JMIR Publications

We are scheduled to perform a server upgrade on Thursday, November 30, 2017 between 4 and 6 PM Eastern Time.

Please refrain from submitting support requests related to server downtime during this window.

Journal of Medical Internet Research


Citing this Article

Right click to copy or hit: ctrl+c (cmd+c on mac)

Published on 02.09.15 in Vol 17, No 9 (2015): September

This paper is in the following e-collection/theme issue:

    Original Paper

    Predictors of Response to Web-Based Cognitive Behavioral Therapy With High-Intensity Face-to-Face Therapist Guidance for Depression: A Bayesian Analysis

    1Department of Psychology, Faculty of Health Sciences, UiT The Arctic University of Norway, Tromsø, Norway

    2Division of Substance Abuse and Addiction, Vestre Viken Hospital Trust, Drammen, Norway

    3Department of Community Medicine, General Practice Research Unit, Faculty of Health Sciences, UiT The Arctic University of Norway, Tromsø, Norway

    Corresponding Author:

    Ragnhild Sørensen Høifødt, PsyD

    Department of Psychology

    Faculty of Health Sciences

    UiT The Arctic University of Norway

    9037 Tromsø

    Tromsø, 9037


    Phone: 47 776 49230

    Fax:47 776 45291



    Background: Several studies have demonstrated the effect of guided Internet-based cognitive behavioral therapy (ICBT) for depression. However, ICBT is not suitable for all depressed patients and there is a considerable level of nonresponse. Research on predictors and moderators of outcome in ICBT is inconclusive.

    Objective: This paper explored predictors of response to an intervention combining the Web-based program MoodGYM and face-to-face therapist guidance in a sample of primary care patients with mild to moderate depressive symptoms.

    Methods: Participants (N=106) aged between 18 and 65 years were recruited from primary care and randomly allocated to a treatment condition or to a delayed treatment condition. The intervention included the Norwegian version of the MoodGYM program, face-to-face guidance from a psychologist, and reminder emails. In this paper, data from the treatment phase of the 2 groups was merged to increase the sample size (n=82). Outcome was improvement in depressive symptoms during treatment as assessed with the Beck Depression Inventory-II (BDI-II). Predictors included demographic variables, severity variables (eg, number of depressive episodes and pretreatment depression and anxiety severity), cognitive variables (eg, dysfunctional thinking), module completion, and treatment expectancy and motivation. Using Bayesian analysis, predictors of response were explored with a latent-class approach and by analyzing whether predictors affected the slope of response.

    Results: A 2-class model distinguished well between responders (74%, 61/82) and nonresponders (26%, 21/82). Our results indicate that having had more depressive episodes, being married or cohabiting, and scoring higher on a measure of life satisfaction had high odds for positively affecting the probability of response. Higher levels of dysfunctional thinking had high odds for a negative effect on the probability of responding. Prediction of the slope of response yielded largely similar results. Bayes factors indicated substantial evidence that being married or cohabiting predicted a more positive treatment response. The effects of life satisfaction and number of depressive episodes were more uncertain. There was substantial evidence that several variables were unrelated to treatment response, including gender, age, and pretreatment symptoms of depression and anxiety.

    Conclusions: Treatment response to ICBT with face-to-face guidance may be comparable across varying levels of depressive severity and irrespective of the presence and severity of comorbid anxiety. Being married or cohabiting, reporting higher life satisfaction, and having had more depressive episodes may predict a more favorable response, whereas higher levels of dysfunctional thinking may be a predictor of poorer response. More studies exploring predictors and moderators of Internet-based treatments are needed to inform for whom this treatment is most effective.

    Trial Registration: Australian New Zealand Clinical Trials Registry number: ACTRN12610000257066; (Archived by WebCite at

    J Med Internet Res 2015;17(9):e197





    Several efficacious psychological and pharmacological treatments for depression exist [1]. A well-documented treatment is cognitive behavioral therapy (CBT), which has shown comparable effects as pharmacotherapy in treating mild to moderate depression with the additional benefit of reducing relapse [2,3].

    The therapy model, structure, and short-term format of CBT make it highly suitable for delivery through self-help material. Delivery through Internet services is one example and several studies have demonstrated the efficacy of Internet-based CBT (ICBT) for depression, especially when guided by a therapist (eg, [4,5-8]). In fact, the treatment effects from guided ICBT and standard face-to-face treatment seem to be comparable [9-11]. Despite the positive results, ICBT is not suitable for all depressed patients because the problem of nonresponse is notable ranging from 50% to 65% (eg, [6,7,12]). Therefore, the question of which patients this treatment is effective for is important to address. The aim of this study is to examine pretreatment variables that can predict response to an ICBT protocol that was published previously [13].

    General Prognostic Factors

    A number of studies have investigated factors predicting the course of depression in primary care and community settings. Factors associated with a poorer course of the depressive disorder include individual characteristics (eg, high levels of neuroticism [14,15]), socioeconomic factors (eg, low educational level [16,17], unemployment [16,17]), relational factors (eg, lack of social support [17-19], loneliness [17]), health-related variables (eg, somatic illness [17,18], severity of somatic symptoms [19,20], poor self-rated health [21], lower levels of mental [20] and global [15] functioning), and factors related to the depressive disorder (eg, baseline depressive severity [17-20,22], history of depression [23], duration of depressive episodes [16,18], dysthymia or double depression [15 24], and comorbidity with anxiety [17,19,24], substance abuse [22], or personality disorders [24]).

    Predictors of Response to Cognitive Behavioral Therapy

    In the literature on treatment response, the concepts of prognostic and prescriptive factors are discussed [25]. The former represent nonspecific predictors of response and the latter represent moderators which refer to variables predicting differential treatment response between treatments [25,26]. The latter is most useful for informing which treatments seem most suitable for which patient characteristics or subpopulations [27].

    Several patient characteristics have been suggested to influence response to CBT for depression. Patient expectancy, perceived treatment credibility, and improvement in the early phases of treatment seem to be powerful predictors of outcome in cognitive therapy and psychotherapy in general [28-32]. Demographic variables such as gender, age, education, and employment status are less consistently related to treatment outcomes [33-36]. However, in a recent study of treatment-resistant depression in primary care, age was found to moderate the effect of CBT with older patients gaining most benefit from this treatment [37]. In addition, married patients seem to respond consistently better to CBT compared to unmarried patients [38-40]. Many studies suggest poorer outcomes in terms of posttreatment symptoms for patients with high baseline depressive severity (eg, [41-43]). This relationship may depend on the definition of outcome; Van et al [34] propose that high initial severity may be associated with more difficulty achieving remission, whereas symptom change may be achieved more readily because higher severity leaves more room for improvement. In addition, regression to the mean effects can be expected to be stronger for those with a higher symptom load. Other features of the depressive disorder, such as high chronicity and younger age of onset, have been found to predict poorer response to CBT [29,38], but the predictive role of number of depressive episodes [29,30,39] and comorbid anxiety remains unclear [44-47]. With its relation to the proposed mechanism of change in CBT, the role of dysfunctional attitudes has received considerable attention and several studies conclude that high baseline levels of dysfunctional attitudes predict a poorer treatment response [29,33,39,48,49].

    Predictors of Response to Internet-Based Cognitive Behavioral Therapy

    For ICBT, results concerning depressive severity are consistent with previous research on face-to-face CBT [4,50-56]. In contrast to previous research, studies of ICBT have found either no association between marital status and treatment response [6,52,57] or a positive association between being separated, widowed, or divorced and symptom reduction [53]. Two studies of younger and older adults, respectively, found more favorable outcomes for females [52,58]. Donker et al [59] found similar results in a sample with a broader age range, whereas others have not replicated this finding [4,6,12,56,57]. Age itself did not significantly predict outcome in these studies, with the exception of Donker et al [59] in which age was found to be a moderator because older individuals responded more favorably to CBT and younger individuals improved more with interpersonal therapy (IPT). Results have been mixed for educational level, employment status, dysfunctional attitudes, and for clinical variables such as number of depressive episodes and the presence of comorbidity (eg, [6,12,50-53,56,57,59]).

    Treatment credibility refers to the extent to which patients endorse a treatment model as logical and meaningful, and 2 studies found this to be unrelated to outcomes of ICBT [60,61]. Results were mixed with respect to treatment expectancy [60,61]. One study indicated that although higher motivation was associated with greater adherence, low and moderate levels were related to better outcomes, perhaps due to unrealistic expectations and proneness to disappointment for highly motivated participants [57]. One may presume that greater adherence leads to better outcomes, but even on this point there are inconsistencies with some studies finding an association [5,60,62-65] and others not [12,54,57,66-68]. A review suggested that the impact of adherence may depend on how it is measured and that module completion may be more consistently related to outcomes for depression than measures such as number of log-ins [69].

    Aim of the Study

    The aim of this study is to identify prognostic predictors of response to an intervention combining the Web-based program MoodGYM and high-intensity face-to-face therapist guidance in a sample of mildly to moderately depressed primary care patients. Data from a randomized controlled trial (RCT) comparing this intervention to a delayed treatment condition was used. Data from the treatment phase of the 2 groups were collapsed. This increased sample size in the treatment group, but also precludes a clear distinction between general prognostic factors and predictors specific to CBT. This limitation must be borne in mind when interpreting the results. Predictor variables were predominately chosen on the basis of previous research on CBT delivered face-to-face and over the Internet, but some measures were included for exploratory purposes.

    Most patients with mild to moderate depression receive all their treatment in primary care where the availability of psychological treatments is often limited [70-73]. If implemented in primary care, this intervention could constitute an alternative to treatment as usual. This paper may indicate which patients in a depressed primary care population may benefit more or less from treatment with MoodGYM and therapist guidance.

