Transdiagnostic factors predicting the 2-year disability outcome in patients with anxiety and depressive disorders

Data were obtained from the Netherlands Study of Depression and Anxiety (NESDA), a multi-site naturalistic longitudinal cohort study investigating the long-term course and consequences of ADD and their predictors looking into psychosocial, biological and genetic determinants [27]. Between September 2004 and February 2007, 2981 participants aged 18–65 years were recruited. Overall, 1701 (57%) people with a current (within the last six months) diagnosis of ADD, 907 (30%) people with life-time diagnoses or at risk of ADD and 373 (13%) healthy controls were included. Patients who were not fluent in Dutch or with a primary diagnosis of psychotic disorder, obsessive–compulsive disorder or severe addiction disorder were excluded. 564 (19%) participants were recruited from community samples, 1610 (54%) participants were recruited from primary care practices and 807 (27%) participants were recruited from mental health organizations. Trained research assistants collected the data and interviews were taped to monitor and analyze data-quality and interviewer performance. After the baseline assessment (T0), follow-up assessments took place at subsequently 2 years (T1; N = 2596), 4 years (T2; N = 2402), 6 years (T3; N = 2256). The research protocol was approved by the Ethical Review Boards of each participating site. Written informed consent was obtained from all participants.

Several transdiagnostic factors of interest were assessed at T2 of NESDA, therefore T2 was set as the baseline of this study. Drop-out from T0 to T2 was as follows: From the 1701 individuals with ADD at T0, 409 dropped out (24%). The remaining 1292 individuals consisted of 632 individuals who had an ADD (37%) at T2 and 660 individuals who were remitted (39%). In addition, out of 1280 individuals without ADD at T0, 134 individuals developed ADD at T2 (10%), which were also added to our ADD sample at T2. Of the participants with a current (6-month) diagnosis of ADD at T2 (n = 766), patients with data on disability at both T2 and T3 (n = 615) were selected.

The presence of ADD was assessed at baseline (T2) using the Composite International Diagnostic Interview (CIDI, version 2.1, Dutch), a frequently used standardized diagnostic interview with good overall reliability and validity for depressive and anxiety disorders [28, 29]. Current ADD was defined as a dysthymic disorder, major depressive disorder, social anxiety disorder, panic disorder with or without agoraphobia, agoraphobia and/or generalized anxiety disorder classified according to Diagnostic and Statistical Manual of Mental Disorders (DSM)-IV criteria in the last 6 months [30].

Disability was assessed at baseline (T2) and after two years of follow-up (T3) using the World Health Organization Disability Assessment Schedule II (WHODAS II) [30]. This questionnaire measures the level of functioning and disability over the past 30 days in six domains of life: cognition (6 items, e.g., ‘in the last 30 days, how much difficulty did you have in concentrating on doing something for ten minutes?’), mobility (5 items, e.g., ‘in the last 30 days, how much difficulty did you have in standing for long periods such as 30 min?’), self-care (4 items, e.g., ‘in the last 30 days, how much difficulty did you have in washing your whole body?’), interpersonal interactions (5 items, e.g., ‘in the last 30 days, how much difficulty did you have in dealing with people you do not know?’), life activities (8 items, e.g., ‘in the last 30 days, how much difficulty did you have in your day to day work/school?’) and participation in society (8 items, e.g., ‘how much of a problem did you have in joining in community activities (for example, festivities, religious or other activities) in the same way as anyone else can?’). It consists of 36 items scored on a 5-point Likert-scale. Scores were transformed into a standard scale with a range from 0 to 100. It has a good reliability and a solid factor structure that remains consistent across different types of patient populations. With respect to validity, results are consistent with those from other measures of disability [31]. Since only 342 (55.6%) of the participants were now employed, the 32-item version of the WHODAS II (excluding work disability) was used in line with other studies, for example Iancu et al. 2020 [4, 31]. The 2-year disability outcome was defined as the disability change score over 2 years (disability T3 – disability T2).

