The inclusion criteria are as follows: (1) individuals 18-60 years of age; (2) BDI-II of at least mild severity (score ≥14) with no upper severity limit [23]; (3) MINI (Mini-International Neuropsychiatric Interview)-confirmed diagnosis of MDD [24]; and (4) fluency in English. All participants are diagnosed by a CAMH staff psychiatrist with diagnosis confirmed via the MINI [24], administered at the screening visit.

The exclusion criteria are as follows: (1) individuals currently receiving weekly structured psychotherapy; (2) individuals who meet DSM-V (Diagnostic and Statistical Manual of Mental Disorders, 5th Edition) criteria for severe alcohol or substance use disorder (in the past 3 months), borderline personality disorder, schizophrenia, or any other primary psychotic disorder, bipolar disorder, or obsessive-compulsive disorder; (3) individuals who manifest clinically significant suicidal ideation defined as imminent intent or attempted suicide (in the past 6 months); and (4) individuals who are judged to have treatment-resistant depression, as defined by failure in at least 2 trials of antidepressant medications or a course of psychotherapy during the current depressive episode [25].

The online group CBT-M program combines online-accessed workbooks with phone-based navigator coaching that assists with the coordination and support of multiple software interactions (eg, secure text messaging, Fitbit tracked walking, and food monitoring via photography). Each participant is loaned a Fitbit HR Charge 4 (Fitbit Inc), which assesses physical steps and 24-hour heart rate, averaged in 5-second intervals. Intervention content builds on 6 online RCTs with students [26-32], 1 RCT with adults with type 2 diabetes [11,33,34] (where significant mental health and blood glucose benefits resulted), and 1 online RCT with youth diagnosed with MDD [22]. The content contains 24 workbook chapters on multiple topics derived from the focus group study (eg, Living By Your Truths, Overcoming Wired-ness and Tired-ness, Mindfulness and Relationships, Loss and Grief, Resilience, Befriending Ourselves, Befriending Your Body with Exercise, Body Image, Intimacy, Forgiveness, Overcoming Procrastination, Dealing with Negative Moods, Stress Resilience, Reducing Performance Anxiety, and Cultivating Inspiration), addressed in sequences mutually agreed on by the participants and navigator-coaches.

The online CBT-M groups address structured tasks that elevate mood and decrease anxiety. Depressive social withdrawals are identified and reflected on while corrective emotional experiences are elicited, structured, and supported. Healthy internet interactions are reinforced by the web-based program content (24 workbooks and 56 videos), with priorities guided by participant interests. While internet-based contacts are sometimes distracting “escapes” (eg, Netflix movies and video game playing), they can be productive, socially engaging events (eg, TED Talks, CBT discussion, and mindfulness instructions) where role-modeling coordinates with learning. Group members stimulate and reinforce health behavior adoptions (exercise, CBT, mindfulness, sleep hygiene, diet, and text message encouragement) and transformations from distractive media fixations to positive learning and enjoyment. Internet-based contacts integrate with face-to-face professional and familial contacts and social intimacy development. Group online CBT-M tasks emphasize applying CBT-M methods in accord with content themes, which are decatastrophizing, overcoming perfectionism, sound sleep, self-befriending, courage development, forgiveness, interpersonal mindfulness, and autonomous generosity. This progression coordinates physical and cognitive changes that integrate interpersonal contacts with self-management. The tracking of walking exercise and heart rate (HR), on a 24-hour basis, captures HR elevations that reflect episodes of anxiety and negative affect (elevated HR with minimal movement). There is selective sharing of Fitbit bio-behavioral data with designated staff and group members. Reduced HR often accompanies mindfulness practice while mildly elevated HR is closely associated with walking (ie, movement-affected HR). Self-modification goals involve improving autonomic nervous system balance through appropriate exercise, improved sleep hygiene, and restorative sleep [35].

The on-site, usual-care CBT group follows the structure of the “Mind Over Mood” workbook in reviewing CBT concepts and procedures [36]. A series of worksheets assist participants in differentiating moods from thoughts and situational influences, leading to modifications of thinking, behavior, emotion, and mood.

Difficulties with cognitive change are assumed to be associated with self-acceptance deficits and negative mood dominance. Increased self-acceptance is linked with mindfulness experiences, which involve “more focusing on the present moment” and on nonjudgmental acceptance of ongoing experience. Mindful breathing awareness is described in content on “balanced deep breathing” and in tension releases achieved via progressive muscle relaxation practice [37].

An emphasized structure is the automatic thought record by which negative, disturbing thoughts are identified, and attention directed to alternate thoughts evaluated as clearer and less distorted. These transitions are more difficult when immersions in negative moods diminish the confidence in finding a “better” thought. Accordingly, self-efficacy is an important goal, emphasized in exchanges of support between group members and in enhanced experiences of self-acceptance.

Behavioral activations are structured for the development of increased awareness and observations of how self-control resources elicit gratifying activities. On the basis of behavioral activation practices, cognitive restructuring strategies utilize the automatic thought record or experiential approximations of the automatic thought record [38].

Another major component of group CBT-M at CAMH involves examinations of core beliefs where multiple techniques are employed (eg, self-observation logs and historical self-reviews) in the group setting and in individual homework assignments between sessions [38].

