An official website of the United States government
NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.
Louisa G. Sylvia, PhD, Evan Albury, BS, Dustin Rabideau, PhD, Nicole Stephan, BS, Heidi Dohse, BA, and Andrew A. Nierenberg, MD.
Author Information and AffiliationsIndividuals with depression are 4 times more likely to die from cardiovascular disease (CVD) than are those without depression, and the cumulative effects of this comorbidity can result in a loss of 8 to 25 years of life expectancy. Conversely, people with CVD have not only 3 times the rate of depression compared with the general population, but they die at higher rates if they have comorbid depression. Exercise improves both depression and risk factors for CVD, yet most Americans do not exercise regularly, especially those with depression or heart disease. Exercise reduces the risk of dying by CVD and, for some individuals, can have antidepressant effects similar to those of medications.
The purpose of this study was to compare empirically supported psychotherapy interventions to increase exercise in individuals with a history of depression who had or were at risk for CVD (defined as <150 minutes of physical activity [PA] per week). The primary aim of this 3-arm randomized trial was to compare internet-based, self-guided cognitive-behavioral therapy (CBT) plus an activity tracker (Fitbit), self-guided mindfulness-based cognitive therapy (MBCT) plus Fitbit, and Fitbit alone. The primary outcome was increased exercise measured by increased daily steps measured by the activity tracker. The secondary aim was to determine the heterogeneity of treatment effects (HTE; predictors and moderators of treatment response) to both interventions.
We recruited adult participants (N = 340; mean [SD] age, 43 [11] years) who had received a diagnosis of unipolar depression and had or were at risk for CVD. Individuals who were already exercising regularly, were using a Fitbit, or were unable to exercise were excluded from the study. Participants were recruited from 2 online patient-powered research networks, MoodNetwork (individuals with mood disorders) and Health eHeart (focused on cardiovascular health). Eligible participants were then randomized to 1 of 3 groups: CBT+Fitbit, MBCT+Fitbit, or Fitbit only through a central randomization program, and then started the intervention (CBT or MBCT). The primary outcome of exercise was defined as the change in the number of steps taken per day as measured by the Fitbit at posttreatment (8 weeks) and follow-up (16 weeks). For the primary aim, we hypothesized that CBT will be superior to MBCT in increasing daily steps over the course of 8 weeks of treatment and after 8 additional weeks of follow-up (16 weeks total). To assess this, we used data provided by the maker of Fitbit to determine the total number of steps. We conducted general linear mixed models that accounted for the covariance of observations within participants. We assessed potential treatment moderators (eg, age, depression, stress, self-efficacy, well-being, [hypo]mania, weight, cigarette smoking status) to explore whether certain characteristics could be used to match treatments to specific subpopulations. We did not have any a priori hypotheses for our exploratory second aim to examine moderators of the main outcome (ie, whether certain populations of individuals might do better with one treatment over the other to increase their daily number of steps); therefore, we conducted HTE analyses. For the HTE analyses, we added appropriate interaction terms to the linear mixed model and used a likelihood ratio test to assess whether the estimated treatment effect differed across levels of the moderator. A significant interaction model would indicate that the HTE variable moderated differences between groups in daily steps changed over time.
The average number of daily steps changed by +2.8 steps per day (95% CI, −1.5 to +7.0) in the MBCT+Fitbit group and by +2.9 steps/day (95% CI, −1.3 to +7.0) in the CBT+Fitbit group (ie, both increased) but changed by −8.2 steps/day (95% CI, −14.5 to −1.9) in the Fitbit-only group (ie, decreased). The changes in average daily steps were not different between the MBCT+Fitbit and CBT+Fitbit groups (P = .97), but both were different from those of the Fitbit-only group across the initial 8-week period (P = .005 and .004, respectively). Group differences were not maintained across the entire 16-week follow-up period. Some moderators were identified in exploratory analyses, including comorbid anxiety disorders, PA (as assessed via the International Physical Activity Questionnaire), and employment status (see Table 4 and Table 5). During the first week of the study, participants took an average (SD) of 4778 (2421) (0%, 25%, 50%, 75%, 100% quantiles = 0, 3013, 4399, 6207, 13,372) steps per day. Over the 16-week follow-up period, participants provided an average (SD) of 81 (34) (0%, 25%, 50%, 75%, 100% quantiles = 1, 61, 95, 111, 112) days of Fitbit data, and the average within-participant SD in daily steps was 2505.
Second Aim: HTE Analysis at 8 Weeks.
Second Aim: HTE Analysis at 16 Weeks.
We found that online behavioral interventions yielded statistically significant but not clinically meaningful changes in daily steps at week 8; moreover, this gain was not maintained at week 16. Our results suggest that an activity tracker with or without self-guided web-based psychotherapy interventions has limited effects on PA in participants with a history of depression who have or are at risk for CVD.
This study had the following limitations: (1) potential treatment moderators, including adverse events, were self-reported; (2) there were technical/syncing issues with the Fitbit; and (3) we had a largely White, female, and highly educated sample, which hindered the generalizability of the results to other populations.