    Based on previous literature, we hypothesized that (1) more positive expectations would predict a more favorable response to treatment, (2) participants with higher baseline depression severity would improve more, and (3) a higher score on a measure of dysfunctional thinking would predict a poorer treatment response. Because the remaining predictor variables have yielded mixed results in previous studies, no specific hypotheses were formulated for these.


    Study Design

    The study was a RCT with 2 conditions: (1) a treatment condition comprising 6 weeks of Web-based CBT with face-to-face therapist guidance and (2) a 6-week waitlist for the same treatment during which time participants could also access treatment as usual. The research protocol was approved by the Regional Committee for Research Ethics in Northern Norway and the Human Ethics Committee of the Australian National University (ANU). The trial was registered in the Australian New Zealand Clinical Registry (ACTRN12610000257066). A more detailed account of the study methods is given in Høifødt et al [13].

    Participants and Procedure

    Participants (N=106) were recruited from general practitioners (GPs), primary care nurses, and from waitlists of primary care referrals at 2 psychiatric outpatient clinics. Local GPs and primary care nurses were informed about the study and provided their patients with information about the project. Patients on waitlists at the psychiatric outpatient clinics at the Psychiatric Centre for Tromsø at the University Hospital of North Norway and at the Department of Psychology at UiT The Arctic University of Norway were invited by postal mail. Patients consented by signing an informed consent form. Consenting participants were screened for inclusion and randomly allocated to the 2 groups. The study inclusion criteria were (1) aged 18 to 65 years, (2) access to the Internet, and (3) a score between 10 and 40 on the Beck Depression Inventory-II (BDI-II), which indicates mild to moderately severe symptoms of depression. Individuals already attending CBT were excluded. Participants with suicidal intentions, concurrent psychosis, or alcohol or drug abuse disorders were excluded. Participants who used antidepressant medication were stabilized for 1 month prior to entering the trial.

    Assessments and treatment took place at the Department of Psychology at UiT The Arctic University of Norway. Because the patients allocated to the 2 study arms showed comparable courses during treatment, data from the treatment phase of the 2 groups was combined to increase statistical power. Seven participants in the control group dropped out during the waiting period and did not complete the pretreatment assessments. Another 7 participants did not meet the inclusion criteria according to the BDI-II at the pretreatment assessment and were excluded, as were 7 participants who provided data only on 1 measurement occasion. In addition, 3 outliers with treatment duration exceeding far beyond that of the rest of the sample (>28 weeks) were excluded. Because slopes of BDI-II were modeled as a function of time, treatment duration is a critical variable; therefore, we chose to base our criterion for outliers on this scale.


    The guided self-help intervention included (1) The Norwegian version of the ICBT program MoodGYM version 3 [74], (2) face-to-face therapist guidance of high-intensity, and (3) reminder emails between sessions.

    The MoodGYM was originally developed at ANU as a free-of-charge automated Web intervention delivered to the public [75]. MoodGYM consists of 5 self-help modules and 29 exercises. The program is based on CBT and was developed to prevent and reduce symptoms of depression and anxiety among adolescents [76], but is efficacious for adult populations also [8,77-79]. MoodGYM focuses on identification and restructuring of dysfunctional thinking, activation of behavioral strategies to increase engagement in positive activities, as well as learning of stress reduction and problem-solving techniques.

    Participants were introduced to the program and instructed to complete one module per week. After each module, participants received face-to-face support (15-30 minutes) from a psychologist (RSH or KL). The main elements of the sessions were reinforcement of progress, discussion of key messages from the modules, and helping participants to relate to the material and employ techniques from the program in their daily life. The full intervention included 8 sessions. The mean number of sessions attended was 7.0 (SD 2.2). Due to delays, some participants attended more sessions (9 sessions: n=8; 10 sessions: n=3; 11 sessions: n=1). Mean session length in minutes (excluding screening) was 28.1 (SD 6.9, range 15.8-48.6). Therapists aimed to meet participants weekly. However, the interval between sessions and the number of sessions were allowed to vary to meet individual needs. Thus, treatment duration varied between participants (mean 9.6, SD 4.8, range 1-22) and there was no fixed posttreatment time-point.


    Several outcome measures were analyzed in the trial focusing on the effect of the intervention [13]. In this paper, analyses are restricted to predicting response on the BDI-II. The BDI-II was administered to all participants at baseline (before randomization) and before every consultation during the intervention phase. The control group also completed an assessment before entering online treatment (pretreatment).

    The BDI-II is a 21-item self-report measure of severity of depressive symptoms during the past 2 weeks [80]. Studies consistently support the BDI-II as a reliable, internally consistent, and valid scale for assessing depression [80-82]. In this study, internal consistency (Cronbach alpha) ranged from .79 to .97 and was generally greater than .90 for the measurement occasions T1 to T11 (baseline to session 11).

    Predictors of Outcome

    Demographic Variables

    The variables gender, age, marital status, and employment status were collected during the screening interview before randomization. Marital status and employment status were dichotomized as married/cohabiting versus not married/cohabiting and being employed versus not being employed, respectively.

    Severity Variables

    This group of variables included pretreatment measures of severity of depressive and anxious symptoms and quality of life, as well as depression and anxiety diagnosis, number of depressive episodes, and alcohol use measured at baseline. In addition, previous treatment was included as a dichotomous variable (1=yes, 0=no) indicating whether participants had previously received pharmacotherapy or psychological treatment for depression.

    Severity of anxiety and depression symptoms pretreatment was assessed with The Hospital Anxiety and Depression Scale (HADS). This inventory has 2 subscales of 7 items each, measuring depression and anxiety, respectively, and is reliable and valid [83,84]. In this study, Cronbach alpha was .67 and .81 for the depression and anxiety subscales, respectively.

    Another measure of anxiety severity was The Beck Anxiety Inventory (BAI) [85]. The inventory possesses robust internal consistency, reliability, and validity [86-88]. Cronbach alpha in the present study was .92.

    The Mini-International Neuropsychiatric Interview (MINI) [89] was used to identify participants who fulfilled the criteria for a major depressive episode (MDE) or any anxiety disorder, and to determine the number of previous depressive episodes (0=no lifetime MDE; 1=single lifetime MDE; 2=2-4 lifetime MDEs; 3=≥5 lifetime MDEs).

    Alcohol use was assessed with the Alcohol Use Disorders Identification Test (AUDIT) [90]. The instrument has favorable internal consistency, reliability, and criterion validity [90,91]. Participants with scores greater than 20 were excluded from the study. Cronbach alpha in this study was .81.

    Health-related quality of life was assessed with the EuroQol 5-Dimension Self-Report Questionnaire (EQ-5D) [92]. Respondents mark their level of functioning for each of 5 dimensions (mobility, self-care, usual activities, pain/ discomfort, and anxiety/depression).

    The Satisfaction With Life Scale (SWLS) measures global life satisfaction according to the individual’s own criteria [93]. The scale has sound psychometric properties [94,95]. Cronbach alpha in this study was .78.

    Cognitive Variables

    Dysfunctional thinking and self-efficacy were explored as potential predictors of response. Dysfunctional thinking patterns were measured with the Warpy Thoughts Quiz, which is part of the first module of MoodGYM [96]. The 42-item quiz covers 7 areas of dysfunctional thinking: the need for approval, love, to succeed, and to be perfect; expectations of rights; influence on others; and the view that happiness depends on external things. Items are rated from 1 (strongly agree) to 5 (strongly disagree). Higher scores indicate more dysfunctional thinking. Norms were based on a sample aged 20 to 32 years (N=153) [97], and the scale demonstrates good internal consistency (Cronbach alpha=.77-.84) [96]. A 20-item short form of the scale correlates strongly with the Automatic Thoughts Questionnaire (r=.51) and moderately (r=.39) with measures of depression and anxiety [98].

    The General Self-Efficacy Scale (GSE) assesses broad and stable beliefs about one’s ability to deal with various demands and challenges [99]. The GSE has satisfactory reliability and construct and criterion cross-cultural validity [100-103]. Cronbach alpha in this study was .89.

    Expectancy, Motivation, and Use

    Expectancy, attitudes toward using an Internet-based program, and motivation were measured after introducing CBT and MoodGYM using questions developed for the purpose of this study:

    1. To which degree do you expect that an Internet-based self-help program can be helpful for your depressive symptoms?
    2. How is your attitude toward using an Internet-based self-help program?
    3. How likely is it that you will use this Internet-based self-help program?

    For the first 2 questions, 5-point Likert scales (1=very high expectations, 5=very low expectations; 1=very negative attitude, 5=very positive attitude) were used. Responses to the item on motivation (question 3) were given on an 11-point scale from 0% to 100%. User data on module completion was registered online and was denoted by a number between zero and 4, with zero indicating no use and 4 indicating completion of the module.

    Statistical Analyses Using Bayesian Statistics

    Motivation for Using Bayesian Methods

    Bayesian methods were used for data analyses instead of the more commonly used null-hypothesis significance testing (NHST) approach (for a general introduction to Bayesian methods, see [104,105]). In a Bayesian framework, we directly estimated the posterior probability distribution of the parameters taking data and model structure into account. Bayesian methods are suitable in the current setting for several reasons. First, the use of Bayesian hierarchical modeling allows the design of custom models that are appropriate for the data without relying on approximations as is necessary in NHST methods. Furthermore, Bayesian modeling is highly flexible because the posterior distribution can be readily transformed into easily interpretable quantities and the uncertainty inherent to the analysis is propagated and available at each level of analysis. As such, Bayesian analysis relies much less on point estimates and an arbitrary choice of significance levels. Indeed, the strong critique on P values (regarding, for example, their biasing impact on which results are trusted/reported and the problems with their interpretation [106,107]) emerging in many relevant scientific fields such as medicine [108] and psychology [106] has triggered the development of Bayesian methods in these fields (eg, [109,110]). Instead of reporting P values and relying on the problematic concept of statistical significance using an arbitrary significance level, Bayesian methods report the results of an analysis in terms of probabilities, odds ratios, and Bayes factors that give a more graded and readily interpretable summary of the conclusions supported by the data.