The primary outcome measure is the BDI-II [23]. It is a 21-item self-report scale, measuring symptoms of depression with a 4-point Likert scale (0 to 3), with higher scores indicating more severe depressive symptoms. The customary cut-off scores for mild depression are 14 to 25, moderate depression 26 to 30, and severe depression 31 and greater.

The secondary outcomes assess the following: anxiety (Beck Anxiety Inventory) [39]; depression (QIDS [Quick Inventory of Depressive Symptomatology] [40] and HDRS-17 [17-item Hamilton Depression Rating Scale], consisting of an interview with a blinded-assessor rater) [41]; mindfulness (FFMQ [Five-Facet Mindfulness Questionnaire]) [42]; quality of life (EQ-5D [European Quality of Life Five Dimension]) [43,44]; patient costs (health care cost diary for major depression) [45]; and pain (brief pain inventory) [46].

All self-report measures and the HDRS-17 interview are carried out at the same CAMH mood and anxiety research clinic in identical assessment rooms. The HDRS-17 interviewer-assessor is blinded to intervention and control conditions for the trial duration and is experienced in the blinded-assessor role with similar conditions conducted at the identical study site used in this trial [22].

The primary analysis will be a Bayesian analysis of covariance (ANCOVA), with change in BDI-II between baseline and 16 weeks as the outcome and 2 covariates, the baseline score and the intervention group variable, coded as “O” (for online) and “I” (for in-office). We refer to the parameter of interest as ∆, the baseline-adjusted difference between the 2 groups in the change in BDI-II from baseline to 16 weeks, coded such that ∆<0 means that the online group has a smaller decrease in the level of depression over the course of the study. Then, if ∆>-3, the improvement in the online group is at most 3 points worse than the improvement in the in-office group, and the online group is noninferior to the in-office group. Using noninformative priors for all study parameters in the ANCOVA model, we will compute and plot the full posterior distribution of ∆, presenting the lower 95% credible interval for ∆, and then calculating the posterior probability that ∆>-3; in other words, the probability (after observing the study data) that online CBT is noninferior, according to our definition. One advantage of a Bayesian approach is that it allows the assigning of posterior probabilities of noninferiority at other margins; notably, other margins near 3 may have equal levels of evidentiary support. For example, without a penalty for multiple testing, we can compute the probability that ∆>-2 or ∆>-4. Another advantage of the Bayesian approach is that it involves a more useful presentation of results than a simple confidence interval or P value. Finally, the outputs of the Bayesian model can be used as probabilistic inputs for the economic analysis. Since this is a noninferiority approach, the primary analysis will be per protocol, which, in the presence of intervention nonadherence, is more conservative than intention-to-treat (ie, less likely to conclude that groups are similar when they are not). Three secondary analyses of the primary outcome will be conducted. First, we will make the above between-group comparison of the changes in BDI-II scores from baseline to each follow-up time point. Secondly, we will perform an intention-to-treat analysis, including outcomes on participants who were nonadherent to their program (standard care CBT vs CBT online) using established cut-offs. Finally, to assess for sensitivity to random imbalance in baseline characteristic, the following covariates will be added to the ANCOVA model: age, baseline anxiety, baseline depression (HDRS-17), baseline pain, and baseline mindfulness. If more than 5% of the participants included in the per protocol are missing outcome data, that analysis will use multiple imputations within the Bayesian model with full baseline covariate data used to impute missing outcomes. In a second approach, we will assume that missing follow-up data are missing not at random and replace missing values by values randomly sampled from the upper half (high BDI-II) of their predictive distribution. This represents the assumption that those who have no follow-up BDI-II have values lower than the median predicted by their baseline characteristics. This will be repeated within the intention-to-treat analysis, where missing outcome data will be more common.

A similar modeling approach will be used for each secondary outcome (Beck Anxiety Inventory, QIDS, and HDRS-24), mindfulness (FFMQ [Five-Facet Mindfulness Questionnaire]), pain (brief pain inventory), health care cost diary for major depression, and quality of life (EQ-5D). While noninferiority margins per outcome are nonexistent, we will present probabilities for a range of margins based on fractions of the minimally clinical important difference for these scales (where available) or fractions of a standard deviation. All primary and secondary outcomes will be analyzed as continuous variables, with outcomes modelled as normal, t, rescaled beta, log-normal, or gamma distributions, whichever is most appropriate.

We will examine BDI-II changes over time across the 2 intervention delivery models during a 6-month post-intervention period. To give a quantitative overview of changes between and within groups over 6 months, all study outcomes will be assessed using linear mixed effects models incorporating within-person correlation. Mixed effects models allow for including partial data (eg, from participants with missed visits or who dropped out prematurely) while accounting for repeated within-person data [47]. The baseline and follow-up scores (up to 10 months from baseline) will form the dependent variable and intervention group, time, and the time-treatment interactions will be the independent variables. Within-person correlation of residuals will be handled by an AR1 error structure. Through the examination of estimated model parameters and their 95% confidence intervals, we can assess whether changes over time are similar between groups and whether differences between groups at any given time are clinically important.