The questionnaire completed by the patients covered socio-demographics (age, sex, marital status, and educational level), work situation and mental health. In the logistic regression analyses, a dichotomous variable was created for educational level (more/less or equal to 13 years of education). The therapists diagnosed each patient and completed a form containing questions about each patient’s present problems and treatment history. After the treatment, the therapist again answered questions about the patient’s present status.
Work Participation
Using information from the questionnaires completed by the patients and therapists, and the medical records, we constructed an index for work participation with four mutually exclusive categories: (1)
Working fully; (2)
Working partly (working part-time and on sick leave, receiving a social benefit or partially unemployed); (3)
Not working (full sick leave or unemployed, and/or receiving social benefits or studying full time, including working additional hours part-time or receiving some form of social benefit); and (4)
Other (respondents who did not fit into any of the other categories, such as being on maternity leave). The category Other included no patients before and after treatment. At the 6-month follow-up, the Other category included six patients: five were on leave from work (two on maternity leave and three on leave from work for unspecified reasons) and one had retired from work. These six patients were omitted from the analyses of changes in work participation. Patients were categorized into four groups based on the International Standard Classification of Occupations (ISCO-88) codes [
19]: (1) high-skilled white-collar workers (e.g., teachers, nurses, social workers and engineers); (2) low-skilled white-collar workers (e.g., childcare workers and nursing assistants); (3) high-skilled blue-collar workers (e.g., construction workers and other skilled workers); and (4) low-skilled blue-collar workers (e.g., cleaners and other elementary occupations). In the logistic regression analysis of predictors of RTW, a dichotomous variable for occupational status was made by merging the last three alternatives.
RTW
Using the categories for work participation after treatment and at the 6-month follow-up, a new variable comprising five categories to define RTW was created. Moving to the Working fully category at the 6-month follow-up from any of the other categories after treatment was defined as Full RTW. Moving to the Working partly category from categories indicating no work participation after treatment was defined as Partial RTW. No change in work participation was defined as Still working fully, Still working partly or Still not working. Moving to a category indicating less work participation was defined as Working less. In the logistic regression analyses, a dichotomous variable for RTW was created as follows. Participants classified as having full or partial RTW were clustered together with those still working fully or partly, and their cases were defined as successful RTW. Those working less were clustered together with those still not working, and their cases were defined as failed RTW.
Mental Health and Psychological Variables
Symptoms of psychological distress were measured using the Clinical Outcomes in Routine Evaluation Outcome Measure (CORE-OM) [
20], a self-administered questionnaire with 34 items related to the preceding week. All items are scored from “Never” (= 0) to “Almost all the time” (= 4). Total mean scores are usually multiplied by ten before being presented as a total score ranging from 0 to 40 [
21]. Forms with fewer than 90% of the items completed were excluded from the analyses of the total score. An internal consistency for the CORE-OM of Cronbach’s coefficient (α) = 0.94 and a 1-week test–retest reliability of Spearman’s r = 0.90 have been reported [
22]. Consistent with this, we found Cronbach’s α = 0.95 for the data at the end of treatment in this study.
The therapists diagnosed the patients according to the International Classification of Diseases 10 (ICD-10) guidelines [
23], and the ICD-10 diagnoses were clustered into five categories:
Depression, including depressive episodes (F32) and recurrent depressive episodes (F33);
Anxiety, including phobic anxiety disorders (F40); other anxiety disorders (F41) and obsessive–compulsive disorder (F42);
Adjustment disorders (F43);
Other psychiatric diagnoses (e.g., substance abuse and eating disorders); and
Z-diagnoses (reasons for contact with health services not resulting in a psychiatric diagnoses, e.g., examination). For the logistic regression analyses, three dummy variables were created, in which each of the diagnoses Depression, Anxiety and Adjustment disorder was contrasted against the other two diagnoses combined. Therapists also scored the patients on the Global Assessment of Functioning (GAF), a 100-point scale from the Diagnostic and Statistical Manual of Mental Disorders—Fourth Edition, which is used to assign a global rating of a patient’s social, occupational and psychological functioning (1 is the lowest and 100 is the highest score) [
24]. Since 1998, Norwegian clinicians have used a split version of the GAF, with one scale for symptoms (GAF-S) and another scale for social functioning (GAF-F).
