Non-linear Impact of Symptom Severity on Effect Size of Neuro-Symptom Variations in Psychosis
Poster No:
481
Submission Type:
Abstract Submission
Authors:
Kangjoo Lee1, Clara Fonteneau1, Ally Price1, Lucie Berkovitch1, Jie Lisa Ji1, Zailyn Tamayo1, Yvette Afriyie-Agyemang1,2, Amber Howell1, Grega Repovš3, John D. Murray1,4, Youngsun Cho1, Alan Anticevic1
Institutions:
1Yale University School of Medicine, New Haven, CT, 2University of Pittsburgh, Pittsburgh PA, 3University of Ljubljana, Ljubljana, Slovenia, 4Dartmouth University, Hanover, NH
First Author:
Co-Author(s):
Yvette Afriyie-Agyemang
Yale University School of Medicine|University of Pittsburgh
New Haven, CT|Pittsburgh PA
Yale University School of Medicine|University of Pittsburgh
New Haven, CT|Pittsburgh PA
Introduction:
The identification of neural circuit abnormalities associated with psychiatric symptoms is crucial for the development of neural markers, where the reproducibility of neural-to-symptom mapping plays a key issue (Greicius 2008). While large sample sizes are critical for discovering and replicating brain-behavior associations with small effect sizes (Marek 2022), strong brain-behavior associations can be found using dense sampling in a small number of subjects exhibiting severe symptoms (Siegel 2023; Lynch 2023). It remains unclear whether the characteristics of studied samples, such as symptom severity, and sample size have a unique or additive impact on the reproducibility of neuro-symptom relationship across psychiatric spectra.
Methods:
We analyzed resting-state functional magnetic resonance imaging (rs-fMRI) and 32 symptom variables from 1,218 young individuals (8-21 y/o) from the Philadelphia Neurodevelopmental Cohort dataset (Fig 1). Among them, 350 subjects were identified as the psychosis spectrum group using raw scales (Calkins 2015). Principal component analysis (PCA) of clinical variables allowed the identification of five reproducible data-driven symptom PCs (Fig. 1) and the corresponding individual PC scores. Varying the sample size ratio (r=a/b, a: psychosis, b: non-psychosis), the pooled standard deviation (SD) of combined sample (a+b) was estimated (Fig 2). For a given ratio (r), we selected a random number (a) of subjects from the psychosis group and a random set of subjects (b=a/r) from the non-psychosis group across 1,000 permutations. Next, to study the impact of symptom severity on effect sizes in neural data (Fig. 2), we computed Pearson's correlation between the rs-fMRI time-course in each parcel and time-courses in all the other parcels of the brain, which were then averaged and Fisher's Z-transformed to estimate global brain connectivity (GBC) for each parcel (Cole-Anticevic Brain Network Parcellation; Ji 2021). For a given ratio, Glass's δ was computed for each parcel per permutation, using individual GBC values from the psychosis versus non-psychosis groups, and then averaged across 1,000 permutations. Finally, we computed how much the effect sizes estimated from the true group comparisons (δ_test) deviates from the effect sizes estimated from the null group comparisons (δ_null).
Results:
Five symptom PCs exhibiting prominent symptom variances were reproducible, representing global psychopathology, cognitive, negative symptoms, lifetime delusion, and lifetime hallucination, as demonstrated by the loading profiles and their correlation to the raw symptom scores (Fig. 1). The distribution of individual PC scores, except PC2, from the psychosis group exhibited a large variance compared to that of the non-psychosis group (Fig. 2A). A robust estimation of pooled SD based on a random subsampling (random 350 subjects from the non-psychosis group to match the number of subjects in the psychosis group) confirmed that this observation is independent of the sample size (gray lines, 1,000 permutations). The sample ratio (r) impacts the pooled SD of symptom PC scores (Fig. 2B), suggesting a combined effect of data quality and sample size on behavioral effects. Using neural data, our data shows a nonlinear curve of effect size deviation as a function of sample size ratio, suggesting a plateau effect.
·Figure 1.
·Figure 2.
Conclusions:
We identified reproducible, data-driven clinical dimensions that have different contributions in psychosis versus non-psychosis groups. Symptom severity combined with sample size ratio exhibited an reproducible effect of psychiatric illnesses in both behavioral and neural data. Testing the hypothesis that neuro-symptom reproducibility plateaus under specific conditions combining symptom severity and sample size may allow us to derive a data-driven estimation of multivariate thresholds that meets the criteria of reproducibility.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 1
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling 2
Task-Independent and Resting-State Analysis
Keywords:
FUNCTIONAL MRI
Psychiatric
Psychiatric Disorders
Schizophrenia
1|2Indicates the priority used for review
Provide references using author date format
Greicius M. (2008), Resting-state functional connectivity in neuropsychiatric disorders. Current Opinion Neurology. 21(4):424-30. doi: 10.1097/WCO.0b013e328306f2c5.
Ji JL et al. (2021), Mapping brain-behavior space relationships along the psychosis spectrum. Elife. 10:e66968. doi: 10.7554/eLife.66968. Erratum in: Elife. 2022 Apr 05;11
Lynch CJ et al. (2023), Expansion of a frontostriatal salience network in individuals with depression. bioRxiv. 14:2023.08.09.551651. doi: 10.1101/2023.08.09.551651.
Marek S et al. (2022), Reproducible brain-wide association studies require thousands of individuals. Nature. 603(7902):654-660. doi: 10.1038/s41586-022-04492-9. Epub 2022 Mar 16. Erratum in: Nature. 2022 May;605(7911):E11.
Siegel JS et al. (2023), Psilocybin desynchronizes brain networks. medRxiv. 24:2023.08.22.23294131. doi: 10.1101/2023.08.22.23294131.