Subjects
This study was approved by the Regional Ethics Board in Stockholm, Sweden (REPN: 2009/1089-31/2; EPM: 2022-00602-02). The sample was derived from the MULTI-PSYCH cohort, described in detail in(20). In summary, participants were recruited between 2009 and 2018 at the Internet Psychiatry Clinic in Stockholm, which specialises in providing ICBT nationwide as part of the public health care system. A total of 2668 individuals were included in this analysis (depression, n=1300, panic disorder (n=728), or social anxiety disorder, n=640). See Table 1 for participant characteristics. The sample had 88% of individuals with higher education, which is notably high compared to 27% of the Swedish general population in 2016(21).
Table 1. Sociodemographic and clinical characteristics at baseline and post-treatment
|
|
Depression
(n=1300)
|
Panic disorder
(n=728)
|
Social anxiety
(n=640)
|
Total
(n=2668)
|
Missing
|
Gender, female
|
859 (66)
|
435 (60)
|
360 (56)
|
1654 (62)
|
1 (0)
|
Age
|
37.6 (11.9)
|
34.6 (10.8)
|
32.7 (10.3)
|
35.6 (11.4)
|
|
In a relationship
|
731 (56)
|
478 (66)
|
363 (57)
|
1572 (59)
|
5 (0)
|
Children
|
603 (46)
|
301 (41)
|
198 (31)
|
1102 (41)
|
5 (0)
|
Highest education attained
|
|
|
|
|
5 (0)
|
Primary
|
25 (2)
|
18 (2)
|
14 (2)
|
57 (2)
|
|
Secondary
|
116 (9)
|
82 (11)
|
56 (9)
|
254 (10)
|
|
Higher
|
1158 (89)
|
626 (86)
|
569 (89)
|
2353 (88)
|
|
Prior inpatient care
|
91 (7)
|
74 (10)
|
27 (4)
|
192 (7)
|
115 (4)
|
Prior suicide attempt
|
76 (6)
|
26 (4)
|
34 (5)
|
136 (6)
|
228 (9)
|
Pre-treatment psychiatric comorbidity
|
395 (30)
|
274 (38)
|
212 (33)
|
881 (33)
|
108 (4)
|
Pre-treatment psychotropic medication
|
788 (61)
|
445 (61)
|
322 (50)
|
1555 (58)
|
454 (17)
|
Number of ICBT modules started
|
6.90 (3.2)
|
6.83 (2.9)
|
7.14 (3.0)
|
6.94 (3.1)
|
|
LSAS pre score
|
|
|
70.6 (23.5)
|
|
62 (10)
|
LSAS post score
|
|
|
50.1 (24.1)
|
|
118 (18)
|
MADRS-S pre score
|
22.7 (6.3)
|
|
|
|
14 (1)
|
MADRS-S post score
|
13.0 (8.0)
|
|
|
|
248 (19)
|
PDSS-SR pre score
|
|
11.0 (4.7)
|
|
|
16 (2.1)
|
PDSS-SR post score
|
|
4.98 (4.5)
|
|
|
158 (21.4)
|
Data are integer count (%) or decimal mean (SD). ICBT=Internet-delivered cognitive behaviour therapy, MADRS-S=Montgomery-Åsberg Depression Rating Scale Self-rated, PDSS-SR=Panic Disorder Severity Scale - Self Report, LSAS=Liebowitz Social Anxiety Scale
|
Study Design
Intervention
The ICBT treatment consisted of 10 treatment modules over 12 weeks for depression(22), panic disorder(23) and social anxiety disorder(24). All participants were treated at the Internet Psychiatry Clinic in Huddinge, Sweden, which specialises in ICBT. The participants did the treatment on a secure platform, including reading 5-20 pages of text per module, and completing weekly homework assignments. They also communicated with their therapist asynchronously through the platform. The format ensured the treatment content followed protocol and facilitated quality control through automatically distributed questionnaires. For more detail, see (20).