The sample size calculation of this predictive factors research is based on an R-squared (R²) test in multivariable linear regression for the outcome variable ‘change in disability’. In testing a set of fourteen predictive factors by means of multivariable regression analysis using a model with ten covariates, including socio-demographic variables and baseline disability scores, that explain at least a conservative percentage of one percent of the outcome variable, assuming alpha = 0.05, a total sample size of n = 600 allows us to detect an effect size of at least an increase in R² = 0.03 with power = 1 – beta = 0.80. This power calculation involved a set of fourteen transdiagnostic factors, assuming controlling for ten covariates that explain only a conservatively one percent of the variability and was obtained from Stata Statistical Software version 17.0 [58] using ‘power rsquared 0.01 0.04, ncontrol(10) ntested(14) n(600)’.

Mean disability at baseline and follow-up were compared using a paired samples T-test. Strength of association between sociodemographic, clinical and transdiagnostic variables and the 2-year disability outcome was identified through univariable regression analysis (significance value set at 0.05). The dependent variable was the 2-year disability outcome defined as the disability change score over a period of 2 years corrected for baseline disability. A correction for baseline disability was applied to avoid distortions of the results due to regression to the mean.

Multicollinearity was suspected if correlation coefficients calculated prior to multivariable analysis were larger than 0.80 and if the variance inflation factor (VIF) was larger than 10 [59]. In case of a highly collinear relationship between variables, adjustment of the structure of the model and selection of independent variables was performed, to prevent a reduction of the statistical significance of the independent variables.

Independent sociodemographic, clinical and transdiagnostic predictors of the 2-year disability outcome were identified through 2 multivariable models. The significance value was set at 0.0035 after Bonferroni correction (0.05/14 transdiagnostic factors). Model 1 included all factors from univariable analysis with a p-value of < 0.25, in the spirit of purposeful variable selection [60]. To further explore the effect of transdiagnostic factors specifically, Model 2 included all transdiagnostic factors from univariable analysis with a p-value of < 0.25. The R-squared estimate indicates the model performance, with R² = 1 (range 0–1) signifying a ‘perfect’ fit.

Data were analyzed using SPSS Statistics for Windows version 25.0 [61].

Descriptive characteristics of the study sample (n = 615) are summarized in Table 1. Mean age was 46.8 years (SD 12.4). Most participants were female (69.6%) and were born in the Netherlands (92.7%). Most of the participants had obtained an intermediate or high education level (93%) and more than half of the participants were currently employed (55.6%). Major depressive disorder (52.7%), social anxiety disorder (34.5%), dysthymic disorder (19.8%) and generalized anxiety disorder (18.5%) were the most common diagnoses. Anxiety disorders were most prevalent and comorbid anxiety and depressive disorder(s) were common (29.9%). Mean disability was 26.2 (SD 15.9) at baseline and 24.3 (SD 16.7) after 2 years of follow-up.

A comparison of individuals with ADD (at T0) who did and did not drop out between T0 and T2 (this study’s baseline), is presented in Supplementary Table S1. Age, gender, education level, IDS, BAI, FQ and WHODAS-II scores were compared. In short, drop-outs were younger, had a lower education level, and had a lower score on the Fear Questionnaire – Social Phobia subscale.

The mean disability change score was -1.9 (11.8), signifying a small improvement of the level of disability over a period of 2 years (t₆₁₄ = 4.059, p = 0.000).

Predictors of the 2-year disability outcome (i.e. disability change scores over a period of 2 years corrected for baseline disability) in patients with ADD are shown in Table 2. In univariable analysis, transdiagnostic factors associated with the 2-year disability outcome were LOC (β = -0.116, p = 0.011), extraversion (β = -0.123 p = 0.004) and experiential avoidance (β = 0.139, p = 0.001).

The VIF of the independent variables varied between 1.0 and 5.0 and the largest correlation coefficient was 0.79 (Supplementary Table S2). Multicollinearity was not a problem.

In multivariable analysis of purposefully selected sociodemographic, clinical and transdiagnostic variables (Model 1), the total explained variance (R², excluding explained variance by baseline disability) was 0.090. The explained variance of the transdiagnostic factors in Model 1 was 0.032. Among the transdiagnostic factors, extraversion was predictive of the 2-year disability outcome (β = -0.173 p = 0.001). In multivariable analysis of purposefully selected transdiagnostic variables (Model 2), the explained variance of the transdiagnostic variables was 0.050.