    Odds ratios are ratios of probabilities or densities indicating the probability of one event occurring relative to another. Similarly, the Bayes factor quantifies how much more likely one hypothesis is with respect to another by dividing the posterior model odds by the prior model odds. Note that the Bayes factor integrates the probability over the complete parameter space and, therefore, automatically punishes overly complex models. Jeffreys [111] discussed how Bayes factors could be interpreted in terms of strength of evidence for and against a hypothesis (see Table 1) and it has been shown that Bayes factors are less prone to overestimating effects from psychological experiments compared to P values [112].

    Using Bayes factors, Bayesian modeling may quantify the support for the null hypothesis and to what extent the null hypothesis (H0) is more likely than the alternative (H1). This is advantageous compared to traditional NHST-based tests which can only “not reject” the null hypothesis. This is a desirable feature when investigating the potential impact of predictor variables on treatment efficiency.

    Table 1. Evidence categories for Bayes factors (BF10).a
    View this table
    Statistical Models

    Depression scores from BDI-II were acquired for each individual over several weeks of treatment. Because the intervention allowed a flexible session schedule there was resulting variation in measurement occasions; therefore, the effects of time from treatment could not be disentangled. Because participants could use the self-help program between sessions, we hypothesized that participants would continuously benefit from the treatment between sessions. Therefore, time (in weeks) was chosen as the repeating variable because this was considered to be the most correct representation of the data. We conducted a model selection procedure (for details see Multimedia Appendix 1) to find the most faithful representation of our data from among a linear, a quadratic, and an exponential model. Based on the results from this procedure, we modeled the BDI-II scores on the individual level as an exponential function of time and constrained the individual regression coefficients by a group-level distribution (hierarchical model). In Bayesian analysis, the specification of prior belief is essential. We specified a weakly informative prior such that the estimates were allowed to vary across a large number of parameter values while constraining them to be in a plausible range [105,113].

    We implemented 2 complementary models, one for predicting probability of responding to treatment and another one for quantifying the strength of the response. The models were fit using Markov chain Monte Carlo (MCMC) algorithms implemented in the Just Another Gibbs Sampler (JAGS) software [114] and convergence was ensured by visual inspection and the Gelman-Rubin diagnostic [115]. We also conducted posterior predictive checks to ensure that the model fit the data well [105] (see Figures S1-S3 in Multimedia Appendix 1).

    Predicting Probability of Response

    Response to depression treatment varies substantially across individuals [27]. Latent-class approaches allow for the modeling of different growth trajectories across subgroups and captures this unobserved heterogeneity in trajectories by employing a categorical latent variable [116,117]. Class membership is initially unknown, but is inferred based on observed data resulting in identified classes of individuals with more similar response patterns within each group than between groups [116]. Thus, different classes of individuals may vary around different mean growth curves with potentially unique forms and parameter values. This can be advantageous compared to conventional growth modeling which assumes that all individuals are drawn from the same population and estimates the average growth curve for this population [118]. Furthermore, covariates can be included in the model to predict class membership and, in this way, individual characteristics predicting differential trajectories may be identified. Previous investigations have successfully employed latent-class methods to identify different distributions for groups of responders and nonresponders to treatment [32,119,120]. Therefore, we chose to fit a model that assumed 2 different distributions from which subject-level parameters could be drawn. Predictor variables were used as regressors on probability of class membership using a logit link function (for details see Multimedia Appendix 1) resulting in estimates βi for each predictor. The resulting model effectively distinguished between responders and nonresponders (see Figure S4 in Multimedia Appendix 1).

    Predicting the Strength of Response

    In a next step, we aimed to explain variation in responsiveness by identifying variables that correlated with the slope of the response. This is an alternative way to look at prediction of response and it has the advantage of being more directly comparable to previous studies because latent-class approaches have not been widely used in the field. We modeled this situation by adding the subject-level covariates as linear predictors on the estimate of the first-level regression slope. Because changes of the slope parameter in the exponential model are not reflected linearly (a unit change on a low slope parameter has strong impact whereas the same change on a higher slope parameter has less impact), we relied on the quadratic model for this approach. This resulted in estimates αi for the regression coefficient for each predictor.


    Sample Characteristics

    A total of 106 participants were included in the study and randomized to an intervention condition (n=52) or a delayed treatment control condition (n=54). Figure 1 describes the flow of participants through the trial. Of the 54 participants in the control group, 47 (87%) showed up for pretreatment assessment after being on a waitlist. For the control and intervention groups, 21 of 47 (45%) and 15 of 52 (29%) participants, respectively, dropped out between pre- and posttreatment assessments.

    Treatment adherence was moderate with 31 of 52 participants (60%) in the intervention group and 20 of 54 participants (37%) in the control group adhering to treatment (completing MoodGYM and attending at least 7 sessions). The average number of completed modules and pretreatment characteristics of the sample are presented in Table 2. Distributions for the predictors are shown in Figure S5 in Multimedia Appendix 1.

    Table 2. Participant characteristics (N=82).
    View this table
    Figure 1. Flow of participants through the trial.
    View this figure

    Predicting Probability of Response

    The restricted 2-class model distinguished well between responders and nonresponders (ie, most participants either have a very low or a very high probability of belonging to the responder group, Presp) (see Figure S4 in Multimedia Appendix 1). Using Presp=.05 as split criterion, we found that 21 of 82 (26%) participants did not respond to treatment, whereas 61 of 82 (74%) did. These results were based on the conditional latent-class exponential model encompassing all predictor variables. A corresponding analysis using the quadratic model found qualitatively similar results. The results of the regression of the covariates on the probability to respond to the treatment are reported in Table 3. The odds ratios indicate the degree of evidence that each covariate has a positive/negative impact relative to the probability of the opposite (eg, as indicated in Table 3, it is almost 15 times more likely that a subject’s score on the Warpy Thoughts Quiz affects the probability of him or her responding to treatment negatively rather than positively). Thus, the odds ratios give an indication of the likely direction of the effect of a covariate on the probability of response, but do not delineate the strength of this effect. To give an indication of the strength of the effect, the probability of being in the responder group as a function of each of the covariates is plotted in Figure 2. This relatively complex reporting of the strength of effects was necessary for this analysis since the estimation of Bayes factors in latent-class models is computationally complicated and still a topic of ongoing research.

    Table 3. Posterior mode, highest density interval (HDI), and odds ratios for the beta coefficients predicting probability of being a responder. The odds ratios indicate the probability that each covariate has a positive/negative impact relative to the probability of the opposite (+: positive effect; -: negative effect), but do not indicate the strength of this effect.
    View this table

    In summary, having had more depressive episodes, being married or cohabiting, and scoring higher on life satisfaction (SWLS) had high odds for positively affecting the probability of response. Tentative positive effects were found for the number of completed modules and having a paid job. Figure 2 shows that the effects are strongest for number of depressive episodes and scores on the SWLS with the probability for response approaching 1 for those with 5 or more depressive episodes and those with highest levels of life satisfaction, whereas those never having had a major depressive episode (only symptoms) and those with the lowest level of life satisfaction had only approximately .50 probability of response.

    In the opposite direction, higher scores on the Warpy Thoughts Quiz were likely to have a negative effect on the probability of responding to treatment. Tentative negative effects were found for health-related quality of life (EQ-5D), motivation, expectancy, scores on both subscales of the HADS, and for alcohol use (AUDIT). Figure 2 shows that high scores on the Warpy Thoughts Quiz appear to be associated with a substantially reduced probability of response (Presp~.40). The impact of the other covariates are more limited (Presp~.60-.80 for participants with scores in the highest range; see Figure 2).

    Figure 2. Probability of being in the responder group as a function of the predictor variables (assuming all other predictors remained at their baseline level). Black line is the mean posterior probability and shaded area is the 95% highest density interval.
    View this figure

    Predicting the Strength of Response

    The analysis of variation in responsiveness indicated that the predictors having the highest impact on response were largely consistent with the results from the latent-class model with the most important variables being the Warpy Thoughts Quiz, number of depressive episodes, life satisfaction (SWLS), module completion, and marital status. Results are summarized in terms of odds ratios in Table 4. Results from a separate analysis exploring the variation in responsiveness in the subgroup of responders (n=61) extracted by the latent-class model described in the previous section are presented in Multimedia Appendix 1.

    Bayes factors quantify the strength of evidence for the null hypothesis (the covariate does not affect treatment response) and for the alternative hypothesis (the covariate affects response to treatment).

    The results were largely consistent with the results from the odds ratio analyses with regard to which variables were most influential (see Table 5). However, the evidence was substantial only for the effect of marital status. There was substantial evidence for the null hypothesis for several variables, indicating that these variables are likely to be unrelated to treatment response in the present trial. This included the variables gender and age, and several severity variables including pretreatment symptoms of depression and anxiety, as well as treatment expectancy, attitude, and motivation. Inconsistent with the results from the odds ratio analyses, there was substantial evidence that the Warpy Thoughts Quiz was unrelated to treatment response.