Approximately 121.5 million adults in the United States have some type of cardiovascular disease (CVD),1 which represents the country's overall leading cause of death2 and is associated with increased health care utilization and costs as well as preventable early loss of life.1,2 An estimated 17.3 million adults in the United States report having a lifetime major depressive episode.3 Up to 40% of adults who have suffered a major cardiac event meet the criteria for major depression disorder, and individuals with CVD experience 3 times the rate of depression relative to the general population,4-9 demonstrating the frequent comorbidity between these 2 conditions.9-11 When co-occurring, CVD and depression are associated with increased financial burden12-15 and reduced quality of life (QOL),16,17 and mortality is twice the rate than that in those who have CVD but no co-occurring depression.18-22 Thus, given the comorbid cardiovascular-related diseases and major depression, we need treatments that improve both conditions.
Physical activity (PA) is robust treatment for physical health and psychological conditions. Among patients with or at risk for cardiovascular health conditions, increased PA decreases blood pressure, mortality rate, risk of heart failure, and inflammatory marker profiles (eg, decreased C-reactive protein).23-26 Studies have shown that increased PA improves symptoms such as lack of interest and low motivation, but exercise may not be more effective than are psychosocial interventions alone.27,28 A combined intervention, or a PA-based psychosocial intervention, may enhance PA among individuals with combined psychological and medical (including cardiovascular) concerns.
Despite the evidence demonstrating that exercise can improve mental and physical health, it remains a challenge for most people to achieve the recommended amount (ie, at least 150 minutes/week). The challenge is even greater for individuals who are at risk for CVD (who typically do <150 minutes of PA/week) and depression. Many patients with cardiovascular conditions do not engage in PA because they believe exercise might lead to pain or exhaustion,29 low self-efficacy about exercise,29-31 and no incentive to exercise.30,32 Similarly, individuals with depressive symptoms lack motivation to exercise (which can be exacerbated by low energy and anhedonia),33,34 have decreased exercise-linked self-efficacy, and possess negative expectancies about exercise outcomes.34,35
Many of the barriers for individuals at risk for depression or CVD may be systematically addressed by psychosocial interventions that facilitate increased present-focused awareness (eg, to help individuals better recognize the distinction between their current behaviors and desired behaviors, thereby motivating behavior change), goal setting, and activity scheduling.36-39 There is evidence that psychosocial interventions can improve exercise,40-42 but these interventions have generally been studied independently rather than being directly compared. Therefore, there is limited evidence as to which interventions, if any, best facilitate behavioral modification, including changes in PA (eg, a more-directive, structured intervention, such as cognitive-behavioral therapy [CBT], or a more awareness-oriented intervention, like mindfulness-based cognitive therapy [MBCT]). To address these challenges, we examined 3 groups, (1) CBT+Fitbit, (2) MBCT+Fitbit, and (3) an activity monitor alone (Fitbit only), for increasing PA among individuals with a history of depression who have or are at risk for CVD. Participants were not given restrictions on how to use their Fitbit; however, they were given guidance for activities and how to increase their steps. The number of daily steps measured by Fitbit was chosen as an outcome measure for both conditions over self-report because it is an objective measure of activity.
The main objective was to help people with a history of depression and who have or are at risk for CVD improve their PA. The primary aim of this study was to evaluate whether internet-based, self-guided CBT or self-guided MBCT adjunctive to an activity tracker (Fitbit) increased exercise measured by increased daily steps compared with the use of an activity tracker alone. The secondary aim was to determine the heterogeneity of treatment effects (HTE; predictors and moderators of treatment response) to both interventions.
Given the high rates of sedentary behavior among individuals with depression at risk for or with CVD, the proposed study could have a major impact on public health, as it aims to increase access to empirically based treatments and further the personalization of interventions.43 Specifically, these interventions will address the known barriers to exercising for individuals with depression and CVD, which include low energy and motivation, fatigue, low mood (eg, feeling useless, “overwhelmed”), negative perceptions of health (eg, coming to terms with having heart disease), and physical limitations (eg, feeling restricted by their heart condition, fear that exercise could cause more damage to the heart, lack of knowledge about exercise, and difficulties getting organized).44-62
For the primary aim, we hypothesized that CBT could be superior to MBCT in increasing daily steps over the course of 8 weeks of treatment and after 8 additional weeks of follow-up (16 weeks total). For the secondary aim, we did not have any a priori hypotheses to examine moderators of the main outcome (ie, whether certain populations of individuals might do better with one treatment over the other to increase their daily number of steps).
The primary outcome (aim 1) was average number of daily steps, measured at 8 weeks (posttreatment) and 16 weeks (follow-up). For aim 1, we hypothesized that the CBT+Fitbit intervention would be more effective than would the MBCT+Fitbit intervention (given that CBT tends to be more prescriptive and focuses on behavioral change) at both posttreatment and follow-up, and that both interventions would be more effective than the Fitbit-only group. For aim 2, we assessed treatment moderators (eg, age, depression, stress, self-efficacy, well-being, [hypo]mania, weight, cigarette smoking status) to explore potential characteristics that could inform treatments to subspecific populations. For aim 2, we hypothesized that there would be clinical and demographic characteristics that may be associated with different treatment responses (eg, age, depression severity, stress, self-efficacy, well-being, [hypo]mania, and cigarette smoking status).
