The Institutional Review Boards for the University of Illinois at Chicago and the Stanford University approved the study. All participants provided written informed consent. Protocols of the RAINBOW and ENGAGE trials were previously published (23, 24).
The ENGAGE sample (n=108) was a subset of the RAINBOW sample (n=409) who also consented to the ENGAGE trial (Figure 1). Participants were recruited from family and internal medicine departments in 4 medical centers of Sutter Health’s Palo Alto Medical Foundation. Adults of any sex and race/ethnicity were eligible if they had body mass index (BMI) ≥30 (≥27 if Asian) and depressive symptoms defined by a 9-item Patient Health Questionnaire (PHQ-9) score ≥10, without serious medical or psychiatric comorbidities or other exclusions.
Potentially eligible patients identified from the electronic health records (EHR) received recruitment invitations to complete initial eligibility screening. Patients passing the screening attended baseline visits to complete a physical exam and questionnaires and separate fMRI visits. A study physician performed final EHR reviews to rule out any comorbidities that might have excluded the participant. If fully eligible, patients were randomized to receive the study intervention or usual care and were followed up at 2 (end of the PST-only component of the intervention), 6 (end of the intensive treatment phase of the intervention), and 12 months (end of the maintenance phase of the intervention). Funding of the ENGAGE study began after about half of the RAINBOW sample had been enrolled, and subsequent participants were offered the options to enroll in RAINBOW only or in RAINBOW and ENGAGE during the informed consent.
Randomization and blinding
RAINBOW participants (including those who also enrolled in ENGAGE) were randomly assigned to receive the Integrated Coaching for Better Mood and Weight (I-CARE) intervention or usual care using an online system (25) based on Pocock’s covariate-adaptive minimization (26). This method was used to achieve better-than-chance marginal balance across multiple baseline characteristics: clinic, age, sex, race/ethnicity, education, BMI, depression Symptom Checklist 20-items (SCL-20) score, antidepressant medications, and number of hospitalizations in the past year. After the ENGAGE study began, participants’ choice of enrolling in RAINBOW only or in RAINBOW and ENGAGE was added as a balancing factor. Investigators not involved in delivery of the study intervention, the data and safety monitoring board, outcome assessors, and the data analyst were blinded to participants’ treatment assignment until the end of the study.
The I-CARE intervention combines the GLB (27) program for weight loss and the PEARLS program for depression care management (28, 29). The GLB program is self-directed using home-viewed GLB videos and promotes modest (5-10%) weight loss through calorie-reduced, healthy dietary changes and gradually increased physical activity. The PEARLS program uses PST with behavioral activation strategies as first-line, plus stepwise increases in doses and number of antidepressant medications as needed. The 12-month intervention included 6 in-person individual PST sessions for depression by 2 months, 3 additional PST sessions for depression and 11 home videos on lifestyle changes for weight loss by 6 months, and 6 maintenance calls by 12 months. Participants received self-care materials, including intervention handouts, a DVD set or online access code for the GLB videos, a wireless activity tracker with replacement batteries, and written instructions for creating a MyFitnessPal.com account to track weight and dietary intake. A trained health coach delivered the in-person PST sessions and met every 1-2 weeks with an intervention manager, a psychiatrist, and a primary care physician to discuss participants with poor progress. Based on these discussions, the psychiatrist might recommend initiating or adjusting antidepressant medications for a patient and would communicate with the patient’s primary care physician who was responsible for medication management.
Participants in the intervention and usual care groups continued to receive medical care from their primary care physicians. All participants also received information on mental health services and weight management and other wellness programs routinely available at their clinic. Control participants received a wireless activity tracker with batteries but not any other intervention materials.
Assessments of BMI, dietary intake, and physical activity occurred at baseline, 6, and 12 months. Following standardized protocols (30), trained research staff measured height (baseline only) and body weight. BMI was calculated as weight (kg) divided by height squared (m2).
Trained staff interviewed participants to conduct multiple-pass 24-hour recalls using the Nutrition Data System for Research (NDSR) (31) and 7-day Physical Activity Recalls per standardized protocol (32, 33). As an index of overall diet quality, the DASH (Dietary Approach to Stop Hypertension) score was computed based on 9 nutrient targets (total fat, saturated fat, protein, cholesterol, fiber, magnesium, calcium, sodium and potassium) (34) provided by the NDSR software. For each nutrient target, participants were assigned a point if they achieved the target and half a point if they achieved an intermediate target (i.e., half-way between the DASH target and the population mean), and the DASH score was the sum of points across all 9 nutrients (35, 36). Additional daily dietary target measures included fruits and vegetables (servings), total caloric intake (kilocalories), and total fat consumption (grams).