    Table 4. Posterior mode, highest density interval (HDI), and odds ratios for the α coefficients predicting the strength of the response. The α coefficients are the group-level regression coefficients on the slope of the treatment effect in the quadratic model (see Equation 5 in Multimedia Appendix 1). The odds ratios indicate the probability that each covariate has a positive/negative impact relative to the probability of the opposite (+: positive effect, –: negative effect), but do not indicate the strength of this effect.
    View this table
    Table 5. Bayes factors (BF10) quantifying the evidence for alternative hypotheses (H1) over the null hypothesis (H0). Variables are sorted with respect to its Bayes factor in ascending order. The null hypothesis is that the predictor does not have an impact on treatment response (H0: α1=0) and the alternative is that it does have an effect (H1: α1≠0). BF10 is the odds for H1 divided by the odds for H0.
    View this table


    Principal Findings

    This paper explored predictors to a treatment combining the MoodGYM program and high-intensity face-to-face guidance. Using Bayesian methods and a latent-class approach, a 2-class model classifying 74% of participants as responders and 26% as nonresponders was identified. The variation in responsiveness was also explored by analyzing whether predictors affected the slope of response. The results suggest that treatment effects were unrelated to baseline depressive severity, gender, and age. In addition, the presence and severity of comorbid anxiety did not predict differential response to treatment. Having a partner and reporting higher life satisfaction at baseline were associated with a more favorable treatment response. Results also indicated that having experienced more depressive episodes may predict more positive treatment effects, whereas higher scores on the Warpy Thoughts Quiz, which is a measure of dysfunctional thinking, may predict poorer response to treatment.


    The results of this study must be interpreted in light of some methodological limitations. Despite merging data from both treatment groups, the size of the sample is limited and may be too small to allow reliable testing of effects. Small sample sizes are a common problem in research on prediction of treatment outcome [121]. Collapsing the data from the 2 groups increased the sample size in the treatment group, but precluded the identification of predictors or moderators of differential treatment response [27]. This means that this study cannot accurately distinguish between nonspecific predictors of good prognosis, nonspecific predictors of response to any treatment, and moderators (predicts differential response to treatments). In an effort to ameliorate this limitation, we separately investigated whether the individual predictors could explain the change in the BDI-II score during the waiting time for the waitlist control group. Due to the limited sample size, Bayes factors indicated no evidence for the alternative hypothesis for any of the predictors and could neither establish confidence in the null hypothesis for most of the variables.

    Multiple comparisons in small samples also introduces a risk of chance findings. Studies with low power have a high chance of overestimating effect sizes or even making sign errors (eg, [122]). Bayesian methods allow us to model all data in a joint context and reduce the multiple comparison problem by constraining individual model coefficients by an overarching distribution (for details see [104]). In addition, formulating the results in terms of probabilities and odds ratios rather than making dichotomized decisions about whether or not a variable serves as a predictor or not can prevent overinterpretation of results.

    Another limitation is that the intervention allowed a flexible session schedule and hence a variation in the spacing between measurement occasions. This means that the effects of time from treatment cannot be disentangled. Because participants could use the self-help program between sessions, we hypothesized that participants would continuously benefit from the treatment also between sessions. Therefore, time was chosen as the repeating variable because this was considered to be the most correct representation of the data.

    The choice of outcome and predictor variables may also be criticized. Although demographic variables and baseline axis-I diagnoses were well covered, several variables that may have important contributions, such as personality variables, were not investigated. Therefore, these results can only give a partial description of factors influencing treatment response. The sole reliance on self-report is another limitation. Furthermore, treatment expectancy, attitudes, and motivation were measured using invalidated single items developed for this study. In addition, the convergent and discriminant validity of the Warpy Thoughts Quiz have not been established. This leaves uncertainty regarding how well these constructs were captured and calls for caution in interpreting the results.

    A limitation of the 2-class model was that we were unable to estimate Bayes factors due to statistical complexity. This would have provided additional information about the strength of effects. Bayesian methods are a field of active research and development, and improved methods will surely be available in the future.

    Finally, although a strength of the study is the recruitment of a relatively heterogeneous sample of primary care patients with regard to the range of depression and anxiety symptoms, the generalizability of the results is uncertain because the sample was a self-selected group. Nevertheless, an estimated uptake of 39% indicates that the trial sample may be representative of a considerable proportion of the targeted patient group [13]. Because some participants were excluded from analyses (eg, participants present at only one measurement occasion), results are based on a subsample of trial participants which further limits generalizability.

    Variables Unrelated to Treatment Response

    Bayesian methods may be used to indicate the likelihood of the null hypotheses. The present analyses provide substantial evidence for the absence of any effect for several variables, such as pretreatment symptoms of depression and anxiety, depression or anxiety diagnosis, earlier treatment, and the demographic variables gender and age. This implies that MoodGYM combined with face-to-face guidance of relatively high intensity may be expected to work equally well for adult patients of varying ages, for women and men, and for various mild to moderate depressive symptom profiles, as well as for patients with comorbid anxiety of varying severity. Previous results regarding the predictive role of anxiety have been mixed [44-47,51,56,59,123]. With regard to depressive severity, several studies of CBT have found a larger response in terms of symptoms change for patients with higher severity (eg, [34,43,50,52,55]). However, these patients also tend to have more difficulties with achieving remission [39,43,124]. This trial did not find evidence for more improvement among participants having higher initial depressive severity; nevertheless, the results suggest that patients with higher depressive severity appear to benefit from treatment. Whether remission was achieved at comparable rates for participants with more or less severe depression cannot be answered by the present analyses. In addition, because the range of symptom severities was restricted because patients with severe depression were excluded and the proportion of patients having severe anxiety was small, no conclusions can be drawn with regard to more severe cases.

    Predictors of Improved Response

    Being married or cohabiting was the most robust predictor of favorable response to treatment. This effect was evident both in the latent-class model and the analysis exploring the strength of response, and the Bayes factor indicated substantial evidence for a predictive effect. These results are in accordance with previous research on CBT delivered face-to-face [38-40]. In fact, some studies have suggested that marital status may be a prescriptive predictor for better outcomes in CBT compared to medications or IPT [38,40]. Although, this study cannot identify moderators, these past results indicate that having a partner is likely to be a predictor of treatment response and not merely of good prognosis. Supportive relationships were emphasized in interviews with participants from the current trial [125]. Participants described how important others encouraged them and facilitated their engagement in treatment (eg, by helping them make time to use MoodGYM or attend sessions). This strengthened their hope for recovery and motivation. Although important others also include friends and other family, one may hypothesize that living with a partner may facilitate such reinforcing processes. Also, being married or cohabiting may reflect a better ability to establish and maintain close relationships and this may in itself be an important factor for success in treatments that include interaction with a therapist [39]. This study included high-intensity face-to-face support. This may explain why this effect was evident in the current trial, whereas most studies of ICBT have failed to find any relation between marital status and response [6,52,57]. Replications within other contexts may decide whether this effect is unique to interventions including face-to-face contact or if similar processes operate also in Internet-based interventions including less support.

    Life satisfaction also emerged as a possible predictor of better response to treatment, although the Bayes factor analysis indicated only anecdotal evidence for this effect. Life satisfaction may be regarded as an indirect measure of illness severity. The SWLS does not directly tap into constructs such as affect, but it is significantly negatively correlated with measures of depression and anxiety [94,95]. This result is consistent with an early study of ICBT in which higher quality of life, although assessed with a different scale, was associated with better outcomes [51]. However, this has not been replicated in other studies [56,59]. Why health-related quality of life (EQ-5D) showed tendencies toward predicting more inferior response in this study is more of a riddle. However, the 2 scales assess quite different constructs with the SWLS focusing on how satisfied individuals are with life according to their own criteria and not based on the presence or absence of specific ailments or impairments. The EQ-5D, on the other hand, focuses on the latter. These 2 constructs need not be highly correlated as is supported in this study’s data (r=.23). Whether life satisfaction is a more potent predictor of better treatment response remains to be replicated.

    The results indicate that more depressive episodes have high odds for predicting a more favorable response. However, the result of the Bayes factor analysis was more ambiguous. This result is puzzling given that high rates of recurrence have been related to poor treatment outcomes [30,51] and treatment resistance [126,127] in previous studies. However, the findings are inconsistent and other studies have found no negative effect of high rates of recurrence on treatment outcomes [4,38,39,53,59]. There are some possible explanations for this finding. Compared with participants with a single or no depressive episodes, more participants with recurrent depression received antidepressant medication or additional psychological therapy. Although most did not receive additional treatment (~65%) and medications were stabilized for 1 month before entering the trial, one cannot rule out the possibility of this influencing the treatment effect. This would be consistent with a meta-analysis finding significantly better effects when adding psychotherapy to pharmacotherapy [128]. Another explanation may be related to the nature of recurrent depression in the general population because studies have suggested that subsequent episodes are shorter in duration than first episodes and have a mean duration of only 3 months [129,130]. This sample was recruited from GPs and is likely to be more similar to a general population sample than to a clinical population recruited from specialist mental health services. In accordance with these epidemiological studies, recurrent depression may be a predictor of shorter episode duration in general population samples. Finally, given that this finding was not fully robust across analyses and was in the opposite direction of most previous results, it may represent a chance finding as a result of random fluctuations in small samples.

    The effect of module completion was ambiguous with the odds ratios indicating a tentative positive effect, but the Bayes factor indicating anecdotal support for no effect. Previous results have been mixed on the association between adherence to treatment and response [56,57,63-65,67-69]. The addition of supportive sessions in this trial may have confounded the effect of module completion and although there was high correlation between completing modules and attending sessions (r=.86), a measure reflecting adherence to both treatment components could have been a more potent predictor.

