depressive Disorder Across Neuroimaging Modalities. JAMA Psychiatry 79, 879-888 (2022).