Primary outcome measure
Symptom severity was measured weekly from the beginning to end of treatment for the primary outcome using disorder-specific instruments: The Montgomery-Åsberg Depression Rating Scale Self-rated (MADRS-S, (25)) for depression, the Panic Disorder Severity Scale - Self Report (PDSS-SR, (26,27)) for panic disorder, and the Liebowitz Social Anxiety Scale (LSAS, (28))for social anxiety disorder. All instruments were self-rated and completed via the online platform. To harmonise symptom values across diagnoses, original disorder-specific values were scaled to a common metric (score range 0-100, see details in the Supplement).
Genotyping
Genotyping was done in three batches, on either Illumina HumanOmniExpress BeadChips (Illumina, USA) or Infinium Global Screening Array 1.0 BeadArray (Illumina, Inc., San Diego, CA, USA), at the Department of Genomics, Life and Brain Centre, University of Bonn, Germany. More details on genotyping and quality control steps are described in the Supplement.
Target dataset
Genotype array data from the target dataset were processed through the Ricopili pipeline v2019_10_15_001 (29). We first used Ricopili to do pre-imputation quality control on array genotypes across each of the three batches using default thresholds (see Supplement). As part of this, we controlled for cryptic relatedness between samples by removing samples where 1) they had a mean PI_HAT relatedness metric above 0.95, 2) they had evidence of being a duplicated sample based on PI_HAT > 0.95, or 3) they had evidence of cryptic relationship based on PI_HAT > 0.2. Genotype imputation was performed using Ricopili using impute2(30) for pre-phasing and minimac3(31) for imputation. To estimate ancestry, principal component analysis (PCA) was done in EIGENSOFT(32) and a value of >6 SD from the mean on any of the first three principal components was considered outlying and therefore removed. PRS were thereafter calculated with PRS-CS(33) from GWAS summary statistics and samples with European ancestry from the 1000 Genomes Project(34). PRS-CS is a Bayesian polygenic prediction method that has demonstrated better predictive accuracy than earlier methods(34). All genetic analyses were conducted using PLINK 1.9(35).
Discovery datasets
Discovery datasets from large GWASs were used to create individual-level aggregated PRS for each phenotype. The following discovery datasets were used: ADHD(36), autism spectrum disorder (ASD (37), bipolar disorder, (BPD, (38), major depressive disorder (MDD, (39)) and schizophrenia (SCZ, (40)), cross disorder PRS (41), educational attainment (EDU, (42)) and intelligence (IQ, (43)). The cross disorder PRS includes genetic effects of ADHD, ASD, BPD, MDD and SCZ. The target dataset was not part of these GWAS meta-analyses.
Statistical analysis
Statistical analysis was done using R(44). To estimate the association between PRS and symptom severity change during ICBT treatment, we used linear mixed effects models for repeated measures (lme4 package, (45)). To obtain p-values for the coefficients of interest we used the package lmerTest(46), which provides p-values via Satterthwaite’s degrees of freedom method. We fitted models that used both the linear and quadratic effects of time, which together provided a better fit for data compared to using either one individually, according to Akaike Information Criterion(47). We estimated the association of time and symptom ratings to investigate the rate of symptom severity change from pre- to post-treatment.
For the main analysis, we first tested all PRS as fixed effects in separate models, and a second step, adjusted for age, sex, batch, and the first five principal components for ancestry (PCs). The interpretation of a significant main effect of a PRS was that the PRS had a constant association with symptom severity through the entire treatment period. As a secondary analysis, to investigate the influence of PRS on the rate of change during treatment, we tested for interaction effects (PRS*time) for those PRS that had a significant main effect. A significant PRS*time effect was interpreted as the PRS being associated with the rate of symptom change during the treatment period.
To account for dropout in our modelling, we incorporated a pattern mixture term by dummy coding participants into two subgroups, non-completers (1) or completers (0); where participants that started <5 modules were classified as non-completers. ICBT patients could drop out of treatment for reasons related or unrelated to the treatment or symptom change, and reasons are in many cases unknown. By accounting for dropout in our model, we decrease the potential bias of our estimates caused by dropout, which a plain mixed model would otherwise ignore. To control for unwanted genotyping batch-effects, analysis also included three dummies (one per batch).