    Predictors of Poorer Response

    The negative predictive effect of high scores on the Warpy Thoughts Quiz was evident in both models, but was not supported by the Bayes factor, which challenges the robustness of the finding. The Warpy Thoughts Quiz has not been used previously in studies of prediction. It is not entirely equivalent to the much-used Dysfunctional Attitude Scale [131], but taps into many of the same constructs including perfectionism and the need for success, love, and approval [96,132]. Worse treatment response has been associated with higher levels of dysfunctional attitudes in previous studies of face-to-face CBT [29,39,48,49] and some studies of ICBT [59], but not others [50,56]. Dysfunctional attitudes moderated treatment response in one study in which those with severe dysfunctional attitudes responded better to IPT and those with lower levels experienced better effects with CBT [29]. Again, this can indicate that this variable may be a predictor of response to treatment rather than a predictor of general prognosis. A proposed explanation is that patients having less severe dysfunctional attitudes may have greater cognitive flexibility [39] making them more able to profit from utilizing cognitive techniques [29].

    The Role of Expectations and Motivation

    The results were somewhat mixed for treatment expectancy, attitude, and motivation with some analyses indicating a possible negative effect of motivation and expectancy, whereas the Bayes factors indicated substantial support for no effect for all 3 variables. The lack of effects in this study is inconsistent with our hypothesis and with previous studies of face-to-face therapy in which expectancy is considered an important predictor of outcome [29,31]. However, results have been inconclusive with respect to ICBT [60,61]. These results may be due to the fact that most individuals entering a research trial have fairly positive attitudes, expectations, and high motivation, which restricts the range of these variables as is reflected by the distributions displayed in Figure S5 in Multimedia Appendix 1. These variables may be more valuable predictors in a regular practice setting. In addition, these constructs were assessed using single items, which call the validity of these measures into question.


    The findings of the present study indicate that within a population of primary care patients with mild to moderate depression, treatment response to Web-based CBT with face-to-face guidance of high intensity was comparable across varying levels of initial depressive severity and irrespective of the presence and severity of comorbid anxiety. Whether the treatment is suitable for more severe depression is still uncertain. Treatment effects were also comparable for men and women and for patients of various ages. Being married or cohabiting and reporting higher life satisfaction predicted more favorable response to treatment. More positive response was also indicated for individuals with more previous depressive episodes, whereas having a higher level of dysfunctional thinking may predict poorer treatment response.

    The purpose of this paper was primarily exploratory. Therefore, the results must be interpreted as hypotheses to inform further research rather than firm conclusions. Nevertheless, the results add to the knowledge base concerning differential treatment response, knowledge that is crucial for further implementation of Internet-based treatments in regular practice. Future studies should continue to explore predictors and, preferably, moderators of different Internet-based treatments compared to face-to-face treatments. In addition, studies exploring different patterns of response may also give important information about the differential response of various subgroups of patients.


    This study was financially funded by the Research Council of Norway (196423/V50). MoodGYM was provided by the Australian National University and the authors would like to acknowledge the contribution to the project by Professor Kathleen M Griffiths, director at the National Institute for Mental Health Research at ANU. We would also like to thank Anthony Bennett, IT manager of the e-hub, the e-mental health research and development group at the National Institute for Mental Health Research at ANU and Kylie Bennett, Collaborative Research and Development Manager at e-hub, who provided technical expertise in administration of the trial; and Ada Tam, Web Developer at ANU, who provided technical support for users in the trial. Last, but not least, we would like to give a special thanks to those participating in the trial.

    Authors' Contributions

    RSH and MM wrote the paper. MM conducted the statistical analyses and OF contributed to the statistical analyses. All authors contributed to project design, data collection, and/or preparation of the manuscript.

    Conflicts of Interest

    KW and ME contributed in the process of translating MoodGYM into Norwegian. RSH, MM, KL, SKK, NK, and OF have no financial or nonfinancial interests to declare in relation to this study.

    Multimedia Appendix 1

    Supplemental material: Details on statistical methods and additional analyses.

    PDF File (Adobe PDF File), 1MB

    Multimedia Appendix 2

    CONSORT-EHEALTH checklist V1.6.2 [133].