Patients and other stakeholders contributed substantially to the study as members of the study stakeholder committee. They were involved in creating and brainstorming our research question and in implementing our research goals and will assist in sharing and disseminating findings. Our patient stakeholders came from 2 groups: MoodNetwork, an online National Patient-Centered Clinical Research Network (PCORnet) patient-powered research network (PPRN) for individuals with mood disorders, and Health eHeart, a PPRN focused on cardiovascular health.
MoodNetwork patients and stakeholders, many of whom said they were at high risk for CVD (eg, obesity, prediabetes, diabetes, high blood pressure) or having CVD, overwhelmingly expressed the need for easy-to-access, effective lifestyle interventions to improve their physical fitness. These stakeholders discussed the reasons they did not live healthier lifestyles and why many had avoided mainstream lifestyle programs (eg, gym programs, Weight Watchers). Specifically, they were concerned that such programs did not sufficiently address key factors that prevented them from exercising. These barriers included low energy and fatigue; difficulty finding motivation, getting organized, setting goals, and following through; staying consistent; having negative thoughts and fear about exercise; having dysfunctional beliefs about their self-efficacy; and experiencing too much stress to engage in exercise, as well as misconceptions about whether exercise could reverse the risk of CVD and concern about how to address environmental factors (eg, family habits and structures).41,44-46,48-66 In November 2014, the Health eHeart Alliance hosted a Patient-Powered Research Summit to which patients, investigators, and other stakeholders were invited for 1.5 days to brainstorm research ideas. Participants self-organized into interest groups, one of which focused on exercise and technology. This group came up with the research question, “Can using activity monitors (mobile devices and apps, the internet) cause people with CVD to be more active?” and an outline for a research protocol that might answer the question. This group was led by the patient principal investigator (PI) of this proposal, Heidi Dohse, who has heart disease and possesses technology expertise as a former, high-level project manager at a successful technology company. Our stakeholders discussed the available tools to enhance exercise, such as peer-to-peer support, community-based programs, and family-focused interventions, as well as the purely technological support originally suggested by the Health eHeart Alliance brainstorming group. The stakeholder review of the available programs/options yielded that 2 modalities, CBT and MBCT, best addressed the psychological barriers to beginning exercise. As a result, we designed a study that incorporated CBT or MBCT and an activity tracker (ie, Fitbit) vs an activity tracker alone.
We conducted 25 stakeholder meetings throughout the duration of this study and typically met on a monthly basis since the development of the study through to its conclusion. Our stakeholders were instrumental in testing and editing the numerous iterations of the online interventions; noting the sections, words, or technical terms which were confusing or needed more explanation; and providing guidance on how to make them more patient friendly. They also were key in designing the onboarding process, from streamlining the baseline assessments to drafting the communications between the study team and participants. Even in the midst of technical challenges to the platforms, stakeholders guided the remediation plans to keep usability and transparency at the forefront. Stakeholders helped the project team decide on the most relevant and patient-centric outcome variables in the creation of this study (eg, the focus on increasing PA rather than weight loss).
Healthy Hearts Healthy Minds was a 16-week (8-week intervention, 8-week follow-up) randomized comparative effectiveness study. Participants were recruited from 2 online research communities: MoodNetwork (participants with mood disorders) and Health e-Heart Alliance (participants with CVD or at risk for CVD).
Study eligibility was determined at the screening session (week 0), and eligible participants were subsequently mailed a wrist activity tracker, Fitbit Charge 2 (Fitbit, Inc). Eligible participants were then asked to sync their Fitbit with the online study platform within 10 days after receiving the device. Notably, only 18% (n = 91) of participants synced their Fitbit within 10 days of receiving the device, and 22% (n = 112) took a month or longer to connect to the study portal. The remaining participants completed the task between 10 days and 1 month of receiving their Fitbit device. Once the Fitbit was synced, participants then completed baseline (week 1) assessments and were randomized to 1 of 3 groups (CBT+Fitbit, MBCT+Fitbit, or Fitbit only).
There were 361 participants randomized to the study interventions at a 2:2:1 ratio (n = 145 in the CBT+Fitbit group, n = 144 in the MBCT+Fitbit group, and n = 72 in the Fitbit-only group). Of the 361 randomized participants, 316 had Fitbit data; therefore, the current study focuses on this subset of the randomized sample. After being randomized, 21 participants (8 CBT+Fitbit, 11 MBCT+Fitbit, 2 Fitbit only) withdrew their consent; thus, data from these 21 participants are not used or presented subsequently.
Participants randomized to the CBT+Fitbit and MBCT+Fitbit groups completed 8 weeks of online intervention sessions with questionnaires to track mood symptoms every other week, whereas participants in the Fitbit-only group completed 8 weeks of questionnaires every other week alone. After the first 8 weeks of the study, participants received no new online intervention materials regardless of their initial group assignment but were instructed to continue wearing their Fitbit for the remaining 8 weeks of the study; the MBCT+Fitbit and CBT+Fitbit groups continued to have access to the intervention material presented in the first 8 weeks of the study. At week 16, participants completed a follow-up assessment consisting of the questionnaire battery participants were given every other week throughout the first 8 weeks of the study. Participants were compensated via Amazon gift cards every other week for every set of questionnaires ($5 for week 2, $5 for week 4, $10 for week 6, $15 for week 8, and $25 for week 16).