The work ability index (WAI) is a seven-item questionnaire used in occupational health services and research to assess work ability [
25]. We used two single items from the WAI. The first item, “Assume that your work ability at its best has a value of 10 points. How many points would you give your current work ability?” is scored from 0 (= worst) to 10 (= best). A high correlation between this single item for rating work ability and the total WAI score has been reported, and this single item is often used instead of the complete instrument [
26]. The second question used was “Do you believe, according to your present state of health, that you will be able to do your current job 2 years from now?” The original alternative answers in this question are: (1) “Unlikely”, (2) “Not certain” and (3) “Relatively certain”. Instead, we used the following answer alternatives: (1) “Yes, definitely”, (2) “Yes, to some degree”, (3) “No, not really” and (4) “No, absolutely not”. In the logistic regression analysis of predictors of RTW, a dichotomous variable was made by merging the first two alternatives and the last two alternatives. A dichotomous variable was also made for change on this variable: answers indicating either a change from a negative to a positive expectancy of future work ability or maintained positive expectancy of future work ability were combined, and answers indicating either a change from a positive to a negative expectancy of future work ability or a maintained negative expectancy of future work ability were combined. By providing the same distribution of individuals within the two categories, this variable to represent change from the beginning to end of treatment became identical to the dichotomous variable describing the status at the end of treatment. This showed a high stability over the time between the two measurement times.
Patients also indicated to what extent the problems they were seeking help for were caused by aspects of their work situation, using the following answer alternatives: (1) Yes, definitely, (2) Yes, to some degree, (3) No, not really and (4) No, absolutely not. In the logistic regression analysis of predictors of RTW, a dichotomous variable was made by merging the first two alternatives and the last two alternatives.
The generalized self-efficacy (GSE) scale assesses an individual’s beliefs in his or her own ability to deal with new or difficult situations [
27]. Possible responses on the 10 items range from “Not at all true” (= 1) to “Exactly true” (= 4). A mean score of 1–4 is calculated for all items. The scale has been used in many research projects and typically yields an internal consistency Cronbach’s α value of 0.75–0.91 [
28]. For the end of treatment data, we found a Cronbach’s α = 0.87 in this study.
Statistical Analyses
Data were analysed using SPSS for Windows, version 24 (IBM Corp., Armonk, NY). Differences between groups were analysed using analysis of variance for continuous and ordinal variables, and the Chi square test for categorical variables. Paired
t tests for continuous variables and marginal homogeneity tests for categorical variables were used to identify differences from before to after the follow-up. In all analyses of differences, a significance level of
p < 0.05 was used. Effect sizes (Cohen’s d) were calculated for changes in work ability and self-efficacy, and was defined as small (d = 0.2), medium (d = 0.5) or large (d = 0.8) [
29]. We adopted Jacobson and Truax’s criterion c for establishing the clinical cut-off for the CORE-OM total score [
30]. Using the SD from a Norwegian non-clinical sample [
31] produced a clinical cut-off of 11.1. Using already published data from before treatment [
18], the changes in clinical outcome measures during treatment were calculated. These analyses are not shown, but all changes were statistically significant. Logistic regression analysis was used to identify factors associated with RTW at the 6-month follow-up. Patient characteristics before treatment, mental health variables after treatment and changes from before to after treatment were used as independent variables. Univariable logistic regression analyses were first performed for all independent variables, with RTW as a dependant variable. Variables that had a
p value of < 0.20 in the univariable analyses were selected for inclusion in the multivariable logistic regression analysis. Multicollinearity between the remaining independent variables was tested by checking the variance influence factor (VIF) statistics. Multicollinearity was assumed when VIF scores were > 4. A multivariable logistic regression analysis with manual backwards stepwise selection was then performed, using a
p value of < 0.05 as the cut-off for inclusion in the final combined model (Wald statistics).