    PDF File (Adobe PDF File), 151KB


    1. Cuijpers P, van Straten A, Andersson G, van Oppen P. Psychotherapy for depression in adults: a meta-analysis of comparative outcome studies. J Consult Clin Psychol 2008 Dec;76(6):909-922. [CrossRef] [Medline]
    2. Churchill R, Hunot V, Corney R, Knapp M, McGuire H, Tylee A, et al. A systematic review of controlled trials of the effectiveness and cost-effectiveness of brief psychological treatments for depression. Health Technol Assess 2001;5(35):1-173 [FREE Full text] [Medline]
    3. Butler AC, Chapman JE, Forman EM, Beck AT. The empirical status of cognitive-behavioral therapy: a review of meta-analyses. Clin Psychol Rev 2006 Jan;26(1):17-31. [CrossRef] [Medline]
    4. Ruwaard J, Schrieken B, Schrijver M, Broeksteeg J, Dekker J, Vermeulen H, et al. Standardized web-based cognitive behavioural therapy of mild to moderate depression: a randomized controlled trial with a long-term follow-up. Cogn Behav Ther 2009 Dec;38(4):206-221. [CrossRef] [Medline]
    5. Andersson G, Bergström J, Holländare F, Carlbring P, Kaldo V, Ekselius L. Internet-based self-help for depression: randomised controlled trial. Br J Psychiatry 2005 Nov;187:456-461 [FREE Full text] [CrossRef] [Medline]
    6. Berger T, Hämmerli K, Gubser N, Andersson G, Caspar F. Internet-based treatment of depression: a randomized controlled trial comparing guided with unguided self-help. Cogn Behav Ther 2011 Dec;40(4):251-266. [CrossRef] [Medline]
    7. Johansson R, Sjöberg E, Sjögren M, Johnsson E, Carlbring P, Andersson T, et al. Tailored vs. standardized internet-based cognitive behavior therapy for depression and comorbid symptoms: a randomized controlled trial. PLoS One 2012;7(5):e36905 [FREE Full text] [CrossRef] [Medline]
    8. Lintvedt OK, Griffiths KM, Sørensen K, Østvik AR, Wang CEA, Eisemann M, et al. Evaluating the effectiveness and efficacy of unguided internet-based self-help intervention for the prevention of depression: a randomized controlled trial. Clin Psychol Psychother 2013;20(1):10-27. [CrossRef] [Medline]
    9. Cuijpers P, Donker T, van Straten A, Li J, Andersson G. Is guided self-help as effective as face-to-face psychotherapy for depression and anxiety disorders? A systematic review and meta-analysis of comparative outcome studies. Psychol Med 2010 Dec;40(12):1943-1957. [CrossRef] [Medline]
    10. Andersson G, Hesser H, Veilord A, Svedling L, Andersson F, Sleman O, et al. Randomised controlled non-inferiority trial with 3-year follow-up of internet-delivered versus face-to-face group cognitive behavioural therapy for depression. J Affect Disord 2013 Dec;151(3):986-994. [CrossRef] [Medline]
    11. Wagner B, Horn AB, Maercker A. Internet-based versus face-to-face cognitive-behavioral intervention for depression: a randomized controlled non-inferiority trial. J Affect Disord 2014 Jan;152-154:113-121. [CrossRef] [Medline]
    12. Vernmark K, Lenndin J, Bjärehed J, Carlsson M, Karlsson J, Öberg J, et al. Internet administered guided self-help versus individualized e-mail therapy: A randomized trial of two versions of CBT for major depression. Behav Res Ther 2010 May;48(5):368-376. [CrossRef] [Medline]
    13. Høifødt RS, Lillevoll KR, Griffiths KM, Wilsgaard T, Eisemann M, Waterloo K, et al. The clinical effectiveness of web-based cognitive behavioral therapy with face-to-face therapist support for depressed primary care patients: randomized controlled trial. J Med Internet Res 2013;15(8):e153 [FREE Full text] [CrossRef] [Medline]
    14. Morris BH, Bylsma LM, Rottenberg J. Does emotion predict the course of major depressive disorder? A review of prospective studies. Br J Clin Psychol 2009 Sep;48(Pt 3):255-273. [CrossRef] [Medline]
    15. Rhebergen D, Beekman AT, Graaf RD, Nolen WA, Spijker J, Hoogendijk WJ, et al. The three-year naturalistic course of major depressive disorder, dysthymic disorder and double depression. J Affect Disord 2009 Jun;115(3):450-459. [CrossRef] [Medline]
    16. Gilchrist G, Gunn J. Observational studies of depression in primary care: what do we know? BMC Fam Pract 2007;8:28 [FREE Full text] [CrossRef] [Medline]
    17. van Beljouw IM, Verhaak PF, Cuijpers P, van Marwijk HW, Penninx BW. The course of untreated anxiety and depression, and determinants of poor one-year outcome: a one-year cohort study. BMC Psychiatry 2010;10:86 [FREE Full text] [CrossRef] [Medline]
    18. Spijker J, de Graaf R, Bijl RV, Beekman AT, Ormel J, Nolen WA. Determinants of persistence of major depressive episodes in the general population. Results from the Netherlands Mental Health Survey and Incidence Study (NEMESIS). J Affect Disord 2004 Sep;81(3):231-240. [CrossRef] [Medline]
    19. Rubenstein LV, Rayburn NR, Keeler EB, Ford DE, Rost KM, Sherbourne CD. Predicting outcomes of primary care patients with major depression: development of a depression prognosis index. Psychiatr Serv 2007 Aug;58(8):1049-1056. [CrossRef] [Medline]
    20. Stegenga BT, Kamphuis MH, King M, Nazareth I, Geerlings MI. The natural course and outcome of major depressive disorder in primary care: the PREDICT-NL study. Soc Psychiatry Psychiatr Epidemiol 2012 Jan;47(1):87-95 [FREE Full text] [CrossRef] [Medline]
    21. Ambresin G, Chondros P, Dowrick C, Herrman H, Gunn JM. Self-rated health and long-term prognosis of depression. Ann Fam Med 2014;12(1):57-65 [FREE Full text] [CrossRef] [Medline]
    22. Riihimäki KA, Vuorilehto MS, Melartin TK, Isometsä ET. Five-year outcome of major depressive disorder in primary health care. Psychol Med 2014 May;44(7):1369-1379. [CrossRef] [Medline]
    23. Dowrick C, Shiels C, Page H, Ayuso-Mateos JL, Casey P, Dalgard OS, et al. Predicting long-term recovery from depression in community settings in Western Europe: evidence from ODIN. Soc Psychiatry Psychiatr Epidemiol 2011 Feb;46(2):119-126. [CrossRef] [Medline]
    24. Skodol AE, Grilo CM, Keyes KM, Geier T, Grant BF, Hasin DS. Relationship of personality disorders to the course of major depressive disorder in a nationally representative sample. Am J Psychiatry 2011 Mar;168(3):257-264 [FREE Full text] [CrossRef] [Medline]
    25. Driessen E, Hollon SD. Cognitive behavioral therapy for mood disorders: efficacy, moderators and mediators. Psychiatr Clin North Am 2010 Sep;33(3):537-555 [FREE Full text] [CrossRef] [Medline]
    26. Kraemer HC, Wilson GT, Fairburn CG, Agras WS. Mediators and moderators of treatment effects in randomized clinical trials. Arch Gen Psychiatry 2002 Oct;59(10):877-883. [Medline]
    27. Simon GE, Perlis RH. Personalized medicine for depression: can we match patients with treatments? Am J Psychiatry 2010 Dec;167(12):1445-1455 [FREE Full text] [CrossRef] [Medline]
    28. Tadić A, Helmreich I, Mergl R, Hautzinger M, Kohnen R, Henkel V, et al. Early improvement is a predictor of treatment outcome in patients with mild major, minor or subsyndromal depression. J Affect Disord 2010 Jan;120(1-3):86-93. [CrossRef] [Medline]
    29. Sotsky SM, Glass DR, Shea MT, Pilkonis PA, Collins JF, Elkin I, et al. Patient predictors of response to psychotherapy and pharmacotherapy: findings in the NIMH Treatment of Depression Collaborative Research Program. Am J Psychiatry 1991 Aug;148(8):997-1008. [CrossRef] [Medline]
    30. Carter JD, Luty SE, McKenzie JM, Mulder RT, Frampton CM, Joyce PR. Patient predictors of response to cognitive behaviour therapy and interpersonal psychotherapy in a randomised clinical trial for depression. J Affect Disord 2011 Feb;128(3):252-261. [CrossRef] [Medline]
    31. Meyer B, Pilkonis PA, Krupnick JL, Egan MK, Simmens SJ, Sotsky SM. Treatment expectancies, patient alliance, and outcome: further analyses from the National Institute of Mental Health Treatment of Depression Collaborative Research Program. J Consult Clin Psychol 2002 Aug;70(4):1051-1055. [Medline]
    32. Lutz W, Stulz N, Köck K. Patterns of early change and their relationship to outcome and follow-up among patients with major depressive disorders. J Affect Disord 2009 Nov;118(1-3):60-68. [CrossRef] [Medline]
    33. Hamilton KE, Dobson KS. Cognitive therapy of depression: pretreatment patient predictors of outcome. Clin Psychol Rev 2002 Jul;22(6):875-893. [Medline]
    34. Van HL, Schoevers RA, Dekker J. Predicting the outcome of antidepressants and psychotherapy for depression: a qualitative, systematic review. Harv Rev Psychiatry 2008;16(4):225-234. [CrossRef] [Medline]
    35. Thase ME, Reynolds CF, Frank E, Simons AD, McGeary J, Fasiczka AL, et al. Do depressed men and women respond similarly to cognitive behavior therapy? Am J Psychiatry 1994 Apr;151(4):500-505. [Medline]
    36. Cuijpers P, Weitz E, Twisk J, Kuehner C, Cristea I, David D, et al. Gender as predictor and moderator of outcome in cognitive behavior therapy and pharmacotherapy for adult depression: an "individual patient data" meta-analysis. Depress Anxiety 2014 Nov;31(11):941-951. [CrossRef] [Medline]
    37. Button KS, Turner N, Campbell J, Kessler D, Kuyken W, Lewis G, et al. Moderators of response to cognitive behavioural therapy as an adjunct to pharmacotherapy for treatment-resistant depression in primary care. J Affect Disord 2015 Mar 15;174:272-280. [CrossRef] [Medline]
    38. Fournier JC, DeRubeis RJ, Shelton RC, Hollon SD, Amsterdam JD, Gallop R. Prediction of response to medication and cognitive therapy in the treatment of moderate to severe depression. J Consult Clin Psychol 2009 Aug;77(4):775-787 [FREE Full text] [CrossRef] [Medline]
    39. Jarrett RB, Eaves GG, Grannemann BD, Rush AJ. Clinical, cognitive, and demographic predictors of response to cognitive therapy for depression: a preliminary report. Psychiatry Res 1991 Jun;37(3):245-260. [Medline]
    40. Barber JP, Muenz LR. The role of avoidance and obsessiveness in matching patients to cognitive and interpersonal psychotherapy: empirical findings from the treatment for depression collaborative research program. J Consult Clin Psychol 1996 Oct;64(5):951-958. [Medline]
    41. Elkin I, Gibbons RD, Shea MT, Sotsky SM, Watkins JT, Pilkonis PA, et al. Initial severity and differential treatment outcome in the National Institute of Mental Health Treatment of Depression Collaborative Research Program. J Consult Clin Psychol 1995 Oct;63(5):841-847. [Medline]
    42. Shapiro DA, Barkham M, Rees A, Hardy GE, Reynolds S, Startup M. Effects of treatment duration and severity of depression on the effectiveness of cognitive-behavioral and psychodynamic-interpersonal psychotherapy. J Consult Clin Psychol 1994 Jun;62(3):522-534. [Medline]
    43. Thase ME, Simons AD, Cahalane J, McGeary J, Harden T. Severity of depression and response to cognitive behavior therapy. Am J Psychiatry 1991 Jun;148(6):784-789. [Medline]
    44. Gelhart RP, King HL. The influence of comorbid risk factors on the effectiveness of cognitive-behavioral treatment of depression. Cognitive and Behavioral Practice 2001 Dec;8(1):18-28. [CrossRef]
    45. Scheibe G, Albus M. Prospective follow-up study lasting 2 years in patients with panic disorder with and without depressive disorders. Eur Arch Psychiatry Clin Neurosci 1994;244(1):39-44. [Medline]
    46. Forand NR, Derubeis RJ. Pretreatment anxiety predicts patterns of change in cognitive behavioral therapy and medications for depression. J Consult Clin Psychol 2013 Oct;81(5):774-782 [FREE Full text] [CrossRef] [Medline]
    47. Smits JA, Minhajuddin A, Thase ME, Jarrett RB. Outcomes of acute phase cognitive therapy in outpatients with anxious versus nonanxious depression. Psychother Psychosom 2012;81(3):153-160 [FREE Full text] [CrossRef] [Medline]
    48. Rude SS, Rehm LP. Response to treatments for depression: The role of initial status on targeted cognitive and behavioral skills. Clinical Psychology Review 1991 Jan;11(5):493-514. [CrossRef]
    49. Blatt SJ, Zuroff DC, Hawley LL, Auerbach JS. Predictors of sustained therapeutic change. Psychother Res 2010 Jan;20(1):37-54. [CrossRef] [Medline]
    50. de Graaf LE, Hollon SD, Huibers MJ. Predicting outcome in computerized cognitive behavioral therapy for depression in primary care: A randomized trial. J Consult Clin Psychol 2010 Apr;78(2):184-189. [CrossRef] [Medline]
    51. Andersson G, Bergström J, Holländare F, Ekselius L, Carlbring P. Delivering cognitive behavioural therapy for mild to moderate depression via the Internet: predicting outcome at 6-month follow-up. Verhaltenstherapie 2004;14(3):185-189. [CrossRef]
    52. Spek V, Nyklícek I, Cuijpers P, Pop V. Predictors of outcome of group and internet-based cognitive behavior therapy. J Affect Disord 2008 Jan;105(1-3):137-145. [CrossRef] [Medline]
    53. Button KS, Wiles NJ, Lewis G, Peters TJ, Kessler D. Factors associated with differential response to online cognitive behavioural therapy. Soc Psychiatry Psychiatr Epidemiol 2012 May;47(5):827-833. [CrossRef] [Medline]
    54. Moritz S, Schilling L, Hauschildt M, Schröder J, Treszl A. A randomized controlled trial of internet-based therapy in depression. Behav Res Ther 2012 Aug;50(7-8):513-521. [CrossRef] [Medline]
    55. Bower P, Kontopantelis E, Sutton A, Kendrick T, Richards DA, Gilbody S, et al. Influence of initial severity of depression on effectiveness of low intensity interventions: meta-analysis of individual patient data. BMJ 2013;346:f540 [FREE Full text] [Medline]
    56. Warmerdam L, Van Straten A, Twisk J, Cuijpers P. Predicting outcome of Internet-based treatment for depressive symptoms. Psychother Res 2013;23(5):559-567. [CrossRef] [Medline]
    57. Farrer LM, Griffiths KM, Christensen H, Mackinnon AJ, Batterham PJ. Predictors of adherence and outcome in Internet-based cognitive behavior therapy delivered in a telephone counseling setting. Cogn Ther Res 2013 Oct 27;38(3):358-367. [CrossRef]
    58. Clarke G, Kelleher C, Hornbrook M, Debar L, Dickerson J, Gullion C. Randomized effectiveness trial of an Internet, pure self-help, cognitive behavioral intervention for depressive symptoms in young adults. Cogn Behav Ther 2009;38(4):222-234 [FREE Full text] [CrossRef] [Medline]
    59. Donker T, Batterham PJ, Warmerdam L, Bennett K, Bennett A, Cuijpers P, et al. Predictors and moderators of response to internet-delivered Interpersonal psychotherapy and cognitive behavior therapy for depression. J Affect Disord 2013 Oct;151(1):343-351. [CrossRef] [Medline]
    60. de Graaf LE, Huibers MJ, Riper H, Gerhards SA, Arntz A. Use and acceptability of unsupported online computerized cognitive behavioral therapy for depression and associations with clinical outcome. J Affect Disord 2009 Aug;116(3):227-231. [CrossRef] [Medline]
    61. Cavanagh K, Shapiro DA, Van Den Berg S, Swain S, Barkham M, Proudfoot J. The acceptability of computer-aided cognitive behavioural therapy: a pragmatic study. Cogn Behav Ther 2009;38(4):235-246. [CrossRef] [Medline]
    62. Meyer B, Berger T, Caspar F, Beevers CG, Andersson G, Weiss M. Effectiveness of a novel integrative online treatment for depression (Deprexis): randomized controlled trial. J Med Internet Res 2009;11(2):e15 [FREE Full text] [CrossRef] [Medline]