This study was hosted on both the MoodNetwork platform, which is hosted at Massachusetts General Hospital (MGH), and Eureka, a widely available digital platform designed to support direct-to-participant research and mHealth data collection used for a variety of research purposes. Participants completed the electronic informed consent and baseline diagnostic/eligibility assessments on the Eureka platform. Eligible participants were then mailed a Fitbit device and randomized to an intervention group. MoodNetwork deployed the study interventions, and Eureka deployed the weekly assessments.
This study involved two online PCORI-funded communities, MoodNetwork and Health eHeart Alliance, and used the Eureka Research Platform for eConsent and data collection. Eureka is designed to support direct-to-participant research and mHealth data collection and is funded by the National Institutes of Health (grant number U2CEB021881). All participants gave written informed consent in accordance with the Declaration of Helsinki. After registering and consenting for the study, participants completed online self-report assessments to determine eligibility. Study eligibility criteria included being between the ages of 18 and 65 years, a self-report of at least 1 episode of depression in their lifetime, and a self-report of increased risk for or current CVD (defined as <150 minutes of PA/week). If eligible, participants were mailed a Fitbit device. Once the device was received, participants were directed back to the study portal to complete baseline assessments, sync their Fitbit, and begin the study. Participants were randomized to either the CBT+Fitbit, MBCT+Fitbit, or the Fitbit-only control group. Participants randomized to 1 of the 2 online interventions (ie, CBT or MBCT) were automatically redirected to a web portal hosted by the MoodNetwork community where they could access their assigned study intervention. Integration between and authentication across the 2 systems was not patient facing so that participants logged into the study portal hosted on Eureka were able to access the interventions maintained by MoodNetwork. In sum, the consent, eligibility, Fitbit integration, and study assessments were hosted on Eureka, and the study interventions were hosted on the MoodNetwork platform. Those randomized to the Fitbit-only treatment group were reminded to wear their Fitbit for the next 16 weeks. All participants, regardless of study group, were asked to return to the study portal at weeks 2, 4, 6, 8, and 16 to complete a battery of assessments [Appendix A]). The protocol was approved by the Partners IRB, has been submitted to ClinicalTrials.gov (identifier NCT03373110), and conforms to the tenets of the Declaration of Helsinki.
We recruited adult individuals with a history of depression (DSM-V major depression or bipolar disorder) and a sedentary lifestyle who had CVD or were at risk for CVD (<150 minutes of PA/week). Please refer to Table 1, Appendix B, and Appendix C for full baseline demographic and clinical characteristics of the 340 consented and randomized study participants. Participants (mean [SD] age, 43 [11] years) were mostly female (82% [n = 279]), White (83% [n = 281]), non-Hispanic (92% [n = 310]), and college educated (92% [n = 311]), with moderate depressive symptoms at baseline (mean [SD] Patient Health Questionnaire-9 [PHQ-9] score, 10 [6]). There were some observable differences across groups in employment status and cigarette smoking status.
Baseline Demographics and Clinical Characteristics by Randomized Intervention Group.
Participants were recruited via the Health eHeart and MoodNetwork PPRNs' websites and email accounts. We also utilized our advocacy partners to promote the study on their respective social media platforms. Inclusion criteria were as follows: ability to give informed consent, fluent in English, ages 18 to 65 years (8 participants over the age of 65 years completed the study), US resident, had a past experience of depression (via a self-report form of the MINI International Neuropsychiatric Interview depression module), self-report of elevated risk for CVD (<150 minutes of PA/week) or of having CVD, and having registered a first-time personal account with Fitbit following screening and prior to enrollment in the study (this inclusion criterion was required because participants needed to agree to the Fitbit terms and agreements before joining the study research study to highlight that this was participants' own decision to register an account with Fitbit. Once registered, participants were then eligible for the study).
Exclusion criteria were as follows (please see the “Changes to the Original Study Protocol” section for original exclusion criteria that are not included in the list): pregnant; active suicidal ideation as measured by question 3 on the PHQ-9, “thoughts that you would be better off dead or of hurting yourself in some way nearly every day”; contraindications to exercise (ie, score less than 5 metabolic equivalents of task via the Duke Activity Status Index [DASI]); current or past user of a Fitbit device or any other activity tracker; self-reported recurrent (ie, >1) incident of “blacking out” or “fainting”; survived a cardiac arrest; have recurrent chest discomfort with activity; current or past heart failure; or hospitalized recently (within past 6 weeks) for a cardiovascular problem.