Our aim was to examine transdiagnostic factors predictive of the 2-year disability outcome in patients with ADD, focusing on potentially malleable factors. The mean disability change score was negative, signifying a small improvement of the level of disability over a period of 2 years. Unfortunately, there are no established thresholds for interpreting the global or domain-specific scores in relation to the criterion of clinically significant impairment.

The first multivariable model considering a combination of purposefully selected sociodemographic, clinical and transdiagnostic factors (i.e. Model 1) explained 9% of the total variability in the 2-year disability outcome. Remarkably, the total variance explained by the model is about five times lower than reported by Iancu et al. 2020 describing comparable predictors of the six-year residual disability in patients with remitting MDD [4]. This is partly explained by their choice to include baseline disability as a predictor instead of performing a statistical correction. Another possible explanation is their more restricted study sample; the differences in results might indicate that there is a lack of overlap between predictors of the course of disability in anxiety and depressive disorders. However, ter Meulen et al. 2021 reported > 75% comorbidity in subjects with ADD. Transitions between depressive and anxiety disorders were common [62]. The beforementioned simultaneous and alternating occurrence of ADD indicate that anxiety and depressive disorders are less distinct entities than suggested by current classification systems and that they are likely to share predictors of the course of the disease. Other possible explanations for the difference in results are a stronger association of the predictors with more long-term disability.

Our study complements earlier research by providing evidence regarding the association of transdiagnostic factors with the course of disability. Results of univariable analysis showed that a higher degree of LOC, a higher level of extraversion and a lower level of experiential avoidance were transdiagnostic factors predictive of a more favorable 2-year disability outcome.

In multivariable analysis, the additional amount of variance explained by transdiagnostic variables was small to medium, both apart from and in combination with sociodemographic and clinical variables. These findings indicate that the examined transdiagnostic factors explain a small but unique part of the degree of variability in the 2-year disability outcome.

In both multivariable analyses (i.e. Model 1 and 2), a lower level of extraversion was predictive of an unfavorable 2-year disability outcome. Personality traits were originally believed to be stable between individuals over time, and changes over the life course were thought to follow common patterns induced primarily by brain maturation and genetic factors [63]. However, together with cumulating evidence, newer theories, such as the Experience-dependent and Mixed set-point model, emphasize that both genetic influences as well as cognitive and environmental influences can drive long-lasting changes [63]. In line with this theory, in their literature review, Roberts et al. 2017 describe evidence that the level of extraversion can change because of intervention [64]. Consequently, targeting extraversion may improve the 2-year disability outcome, but due to the small contribution to the variance in the disability outcome, the clinical relevance of targeting extraversion seems limited. Yet, it is noteworthy that extraversion has a comparable strength of association with the 2-year disability outcome as compared to the accepted and widely used disease severity measures (i.e. IDS and BAI), supporting the importance of looking beyond using disease severity measures as predictors.

This study has a few limitations. First, the assessment of several variables relies on self-report questionnaires. This is important to obtain the patient’s perspective, but results might be affected by information-processing bias. Although the 32-item version of the WHODAS II was used, this does not hinder comparison to other studies, because the scores of the 32-item and 36-item WHODAS II are comparable [30]. Secondly, sufficient data on the course and type of psychotherapy and psychopharmacotherapy and the level of pre-morbid functioning were not available and therefore, their impact on the results could not be determined. Also, other potential transdiagnostic factors that were not assessed in this study, such as other personality characteristics, emotion regulation, alexithymia, self-esteem, intolerance of uncertainty, loneliness, social support and attachment style might turn out to predict the yet unexplained part of the variability. And besides personal factors, environmental factors may account for a part of the variability as well [6].

Thirdly, setting the baseline at T2 of NESDA could have led to bias, due to selective loss-to-follow up between T0 and T2. However, this is not likely as Lamers et al. 2012 found that the overall attrition over a period of 2 years was very limited [65]. Fourthly, due to the observational nature of this study, no definitive conclusions regarding causality can be substantiated.

Strengths of this study are the prospective design and the inclusion of a large mixed sample recruited from different settings, resulting in a higher degree of external validity. Another strength is the use of a rich dataset with multiple putative transdiagnostic predictors.