Data was provided by NIPH. It was obtained from the PMHC treatment arm of a pragmatic Randomized Controlled Trial (RCT) conducted in two Norwegian municipalities, Sandnes and Kristiansand. We looked into predictors of dropout among those who received the intervention, thereby making this a prospective cohort study design. The descriptions of subjects, materials, and methods were first described in the primary evaluation of the RCT by Knapstad et al. [
24].
Data collection and procedure
Participants in this study were recruited between November 2015 and August 2017 [
24]. The trial sites were found to be relatively similar to each other as well as representative for the Norwegian population on several sociodemographic variables, for instance, rates of immigrant background, higher education, and unemployment [
24].
Psychologists had professional responsibility for the service at each site. Ten therapists were included in the current study. The number of clients per therapist ranged from eight to 90 clients (m = 52). The majority of clients started with a four-session psychoeducational course. Low-intensity self-help programs were to a limited extent accessible throughout the trial period. Most clients received only low-intensity treatment in terms of group-based psychoeducation (36.5%) or a combination of low and high-intensity interventions (33%). Furthermore, 29.4% primarily received high-intensity treatment. Only 1% received guided self-help [
28]
.
Recruitment and participants
Information about the study was conveyed both through an information letter from NIPH to all GPs in the area and directly from the services at local GP association meetings. Citizens could get information about the study from their GP, through the municipality web page, local newspapers, and local radio. People who contacted PMHC in Sandnes or Kristiansand got an appointment for an initial assessment. This assessment consisted of a clinical interview to evaluate the client's mental health problems and motivation for treatment, in addition to providing information about the study.
There were predefined inclusion and exclusion criteria to evaluate participants’ eligibility for PMHC during the trial period. The criteria were supposed to resemble ordinary care. The primary inclusion criterion was anxiety and/or mild to moderate depression. The Patient Health Questionnaire (PHQ-9) and Generalized Anxiety Disorder scale (GAD-7) were used as screening instruments with predetermined cut-offs (PHQ-9 > = 10 and/or GAD-7 > = 8) [
24]. Upper cut-offs for excision of severity were not predefined as severity was also based on clinical judgment in the clinical interview. Further requirements were a minimum age of 18 years, place of residence in the relevant municipalities, and basic Norwegian language proficiency.
People were excluded if they met the criteria of more profound mental problems such as eating disorder, severe suicidal risk, bipolar disorder, severe depression, incapacitating anxiety, psychotic symptoms, substance abuse, or personality disorder. Another exclusion criteria was two or more previous attempts at treatment in the secondary services, without satisfactory effect. People with serious physical health problems as their primary challenge were also excluded. Those not considered eligible for PMHC were referred to their GP, secondary services, or other services suitable for their main challenge.
Those who met the inclusion criteria were asked to participate, gave their written consent and registered on a secure online data portal. The portal was developed by the Norwegian Social Science Data Services (NSD) and was used to collect all data and questionnaires from clients and therapists. It was also used to randomize the clients to either PMHC treatment or treatment as usual (TAU) [
24,
40]. There were 774 participants who were included in the trial, whereof 526 were randomized to PMHC treatment [
24,
40]. Participant data from the PMHC group was used for the analysis in this paper.
Clinical variables
PHQ-9 asks the responder to evaluate nine items describing each criterion for depression based on DSM-V. The response options vary from 0 (
not at all) to 3 (
nearly every day), which allows a maximum sum score of 27. Caseness was defined as a minimum score of 10. A score above 14 was defined as moderate to severe symptoms of depression. The scores were coded into three different categories, namely below cut-off (0–9), mild depression (10–14), and moderate to severe depression (15–27). The variable
below cut-off was used as a reference category. The PHQ-9 has been tested as a reliable and valid measure for making criteria-based diagnoses for depression, assessing symptom severity, and monitoring change over time [
27]. The internal reliability of PHQ-9 has been measured and evaluated, showing excellent test–retest reliability and Cronbach’s α between 0.86–0.89 [
27]. Cronbach’s α based on our data was 0.80.