This study compared (1) web-based CBT adjunctive to an activity monitor (CBT+Fitbit), (2) web-based MBCT adjunctive to an activity monitor (MBCT+Fitbit), and (3) an activity monitor alone (Fitbit only) for increasing PA (defined as daily steps over the course of an 8-week treatment period and 8-week follow-up period). Participants in both therapy-based interventions accessed materials (eg, client vignettes, videos, activities) that illustrated ways to increase PA (eg, walks, runs, climbing stairs). They were guided (through the online portal) to gradually increase their daily steps (PA) at their own pace without a minimum or maximum recommended number of steps. Both CBT and MBCT were provided online, as prerecorded interventions through the study portal, and guided participants through the elements of the treatments and were addressing PA for enhancing cardiovascular health in a population that was not exercising at the minimum recommended level. The interventions focused on PA for cardiovascular health in a population of individuals with a history of depression with or at risk for CVD. Both CBT and MBCT interventions were composed of 8 sessions. Each session was approximately 30 minutes depending on how long participants took with completing the interactive exercises (eg, identifying obstacles to increasing the number of daily steps). Both interventions were provided via online-accessible videos. In addition to structured guidance to increase exercise, both treatments addressed ways to overcome obstacles to individuals from engaging in exercise. Participants in both groups received Fitbit devices to measure their steps throughout the study. The structure and format of the 2 online interventions (ie, CBT and MBCT) were equivalent, such that participants accessed the intervention material via the same online study portal. Participation was passively monitored by the study platform (eg, how many times the portal was accessed to view the study intervention). While data on uptake of the intervention were collected, these data were not reported due to validity concerns. The difference between the groups was the content of each session. Specifically, MBCT focused more on building awareness of daily experiences (with the goal of recognizing, as a result, factors that impede a balanced lifestyle), whereas CBT focused on behavior change. With regard to the structure, the first session explained the study platform and how to use it as well as introduced the rationale for using CBT or MBCT (depending on group assignment) to increase participants' daily PA. One or 2 skills were taught each week, and participants were requested to practice skills until the following week when another session would be released. During the last session at week 8, participants were encouraged to continue to use their Fitbit to track their steps as well as to come back to the intervention material to access their worksheets and the other intervention material that would remain stored and accessible in their study account for the duration of the follow-up. In addition to the 2 treatment-based intervention groups, the study included a control group. Participants in the control group received Fitbits, but they did not have access to the online therapy material. By including this group, we were able to assess the treatment effect of the therapies beyond the use of a Fitbit alone.
Randomization was stratified (and blocked within strata) through a central randomization program (ie, the study platform) according to risk for or current CVD (<150 minutes PA/week) because we expect potential HTE to be determined by this factor. Block randomization was used with block sizes of 2 and 4 in a 2:2:1 ratio. This randomization schedule was visible only to study platform programming staff. All participants were randomized after agreeing to participate and providing informed consent.
The CBT program was developed and implemented by the co-I (Dr Louisa G. Sylvia) and was based on previous work using CBT to enhance exercise8,12,13 in individuals with depression. It included the following elements: (1) identifying and setting realistic exercise-based goals to maximize success at achieving exercise goals, and thereby increase motivation; (2) scheduling to optimize when to exercise, to problem solve obstacles to engaging with exercise, and to reinforce engagement with exercise by determining rewards for meeting behavioral goals; and (3) recognizing dysfunctional, maladaptive thoughts about exercise which decrease motivation and develop skills to encourage the generation of more-adaptive, positive thoughts surrounding exercise to overcome thoughts, for example, of being too tired or experiencing too much stress to exercise.
The MBCT program was based on previous work among individuals with mood disorders1-5 and adapted for online use. A central aspect of MBCT is the concept of awareness. Participants practice a variety of meditation types (eg, breath awareness) and learn to bring mindfulness to everyday situations. Awareness is directed to elements in participants' lives that interfere with living a more productive, physically active life (eg, thoughts and feelings that interfere with becoming more physically active, stressful situations and circumstances that prevent engagement with exercise). The program guides participants to a mindful consideration about how to “respond” to factors that interfere with a healthy lifestyle and encourages participants to bring mindfulness to PA and its pleasant after-effects (as a way of enhancing motivation).
MBCT employs a more explorative, self-guided approach to increasing awareness than does CBT, which utilizes a more directive, prescriptive approach. CBT focuses on goal setting, planning, behavioral activation (increasing engagement with activities that promote positive feelings) and reward, and reconceptualization of dysfunctional thoughts that interfere with goal achievement, whereas MBCT strives to gradually increase awareness of personal and situational factors that interfere with a healthy, physically active lifestyle, and to change one's experience of engaging in PA by helping individuals become more mindful while exercising. Because participants were not able to choose the intervention to which they would be assigned, we did assess treatment preferences and will explore these preferences as a potential source of HTE.
We chose the Fitbit Charge 2, a bracelet-like band that is worn on the wrist, to measure the number of daily steps (our main outcome variable) of the study participants. Fitbit products are the most commonly purchased wearable trackers; in June 2015, market share from first-quarter sales indicated that the top wearable tracker vendor was Fitbit (34% of all wearable tracker sales). Additionally, there is evidence that the step counts measured by the Fitbit Charge 2 are significantly correlated with ActiGraph steps19-21 and with visually counted steps.22 Additionally, a systematic review of the validity and reliability of consumer-wearable activity trackers found that Fitbit's interdevice reliability of step counts was generally high, particularly during laboratory-based treadmill tests.21
Both the CBT and MBT treatments have demonstrated effectiveness in addressing barriers that prevent exercising and in improving mood symptoms and QOL, and increasing exercise/PA.41,44-46,48-59,60-63 If the question is whether a treatment is efficacious, a randomized controlled trial with placebo would be the first choice. In this case, with already-existing efficacy information, and consistent with PCORnet's guidelines for demonstration projects to conduct comparative effectiveness research, we opted to compare the effectiveness of CBT vs MBCT, with Fitbit only serving as a control group (aim 1) and to determine who (eg, sex, education, psychological functioning) responded better to which treatment (aim 2) to advance to more-personalized treatment recommendations.