Physical activity was assessed as Metabolic Equivalent Task (MET) minutes per week of leisure time physical activity based on the sum of the weighted physical activity minutes for moderate (weight: 4 METs), hard (weight: 6 METs), and very hard (weight: 10 METs) activities from the 7-day Physical Activity Recall (37, 38). Also, total energy expenditure in kilocalories per kilogram per day was derived from MET-minutes/day using the conversion 1 MET = 1 kcal/kg/hour (37, 38).
Potential mediators: neural circuit targets
fMRI data were collected at baseline and 2 months. Neural circuit targets were defined and quantified by the following procedures:
- i) Sequences
We implemented previously established functional neuroimaging sequences and parameters as defined in our pre-specified protocol ((23); see Additional File 1 for details).
- ii) Paradigms
Neural circuits of interest, anchored in the amygdala and ACC, were engaged using established experimental paradigms in which participants were instructed to view standardized facial expressions of emotion depicting both threat and sadness (22, 39, 40). Informed by findings for depression we focused on the non-conscious viewing condition for threat and the conscious viewing condition for sad. See Additional File 1 for description of the task and imaging parameters).
iii) Neural circuit target regions of interest
Regions of interest for the neural circuit engaged by threat were the amygdala (bilaterally) and the sgACC and for the neural circuit engaged by sad, the amygdala (bilaterally), dorsal anterior cingulate, and anterior insula (bilaterally). These regions of interest were defined previously using a systematic procedure (41) informed by our theoretical taxonomy (17, 18). See Additional File 1 for additional details of region definitions.
- iv) Computing circuit function for a healthy reference standard sample
Regional activation and region-to-region connectivity values for these regions of interest were quantified for a healthy reference sample available to this study as a healthy reference standard for computing extent of neural circuit dysfunction in the clinical participants. We previously established that the healthy sample was characterized by good quality data (62). Healthy reference values for activation of regions of interest for the contrast of each emotion minus neutral and psychophysiological interaction (PPI) analysis for region-to-region functional connectivity were computed. These values were mean-centered, and scaled to be expressed as standard deviation units. A global circuit dysfunction score was computed by averaging the constituent activation and connectivity values for circuits engaged by threat and sad.
The reference sample included 40 healthy individuals without depression or obesity (mean age 33.9 years, SD 13.4, 50% female, 52% non-Hispanic White, mean BMI 24.1, SD 3.5, and mean PHQ-9 1.1, SD 2.2) completed identical fMRI assays.
- v) Computing circuit dysfunction for the present clinical trial sample
Patient-level activation for the regions of interest were derived in a manner consistent with the methods used for the healthy reference sample and that have been published previously (22, 39-41) and described in detail in Additional File 1. Similarly, PPI analyses, using an established approach (42) were used to quantify functional connectivity between regions of interest. See Additional File 1 for details.
- vi) Standardizing circuit dysfunction
Having identified the circuit dysfunction values for our clinical sample, we expressed each of these activation and connectivity values in standard deviation units relative to our healthy reference sample in order to interpret circuit dysfunction in clinical participants relative to a healthy reference group.(41) Through this procedure, global circuit dysfunction scores were interpretable relative to a healthy reference mean of zero. Then the values were winsorized using +/- 3 standard deviations.
As part of the standardization process we ensured that the directionality of each dysfunction value (i.e., activity of a region or region-to-region connectivity) was oriented so that greater scores always indicated greater dysfunction, according to our theoretical framework (17, 18). This process resulted in a single score for activation and connectivity value for each region of interest or pairs of regions. We also averaged these values in order to generate a global circuit dysfunction score for the neural circuit engaged by threat and sad.
Longitudinal mediation was analyzed to assess whether the change in a potential mediator from baseline to 2 months mediated the change in an outcome from baseline to 6 or 12 months. Based on the approach described by Kraemer et al. (43), mediation exists if there is a significant effect of the intervention (X) on the potential mediator (M, XàM Path A) and the potential mediator is significantly associated with the outcome either as a simple effect in the usual care group (or usual care effect) or an interaction effect with the intervention compared to the usual care group (MàY Path B) (Figure 2). Path A was assessed using the t test to obtain between-group differences of a potential mediator and 95% confidence intervals (CIs). Path B was tested using the ordinary least square regression to assess the association of a potential mediator with an outcome in the usual care group (the usual care effect) and by interaction with the intervention relative to the usual care group (the interaction effect), adjusting for the outcome at baseline. The usual care effect reflects the effect in the usual care group and the interaction effect reflects the effect in the intervention group relative to the usual care group. Participants were analyzed based on the group to which they were randomly assigned using all available data.
All analyses were conducted using SAS, version 9.4 (SAS Institute Inc., Cary, North Carolina); statistical significance defined by 2-sided P<0.05. P values were not adjusted for multiple comparisons as this was an exploratory study, and additional dedicated studies are needed to confirm the results.(44) Figures were generated using the ggplot2 package in R.