    63. Hilvert-Bruce Z, Rossouw PJ, Wong N, Sunderland M, Andrews G. Adherence as a determinant of effectiveness of internet cognitive behavioural therapy for anxiety and depressive disorders. Behav Res Ther 2012 Aug;50(7-8):463-468. [CrossRef] [Medline]
    64. Christensen H, Griffiths K, Groves C, Korten A. Free range users and one hit wonders: community users of an Internet-based cognitive behaviour therapy program. Aust N Z J Psychiatry 2006 Jan;40(1):59-62. [CrossRef] [Medline]
    65. Christensen H, Griffiths KM, Korten A. Web-based cognitive behavior therapy: analysis of site usage and changes in depression and anxiety scores. J Med Internet Res 2002;4(1):e3 [FREE Full text] [CrossRef] [Medline]
    66. Warmerdam L, van Straten A, Twisk J, Riper H, Cuijpers P. Internet-based treatment for adults with depressive symptoms: randomized controlled trial. J Med Internet Res 2008;10(4):e44 [FREE Full text] [CrossRef] [Medline]
    67. Clarke G, Eubanks D, Reid E, Kelleher C, O'Connor E, DeBar LL, et al. Overcoming Depression on the Internet (ODIN) (2): a randomized trial of a self-help depression skills program with reminders. J Med Internet Res 2005;7(2):e16 [FREE Full text] [CrossRef] [Medline]
    68. Clarke G, Reid E, Eubanks D, O'Connor E, DeBar LL, Kelleher C, et al. Overcoming Depression on the Internet (ODIN): a randomized controlled trial of an Internet depression skills intervention program. J Med Internet Res 2002 Dec;4(3):E14 [FREE Full text] [CrossRef] [Medline]
    69. Donkin L, Christensen H, Naismith SL, Neal B, Hickie IB, Glozier N. A systematic review of the impact of adherence on the effectiveness of e-therapies. J Med Internet Res 2011;13(3):e52 [FREE Full text] [CrossRef] [Medline]
    70. Davidsen A. Experiences of carrying out talking therapy in general practice: a qualitative interview study. Patient Educ Couns 2008 Aug;72(2):268-275. [CrossRef] [Medline]
    71. Backenstrass M, Joest K, Rosemann T, Szecsenyi J. The care of patients with subthreshold depression in primary care: is it all that bad? A qualitative study on the views of general practitioners and patients. BMC Health Serv Res 2007;7:190 [FREE Full text] [CrossRef] [Medline]
    72. Mykletun A, Knudsen AK, Tangen T, Øverland S. General practitioners' opinions on how to improve treatment of mental disorders in primary health care. Interviews with one hundred Norwegian general practitioners. BMC Health Serv Res 2010;10:35 [FREE Full text] [CrossRef] [Medline]
    73. Wittchen HU, Jacobi F, Rehm J, Gustavsson A, Svensson M, Jönsson B, et al. The size and burden of mental disorders and other disorders of the brain in Europe 2010. Eur Neuropsychopharmacol 2011 Sep;21(9):655-679. [CrossRef] [Medline]
    74. Australian National University, Centre for Mental Health Research. The MoodGYM training program: mark III: learn cognitive behaviour therapy skills for preventing and coping with depression. 2008.   URL: [accessed 2013-05-10] [WebCite Cache]
    75. Bennett K, Reynolds J, Christensen H, Griffiths KM. e-hub: an online self-help mental health service in the community. Med J Aust 2010 Jun 7;192(11 Suppl):S48-S52. [Medline]
    76. Calear AL, Christensen H, Mackinnon A, Griffiths KM, O'Kearney R. The YouthMood Project: a cluster randomized controlled trial of an online cognitive behavioral program with adolescents. J Consult Clin Psychol 2009 Dec;77(6):1021-1032. [CrossRef] [Medline]
    77. Powell J, Hamborg T, Stallard N, Burls A, McSorley J, Bennett K, et al. Effectiveness of a web-based cognitive-behavioral tool to improve mental well-being in the general population: randomized controlled trial. J Med Internet Res 2013;15(1):e2 [FREE Full text] [CrossRef] [Medline]
    78. Christensen H, Griffiths KM, Jorm AF. Delivering interventions for depression by using the internet: randomised controlled trial. BMJ 2004 Jan 31;328(7434):265 [FREE Full text] [CrossRef] [Medline]
    79. Farrer L, Christensen H, Griffiths KM, Mackinnon A. Internet-based CBT for depression with and without telephone tracking in a national helpline: randomised controlled trial. PLoS One 2011;6(11):e28099 [FREE Full text] [CrossRef] [Medline]
    80. Beck A, Steer R, Brown G. BDI-II, Beck Depression Inventory: Manual. San Antonio, TX: The Psychological Corporation; 1996.
    81. Arnau RC, Meagher MW, Norris MP, Bramson R. Psychometric evaluation of the Beck Depression Inventory-II with primary care medical patients. Health Psychol 2001 Mar;20(2):112-119. [Medline]
    82. Dozois DJA, Dobson KS, Ahnberg JL. A psychometric evaluation of the Beck Depression Inventory-II. Psychological Assessment 1998;10(2):83-89. [CrossRef]
    83. Bjelland I, Dahl AA, Haug TT, Neckelmann D. The validity of the Hospital Anxiety and Depression Scale. An updated literature review. J Psychosom Res 2002 Feb;52(2):69-77. [Medline]
    84. Zigmond AS, Snaith RP. The hospital anxiety and depression scale. Acta Psychiatr Scand 1983 Jun;67(6):361-370. [Medline]
    85. Beck A, Steer R. Beck Anxiety Inventory: Manual. San Antonio, TX: The Psychological Corporation; 1993.
    86. Fydrich T, Dowdall D, Chambless DL. Reliability and validity of the Beck Anxiety Inventory. Journal of Anxiety Disorders 1992 Jan;6(1):55-61. [CrossRef]
    87. Steer RA, Ranieri WF, Beck AT, Clark DA. Further evidence for the validity of the beck anxiety inventory with psychiatric outpatients. Journal of Anxiety Disorders 1993 Jul;7(3):195-205. [CrossRef]
    88. Beck AT, Epstein N, Brown G, Steer RA. An inventory for measuring clinical anxiety: psychometric properties. J Consult Clin Psychol 1988 Dec;56(6):893-897. [Medline]
    89. Sheehan DV, Lecrubier Y, Sheehan KH, Amorim P, Janavs J, Weiller E, et al. The Mini-International Neuropsychiatric Interview (M.I.N.I.): the development and validation of a structured diagnostic psychiatric interview for DSM-IV and ICD-10. J Clin Psychiatry 1998;59 Suppl 20:22-33;quiz 34. [Medline]
    90. Saunders JB, Aasland OG, Babor TF, de la Fuente JR, Grant M. Development of the Alcohol Use Disorders Identification Test (AUDIT): WHO Collaborative Project on Early Detection of Persons with Harmful Alcohol Consumption--II. Addiction 1993 Jun;88(6):791-804. [Medline]
    91. Reinert DF, Allen JP. The alcohol use disorders identification test: an update of research findings. Alcohol Clin Exp Res 2007 Feb;31(2):185-199. [CrossRef] [Medline]
    92. EuroQol Group. EQ-5D. 2013. About EQ-5D   URL: [accessed 2013-05-08] [WebCite Cache]
    93. Diener E, Emmons RA, Larsen RJ, Griffin S. The Satisfaction With Life Scale. J Pers Assess 1985 Feb;49(1):71-75. [CrossRef] [Medline]
    94. Pavot W, Diener E. Review of the Satisfaction With Life Scale. Psychological Assessment 1993;5(2):164-172. [CrossRef]
    95. Glaesmer H, Grande G, Braehler E, Roth M. The German version of the Satisfaction With Life Scale (SWLS). European Journal of Psychological Assessment 2011 May 3;27(2):127-132. [CrossRef]
    96. Christensen H, Griffiths K, Groves C. MoodGYM Training Program: Clinician’s Manual: A Manual for the Use of the MoodGYM Training Program in Clinical Practice. Canberra: Centre for Mental Health Research; 2004.
    97. Jorm AF, Medway J, Christensen H, Korten AE, Jacomb PA, Rodgers B. Attitudes towards people with depression: effects on the public's help-seeking and outcome when experiencing common psychiatric symptoms. Aust N Z J Psychiatry 2000 Aug;34(4):612-618. [Medline]
    98. Parslow RA, Christensen H, Griffiths KM, Groves C. The Warpy Thoughts Scale: a new 20-item instrument to measure dysfunctional attitudes. Cogn Behav Ther 2006;35(2):106-116. [CrossRef] [Medline]
    99. Schwarzer R, Jerusalem M. Generalized Self-Efficacy Scale. In: Weinman J, Wright S, Johnston M, editors. Measures in Health Psychology: A User’s Portfolio Causal and Control Beliefs. Windsor, UK: NFER-NELSON; 1995:35-37.
    100. Leganger A, Kraft P, R⊘ysamb E. Perceived self-efficacy in health behaviour research: conceptualisation, measurement and correlates. Psychology & Health 2000 Feb;15(1):51-69. [CrossRef]
    101. Luszczynska A, Gutiérrez‐Doña B, Schwarzer R. General self‐efficacy in various domains of human functioning: evidence from five countries. International Journal of Psychology 2005 Apr;40(2):80-89. [CrossRef]
    102. Luszczynska A, Scholz U, Schwarzer R. The general self-efficacy scale: multicultural validation studies. J Psychol 2005 Sep;139(5):439-457. [CrossRef] [Medline]
    103. Scholz U, Gutiérrez Doña B, Sud S, Schwarzer R. Is general self-efficacy a universal construct? European Journal of Psychological Assessment 2002 Sep 1;18(3):242-251. [CrossRef]
    104. Kruschke J. Doing Bayesian Data Analysis: A Tutorial with R and BUGS. Oxford: Academic Press; 2011.
    105. Gelman A, Carlin JB, Stern HS, Dunson DB, Vehtari A, Rubin DB. Bayesian Data Analysis. 3rd edition. Boca Raton, FL: Chapman & Hall/CRC Press; 2013.
    106. Wagenmakers E. A practical solution to the pervasive problems of p values. Psychon Bull Rev 2007 Oct;14(5):779-804. [Medline]
    107. Wagenmakers E, Wetzels R, Borsboom D, van der Maas HL. Why psychologists must change the way they analyze their data: the case of psi: comment on Bem (2011). J Pers Soc Psychol 2011 Mar;100(3):426-432. [CrossRef] [Medline]
    108. Goodman SN. Toward evidence-based medical statistics. 1: The P value fallacy. Ann Intern Med 1999 Jun 15;130(12):995-1004. [Medline]
    109. Kruschke JK. Bayesian estimation supersedes the t test. J Exp Psychol Gen 2013 May;142(2):573-603. [CrossRef] [Medline]
    110. Rouder JN, Speckman PL, Sun D, Morey RD, Iverson G. Bayesian t tests for accepting and rejecting the null hypothesis. Psychon Bull Rev 2009 Apr;16(2):225-237. [CrossRef] [Medline]
    111. Jeffreys H. Theory of probability. Oxford: Oxford University Press; 1961.
    112. Wetzels R, Matzke D, Lee MD, Rouder JN, Iverson GJ, Wagenmakers E. Statistical evidence in experimental psychology: an empirical comparison using 855 t tests. Perspectives on Psychological Science 2011 May 18;6(3):291-298. [CrossRef]
    113. Gelman A, Shalizi CR. Philosophy and the practice of Bayesian statistics. Br J Math Stat Psychol 2013 Feb;66(1):8-38. [CrossRef] [Medline]
    114. Plummer M. JAGS: a program for analysis of Bayesian graphical models using Gibbs sampling. 2003 Presented at: Proceedings of the 3rd International Workshop on Distributed Statistical Computing; March 20–22, 2003; Vienna, Austria.
    115. Gelman A, Rubin D. Inference from iterative simulation using multiple sequences. Stat Sci 1992;7(4):457-472.
    116. Jung T, Wickrama KA. An introduction to latent class growth analysis and growth mixture modeling. Social Pers Psych Compass 2008 Jan;2(1):302-317. [CrossRef]
    117. Tofighi D, Enders CK. Identifying the correct number of classes in growth mixture models. In: Hancock GR, Samuelsen KM, editors. Advances in Latent Variable Mixture Models. Greenwich, CT: Information Age Publishing; 2007.
    118. Muthén B. Latent variable analysis: growth mixture modeling and related techniques for longitudinal data. In: Kaplan D, editor. The Sage Handbook of Quantitative Methodology for the Social Sciences. Thousand Oaks, CA: SAGE Publications; 2004:345-368.
    119. Stulz N, Lutz W, Leach C, Lucock M, Barkham M. Shapes of early change in psychotherapy under routine outpatient conditions. J Consult Clin Psychol 2007 Dec;75(6):864-874. [CrossRef] [Medline]
    120. Muthén B, Brown HC, Hunter AM, Cook IA, Leuchter AF. General approaches to analysis of course: applying growth mixture modeling to randomized trials of depression medication. In: Shrout PE, Keyes K, Ornstein K, editors. Causality and Psychopathology: Finding the Determinants of Disorders and their Cures. New York: Oxford University Press; 2010:159-177.
    121. Cuijpers P, Reynolds CF, Donker T, Li J, Andersson G, Beekman A. Personalized treatment of adult depression: medication, psychotherapy, or both? A systematic review. Depress Anxiety 2012 Oct;29(10):855-864. [CrossRef] [Medline]
    122. Gelman A, Carlin J. Beyond power calculations: assessing type S (sign) and type M (magnitude) errors. Perspectives on Psychological Science 2014 Nov 17;9(6):641-651. [CrossRef]
    123. Köhler S, Unger T, Hoffmann S, Steinacher B, Fydrich T. Acute and long-term treatment outcome in depressed inpatients with vs. without anxious features: results of a one-year follow-up study. J Affect Disord 2013 Sep 25;150(3):1055-1061. [CrossRef] [Medline]
    124. Persons J, Bostrom A, Bertagnolli A. Results of randomized controlled trials of cognitive therapy for depression generalize to private practice. Cognit Ther Res 1999;23(5):535-548. [CrossRef]
    125. Wilhelmsen M, Lillevoll K, Risør MB, Høifødt R, Johansen M, Waterloo K, et al. Motivation to persist with internet-based cognitive behavioural treatment using blended care: a qualitative study. BMC Psychiatry 2013;13:296 [FREE Full text] [CrossRef] [Medline]
    126. Dudek D, Rybakowski JK, Siwek M, Pawłowski T, Lojko D, Roczeń R, et al. Risk factors of treatment resistance in major depression: association with bipolarity. J Affect Disord 2010 Oct;126(1-2):268-271. [CrossRef] [Medline]
    127. Bennabi D, Aouizerate B, El-Hage W, Doumy O, Moliere F, Courtet P, et al. Risk factors for treatment resistance in unipolar depression: a systematic review. J Affect Disord 2015 Jan 15;171:137-141. [CrossRef] [Medline]
    128. Cuijpers P, Dekker J, Hollon SD, Andersson G. Adding psychotherapy to pharmacotherapy in the treatment of depressive disorders in adults: a meta-analysis. J Clin Psychiatry 2009 Sep;70(9):1219-1229. [CrossRef] [Medline]
    129. Spijker J, de Graaf R, Bijl RV, Beekman ATF, Ormel J, Nolen WA. Duration of major depressive episodes in the general population: results from The Netherlands Mental Health Survey and Incidence Study (NEMESIS). Br J Psychiatry 2002 Sep;181:208-213 [FREE Full text] [Medline]
    130. Eaton WW, Anthony JC, Gallo J, Cai G, Tien A, Romanoski A, et al. Natural history of Diagnostic Interview Schedule/DSM-IV major depression. The Baltimore Epidemiologic Catchment Area follow-up. Arch Gen Psychiatry 1997 Nov;54(11):993-999. [Medline]
    131. Weissman A, Beck A. Development and validation of the Dysfunctional Attitude Scale: a preliminary investigation. 1978 Presented at: Annual Meeting of the American Educational Research Association (62nd); March 27-31, 1978; Toronto, ON.
    132. Chioqueta AP, Stiles TC. Factor structure of the Dysfunctional Attitude Scale (Form A) and the Automatic Thoughts Questionnaire: an exploratory study. Psychol Rep 2006 Aug;99(1):239-247. [CrossRef] [Medline]
    133. Eysenbach G, Consort- E. CONSORT-EHEALTH: improving and standardizing evaluation reports of Web-based and mobile health interventions. J Med Internet Res 2011;13(4):e126 [FREE Full text] [CrossRef] [Medline]