We used linear mixed-effects models to examine the effect of the interventions on total daily steps (dependent variable). The models included random participant intercepts and slopes and fixed effects for intervention, (linear) time, and an intervention × time interaction. We used separate models to assess the intervention effects across the first 8 weeks of the study as well as the complete 16-week follow-up period. For each time period, we used a single omnibus likelihood ratio test to assess whether there were any group differences in average daily step improvement over time, followed by 3 Wald tests to assess all pairwise group comparisons. We reported model-based, within-group average slopes, between-group differences in slopes, and corresponding 95% CIs. The Fitbit recorded start and stop times of the day (eg, we excluded any steps recorded during Fitbit-indicated sleeping hours), and therefore, steps were counted only during these times. Participants without any step count data for a given day were treated as missing the outcome for that day. Participants without any step count data for a given day were treated as missing values for that day. There was no minimum requirement or cutoff for daily steps. Statistical analyses were performed using the R software6 (ie, version 4.0.2, r-project.org; ImerTest, version 3.1-2; Ime4, version 1.1-23; multcomp, version 1.4-13; ggplot2, version 3.3.2). A 2-sided significance level of .05 was used for all analyses without adjustment for multiple comparisons.
The primary outcome was number of daily steps measured via Fitbit across 8 weeks and 16 weeks.
Specific aim 2 was to explore the HTE on the primary outcome (daily steps). This aim allowed us to explore who benefited most and least from either CBT or MBCT (predictors) and determine the patient factors that would identify patients likely to have a larger- or smaller-than average response to either intervention (moderators or predictors of better outcomes with CBT or MBCT). Moderators included age, sex, smoking status, baseline PA, employment status, comorbid anxiety, depression symptoms, (hypo)mania symptoms, self-efficacy for exercise, and perceived stress.
Participants in this study completed self-report assessments at 7 time points over the 16-week study period. Please refer to Appendix A for a full schedule of the study assessments and time points.
The Mini International Neuropsychiatric Interview (MINI) can assess for current 17 DSM-V Axis I diagnoses, exploring lifetime diagnoses where clinically relevant (ie, previous manic episode for a diagnosis of bipolar disorder).64 Diagnoses can be ruled out by answering “no” to 1 or 2 screening questions. Positive responses to screening questions are followed by further exploration of other diagnostic criteria. The MINI shows good specificity and sensitivity for most psychiatric diagnoses and concordance with other structured diagnostic interviews. A self-report form of the major depressive episode module was administered only at the screening session to evaluate lifetime experience of depressive symptoms.
The World Health Organization Composite International Diagnostic Interview for Bipolar Spectrum Disorders (CIDI) is a 12-item, fully structured, lay-administered diagnostic interview that was validated as being capable of generating conservative diagnoses of both threshold and subthreshold bipolar disorder.65
Participants provided information at study entry on demographic variables such as age, biological sex, gender identity, race, ethnicity, sexual orientation, marital status, smoking status, employment status, and education history.
At the screening session, participants supplied information on current psychiatric medications, comorbid anxiety, and substance use:
Our ethnic, racial and age distribution in this study (see Table 2) was modeled based on research showing THAT more than two-thirds (70.2%) of adults aged 20 years or older are considered to be overweight or obese (White, 66.6%; Hispanic, 78.8%; African American, 76.6%; Asian, 40.4%; American Indian/Alaskan Native, 67.3%; Native Hawaiian/other Pacific Islander, 71.3%75). Ten percent to 12% of adults ages 20 to 39 years and 36% to 41% of adults ages 40 to 59 years currently have CVD.76 In 2013, an estimated 15.7 million adults aged 18 years or older in the United States had at least 1 major depressive episode in the past year. This represents 6.7% of all US adults. Of adults in the United States, 2.6% have some form of bipolar disorder,77 with recurrent depression dominating the clinical picture.78
Anticipated Final Racial or Ethnic and Gender Enrollment.
We followed FDA guidelines for clinical trials and thus did not correct our single primary outcome for multiple comparisons; similarly, other secondary and exploratory analyses were not adjusted, but we present the results regardless of statistical significance. We also assumed a 10% loss of participants, which left us with 225 participants per group. With these assumptions, we assessed power with PASS 14.0.79 We computed the power to detect various levels of the standardized mean difference between group means of the primary outcome (daily steps) based on a repeated-measures analysis of variance with an AR(1) covariance structure (see Figure 1 and Figure 2). This analysis model does not correspond exactly to the mixed model we use in the study. Computing power for mixed models is not feasible because it relies on too many unknown features of the data.
Table 3 outlines the overall study timeline. IRB approval was delayed because the study utilized a commercially available activity tracker and 2 web platforms, which meant there were 2 ways to store, retrieve, and transfer data. IRB approval required that multiple groups within MGH reviewed the study. The initial approval took over 3 months.