GAD-7 measures the frequency of seven common symptoms of general anxiety. Similar to PHQ-9, the response options vary from 0 (
not at all) to 3 (
nearly every day). The maximum sum score is 21. Caseness was set at 8, and a score above 14 was defined as severe symptoms of anxiety. GAD scores were coded into three categories, namely below cut-off (0–7), mild-moderate anxiety (8–14), and severe anxiety (15–21).
Below cut-off was used as a reference category. GAD-7 has been found to have good validity and reliability for measuring general anxiety. The instrument can be used both to assess symptom severity and monitor change over time [
24,
47]. It has shown excellent test–retest reliability and Cronbach’s α of 0.92 [
47]. Cronbach’s α based on our data was 0.83.
The Work and Social Adjustment Scale (WSAS) measures impairment of daily function by evaluating five items on a scale ranging from 0 (
not at all) to 8 (
very severely). The answers are based on function at work and in social relations during the last month [
53]. The sum scores reported were converted to a binary variable. Scores within the highest tertile were coded as 1 (
low functional status), while scores in the lowest two tertiles were coded as 0 (
high functional status). WSAS has been used in former PMHC evaluations [
46]. Furthermore, WSAS has comparable reliability, sensitivity, and discriminant validity to PHQ-9 and GAD-7 [
53].
Duration of problems was measured in months. The variable was recoded into three categories: less than or equal to 6 months, between 7 and 24 months, and longer than 24 months. The middle category was used as reference based on findings from the literature review.
Sociodemographic variables
The sociodemographic questions were used as binary variables. These questions included sex (female: yes/no), higher education (university/college: yes/no), and immigration background (1st or 2nd generation immigrant: yes/no). Employment was assessed by two multiple response questions regarding current work status and source of income. Based on their answers, participants were coded into five different categories. These were employed, employed while receiving benefits, unemployed, students and other (e.g. retirees, full disability pensionists). The employed category was used as a reference category. Age was also used as a binary variable (above 30 years: yes/no) as the literature suggests that particularly younger people are at risk to drop-out. Even though there is always a degree of arbitrariness in choosing a cut-off, our observed data suggested a marked drop in the probability of dropping out after age 30.
Questions about lifestyle and social variables were also reported using binary responses. Most relevant for this analysis was the question of social support. The 3-item Oslo Social Support Scale (OSSS-3) covers the number of close confidants, the sense of concern shown by others, and perceived availability of practical help from neighbors [
26]. A sum score ranging from 3 to 14 was calculated. Clients scoring 3 to 8 were coded as 1 (
low social support), whereas those scoring 9 to 15 were coded as 0 (
medium to high social support). Validity and reliability for OSSS-3 have been reported as satisfying [
26]. Cronbach’s α of the OSSS-3 was relatively low based on our data (0.58).
Statistical analyses
Preliminary analyses were undertaken to prepare the specific statistical techniques to address the research question. All variables were checked for errors, outliers, normality of distribution, variance, and missing data. Within the variables higher education, duration of problems, and immigrant background, we found some missing data (< 3%). Missing data were handled by listwise deletion in the regression analyses. Logistic regression was considered the most appropriate analysis as the dependent variable was dichotomous [
35].
To examine possible relationships between dropout as a dependent variable and client factors as independent variables, we first did bivariate logistic regression analyses for nine variables of relevance according to the literature. Of sociodemographic variables, these were age, sex, immigrant background, work status, level of education, and social support. Of clinical variables, these were symptom severity, duration of problems, and daily function.
The independent variables reaching p values < 0.05 in the logistic regression analyses were subsequently included in a multivariate logistic regression model. If the strength of an association changed when included in the multivariate analysis, further analyses were conducted to understand what accounted for the variation in the outcome variable. This was done by exploring different combinations of variables using logistic regression analysis, and observing possible changes. Therapists and municipalities were included in all analyses as fixed effects. All statistical analyses were performed using IBM SPSS Statistics, version 28.0.1.0.