    ANU: Australian National University
    AUDIT: Alcohol Use Disorders Identification Test
    BAI: Beck Anxiety Inventory
    BDI-II: Beck Depression Inventory-II
    BF: Bayes factor
    CBT: cognitive behavioral therapy
    EQ-5D: EuroQol 5-Dimension Self-Report Questionnaire
    GP: general practitioner
    GSE: General Self-efficacy Scale
    HADS-A: Hospital Anxiety and Depression Scale-Anxiety subscale
    HADS-D: Hospital Anxiety and Depression Scale-Depression subscale
    HDI: highest density interval
    ICBT: Internet-based cognitive behavioral therapy
    IPT: interpersonal therapy
    MCMC: Markov chain Monte Carlo
    MDE: major depressive episode
    NHST: null-hypothesis significance testing
    Presp: probability of response
    RCT: randomized controlled trial
    SWLS: Satisfaction With Life Scale

    Edited by G Eysenbach; submitted 17.02.15; peer-reviewed by K Button, B Meyer; comments to author 12.03.15; revised version received 08.05.15; accepted 08.07.15; published 02.09.15

    ©Ragnhild Sørensen Høifødt, Matthias Mittner, Kjersti Lillevoll, Susanne Kvam Katla, Nils Kolstrup, Martin Eisemann, Oddgeir Friborg, Knut Waterloo. Originally published in the Journal of Medical Internet Research (, 02.09.2015.

    This is an open-access article distributed under the terms of the Creative Commons Attribution License (, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on, as well as this copyright and license information must be included.