Study Timeline.
Recruitment began later than expected due to Fitbit connection issues and email issues, and because some participants bypassed consent (ie, there were technical difficulties within the study portal that allowed some participants to bypass the consenting procedures; these participants were recontacted for consent). Additionally, early recruitment was paused for two months due to a limited number of available Fitbits. Additionally, we found that our inclusion/exclusion criteria were too strict, as 89.4% of participants who consented from Health eHeart (2768 consented) and 68.2% of participants who consented from MoodNetwork (622 consented) were found ineligible, which threatened the feasibility and generalizability of our study. We found that most participants were excluded because they exceeded the threshold for suicidal ideation, had too many depressive symptoms, or were not overweight. We consulted with our study stakeholder committee about including participants who had suicidal thoughts as well as moderate to severe depression symptoms. Our stakeholders and data and safety monitoring board established that allowing participants with a suicide risk to participate in the study was ethically important and closer to real-world clinical practice. We emphasized in the consent form that the study clinicians and staff would not monitor their symptoms in real time and to check in with their regular provider if any safety concerns emerged.
We tried to prevent dropout by using patient-centered interventions (eg, messaging, tools, videos) to maximize retention and reduce missing outcome data. We electronically checked surveys for completion. Health eHeart and MoodNetwork tracked whether participants logged on to the web-based intervention to complete their respective treatment portions/assignments and reminded participants to do so by email and text message. Once assessments or treatment portions were overdue, the research project team was alerted that a participant was overdue; thus, the team could reach out to the participant as needed. When data from active participants were missing, we made every attempt to obtain the reasons why as well as to contact participants when their data were not sent per the protocol schedule.
Unique identifiers (UIDs), or strings of letters and numbers continuing no personal identifications, were used to deidentify the data. MGH securely stored a linking list document, which allowed the UIDs from this project to be linked to the UIDs used by MoodNetwork and Health eHeart, allowing access to historical medical data stored in these databases. However, no data generated by this project were inserted into MGH or Mass General Brigham medical records.
Aim 1 was to determine which intervention (ie, CBT+Fitbit, MBCT+Fitbit, Fitbit-only group) is superior to increase daily steps as measured by Fitbit for the periods of (a) baseline to 8 weeks and (b) baseline to 16 weeks. We used linear mixed-effects models to examine the effect of the interventions on total daily steps. The models included random participant intercepts and slopes and fixed effects for intervention, (linear) time, and an intervention × time interaction. We used separate models to assess the intervention effects across the 8-week intervention period as well as the complete 16-week follow-up period. We reported model-based within-group average slopes, between-group differences in slopes, and corresponding 95% CIs. The Fitbit recorded start and stop times of the day (eg, we excluded any steps recorded during Fitbit-indicated sleeping hours), and therefore, steps were only counted during these times. Participants without any step count data for a given day were treated as missing the outcome for that day. Participants without any step count data for a given day were treated as missing values for that day, but our models incorporated all available daily step measurements from participants. Sensitivity analyses were carried out by adding the number of days between study screening and randomization, history of smoking at baseline, and baseline employment status (all 3 added simultaneously in 1 analysis), as well as post randomization number of intervention sessions completed (separate analysis) to the primary mixed-effects model.
Aim 2 was to determine the HTE (predictors and moderators of treatment response) of both interventions. In exploratory analyses, demographics and total assessment scores measured at the beginning of the intervention period were evaluated as potential moderators of the relationship between intervention and daily steps over time. We added a fixed effect for the moderator, as well as interactions between the moderator and all other terms in the linear mixed model described above (ie, moderator × intervention, moderator × time, and a 3-way moderator × intervention × time interaction). We used a likelihood ratio test to assess the significance of the 3-way interaction term, a term which represents the estimated differential treatment effect across levels of the moderator. Separate moderator analyses were carried out for the 8-week and 16-week periods.
Statistical analyses were performed using R (v4.0.2 r-project.org).80 A 2-sided significance level of .05 was used for all analyses without adjustment for multiple comparisons.
After carrying out an analysis using all available data, we will carry out an exploratory data analysis to summarize the extent and patterns of missingness, paying special attention to the primary outcome (total daily steps), secondary outcomes, and important covariates used in the models for aim 1. Additional analyses will be carried out using multiple imputation or inverse probability weighting to incorporate information on observed covariates that are known/thought to influence differential missingness of the primary outcome. If data are thought to be missing not at random (ie, if missingness is thought to depend on unobserved covariates or outcomes), we will carry out appropriate sensitivity analyses.81
This study was designed to exclude individuals who already owned activity monitors; however, we realized many in our community already had activity monitors, and after discussions with our data and safety monitoring board, we deleted this criterion. Other changes to the original study protocol included allowing participants who had suicidal thoughts as well as moderate to severe depressive symptoms to participate in the study as well as participants who were not overweight. Additionally, our IRB approved the use of the SAQ-19; however, participants were administered the SAQ-11 (short version). This was noted as a minor deviation.
Study data will be available to external collaborators upon request.
