Association of Adverse Outcomes With Emotion Processing and Its Neural Substrate in Individuals at Clinical High Risk for Psychosis

This case-control study analyzes emotion recognition and neuroimaging data as well as clinical and functional outcomes for individuals at risk for transition to psychosis and those without psychiatric or neurological disorders.


eMethods. MRI Data Acquisition and Preprocessing; Image Quality Assessment;
Assessment of Inter-Rater Reliability (IRR)

MRI Data Acquisition and Preprocessing.
For all participants, a three-dimensional T1-weighted inversion recovery prepared gradient echo sequence was obtained at 3T with voxel size: 1 x 1 x 1.2mm. Participants were recruited at 9 different clinical sites: Amsterdam, Basel, Cologne, Copenhagen, London, Melbourne, Paris, The Hague, Vienna.
Amsterdam and The Hague participants were scanned at the same site in Amsterdam; given that this site underwent a scanner change halfway through the project, we modelled "scanner" in all behavioral and imaging analyses instead of "site" (eTable 5).
Structural images were preprocessed using the Voxel-Based Morphometry protocol 1

Image Quality Assessment.
Image quality assessment (QA) was performed for all of the structural images. This involved careful visual inspection of all structural images by an experienced neuroimaging researcher (MK), and resulted in the exclusion of 3 participants (1 with large congenital cyst, 1 with distortion from brace, one with movement artefact). We also applied CAT12 to the structural MRI images of the 265 study participants as additional QA measure (eFigure 2). This analysis revealed two volumes of with a relatively low quality rating; these were carefully checked again and no artefacts were identified.

Assessment of Inter-Rater Reliability (IRR).
Krippendorff's α were calculated across the raters of the different EU-GEI sites to generate a measure of IRR. An IRR of > 0.7 was deemed acceptable. In order to become an EU-GEI rater, researchers had to pass the online training course, which entailed rating GAF and CAARMS training videos. After the initial training, new IRR videos appeared online and had to be scored at least once a year. These videos and vignettes had to be rated by all researchers recruiting participants for the EU-GEI project. Analysis of IRR across raters across the EU-GEI sites are shown in eTable 2. eResults. BFRT Results; Global Functioning Change Over Time; Sensitivity Analyses; Potential Confounders; Analysis of Normality for DFAR-GMV Interaction Data

Benton Facial Recognition Test
The BFRT short form was used, 2 in which a target face is presented centered above six stimulus faces.
In the first six trials, only one of the six stimulus faces is identical to the target face. In the following seven trials, three of the stimulus faces match the target. The test involves a total of 13 trials requiring a total of 27 responses (scoring ranges from 0-27). Scores of 21 or above are interpreted as being well within the normal range, and scores of 16 or below are interpreted as being impaired. Out of the 265 participants included in the study, only one participant scored below 16 (score = 15). BFRT scores were included as covariates of no interest in all analyses, as reported in the manuscript (group differences in DFAR performance, group differences in DFAR-GMV interactions).
Additional analyses removing the participant with a score below 16 from the tests of group differences in DFAR performance revealed that all results remained unchanged (eTable 3). Similarly, additional analyses with this participant removed from the tests of group differences between HC and CHR in DFAR-GMV associations did not change the results (eFigure 1). The DFAR anger x GMV interaction in the MPFC remained significant (xyz=0, 60, 18, Z=3.81; pFWE=0.03), as did the group x DFAR happy x GMV interaction in the left MPFC (xyz=-12, 54, 0; Z=3.98; pFWE=0.03), and the lack of group effects for interactions with neutral or fearful emotion. Removing this participant from the tests of group differences between CHR-GO and CHR-PO in DFAR-GMV associations did not change the results either: the left hippocampal finding for DFAR anger remained unchanged (xyz=−32, −40, −3; Z=3.79; pFWE=0.02), as did the left MPFC association with DFAR fear (xyz=−12, 38, -9; Z=3.69; pFWE=0.049).
Other group interactions with neutral or happy emotion remained non-significant. Finally, the lack of significant group x DFAR x GMV interactions based on transition vs nontransition outcomes also remained unchanged after removing the participant with a low BFRT score.

Analysis of global functioning change over time
We calculated GAF change over time as the difference between scores at the baseline and follow-up scores (ΔGAF = GAFfollow-up -GAFbaseline) for those CHR participants in whom follow-up GAF ratings were available (n=130, as per main manuscript page 12). The mean [SD] ΔGAF was 4.20 [17.02].

DFAR performance
Partial correlation was performed in SPSS to examine associations between DFAR performance (angry, happy, fearful, and neutral) and GAF change, adjusted for age, sex, IQ, site and BFRT score. This analysis revealed a significant positive correlation (Bonferroni-corrected at p=0.05/4=0.01) between DFAR fear and GAF change (r=0.309, p=0.001), indicating that the better recognition of fear at baseline, the greater the improvement in the GAF score over the follow-up period. There were no other significant correlations (DFAR neutral: r=0.061, p=0.52; DFAR happy: r=0.010, p=0.92; DFAR angry: r=0.073, p=0.44) (eFigure 3).

DFAR-GMV associations
Complementary analyses tested whether regions showing between-group differences in the CHR sample versus healthy controls analysis (MPFC -DFAR happy and MPFC -DFAR Anger) were related to longitudinal changes in GAF score. Individual values from the significant clusters in were extracted from SPM and Pearson's product-moment correlation analyses were performed in SPSS with GAF change scores. This analysis revealed no significant associations with GAF change for either MPFC -DFAR happy (r=0.010, p=0.91) or MPFC -DFAR anger (r=0.103, p=0.24).

DFAR performance
Re-analysis of DFAR performance excluding the sites that did not contribute HC participants did not change the results, which remained non-significant (eTable 7).

DFAR performance
Re-analysis of DFAR performance excluding the site that contributed only one participant to the CHR-GO vs CHR-PO analysis did not change the results: anger recognition at baseline was significantly associated with the level of functioning at 12 months follow-up (p=0.03; eTable 8).

Facial Emotional Processing
The functional outcome results remained unchanged after adjusting for baseline prodromal symptoms, baseline GAF scores, or transition/nontransition outcomes (eTable 9).

Analysis of normality for DFAR-GMV interaction data
We examined whether normality could be assumed for the original DFAR-GMV differences analysis.