See Table 1 for full baseline demographic and clinical characteristics of study participants. Participants (mean [SD] age, 43 [11] years) were mostly female (82% [n = 279]), White (83% [n = 281]), non-Hispanic (92% [n = 310]), and college educated (92% [n = 311]), with moderate depressive symptoms at baseline (mean [SD] PHQ-9 score, 10 [6]).There were some observable differences across groups in employment status and cigarette smoking status. The mean (SD) number of intervention sessions completed for all participants was 6.6 (2.9) and by group was 6.5 (3.0) for MBCT+Fitbit and 6.8 (2.8) for CBT+Fitbit. Participants (N = 361) were randomized to the study interventions (n = 145 in the CBT+Fitbit group, n = 144 in the MBCT+Fitbit group, and n = 72 in the Fitbit-only group). See Figure 3 for a CONSORT chart. After being randomized, 21 participants (8 CBT+Fitbit, 11 MBCT+Fitbit, 2 Fitbit only) withdrew their consent; thus, data from these 21 participants are not used or presented subsequently. The remaining 145 eligible participants were not randomized because they failed to return to the portal to complete their baseline assessments and sync their Fitbit. Among the 340 consented and randomized participants, 314 participants provided Fitbit data (ie, 26 participants did not have Fitbit data recorded for any study days, which may have been caused by syncing issues or not syncing at all); therefore, the current study will focus on this subset of the randomized sample (see Figure 3 for a CONSORT chart).
CONSORT Chart.
A visual depiction of daily steps in each of the intervention groups over time is presented in Figure 4. During the first week of the study, participants took an average (SD) of 4778 (2421) (0%, 25%, 50%, 75%, 100% quantiles = 0, 3013, 4399, 6207, 13,372) steps per day. Over the 16-week follow-up period, participants provided an average (SD) of 81 (34) (0%, 25%, 50%, 75%, 100% quantiles = 1, 61, 95, 111, 112) days of Fitbit data, and the average within-participant SD in daily steps was 2505. Change in average daily steps differed by intervention group across the 8-week intervention period (P = .01). Specifically, in the CBT+Fitbit and MBCT+Fitbit groups, participants had an average change of +2.9 (95% CI, −1.3 to +7.0) steps per day and +2.8 (95% CI, −1.5 to +7.0) steps per day (ie, an average increase), respectively, whereas participants in the Fitbit-only group had an average change of −8.2 (95% CI, −14.5 to −1.9) steps per day (ie, an average decrease), respectively. Thus, both the CBT+Fitbit and MBCT+Fitbit groups experienced more change in average daily steps through the intervention period than did the Fitbit-only group (P = .004 and .005, respectively). We did not find a significant difference between the CBT+Fitbit and MBCT+Fitbit groups in terms of their average changes in daily steps across the 8-week intervention period (P = .973). These models suggest that at 8 weeks, participants in the CBT+Fitbit and MBCT+Fitbit groups were walking an average total (over the 8 weeks, or 56 days of treatment) of 160 (95% CI, −71 to 392) and 155 (95% CI, −85 to 395) more steps per day than they were at baseline (both corresponding to an approximate 3% increase above 4776, the mean daily steps in the first study week), respectively, whereas the Fitbit-only group was walking an average total (over the 8 weeks) of 458 (95% CI, 106-811) fewer steps per day compared with baseline (corresponding to a 10% reduction). These findings were not maintained across the entire 16-week follow-up period. At 16 weeks, relative to baseline, participants in the CBT+Fitbit, MBCT+Fitbit, and Fitbit-only groups all reported average changes of −1.8 (95% CI, −3.5 to −0.1), −1.2 (95% CI, −2.9 to +0.5), and −2.6 (95% CI, −5.1 to −0.1) steps per day, respectively (omnibus P = .65, all pairwise P > .35). All of these general findings were retained when adjusting for number of days between study screening and randomization, history of smoking at baseline, and baseline employment status as well as postrandomization number of intervention sessions completed (Appendix D and Appendix E).
Intervention Groups Over Time.
Table 4 and Table 5 show complete results (within-group slope estimates, 95% CIs, and omnibus interaction P values) of these exploratory moderator analyses across both 8 and 16 weeks. In Appendix F, we additionally present point estimates, 95% CIs, and P values for pairwise model-based group differences within levels of the 3 moderators, with an interaction P < .05 for both 8- and 16-week analyses. The subgroup estimates in Appendix F can be contrasted with their overall counterparts in Table 6. The 3 moderators identified based on omnibus P values included comorbid anxiety disorders, self-reported PA, and employment status. We briefly summarize the results for these 3 moderators below, focusing on how they impacted the primary results; that is, how the model-based differences in slopes between intervention arms were impacted by each moderator.
Analysis of Primary Study End Point (Daily Steps Measured via Fitbit) Using Linear Mixed-Effects Models: 8 Weeks and 16 Weeks.
Over 8 and 16 weeks, group differences favoring MBCT were generally more pronounced among those without comorbid anxiety than in those with comorbid anxiety (eg, model-based difference in slopes MBCT vs Fitbit only, 23.3 [95% CI, 10.3-36.4] among those without anxiety, compared with corresponding overall MBCT vs Fitbit only estimate of 10.9), except for MBCT vs CBT across 16 weeks. Similarly, higher self-reported PA at baseline also tended to favor MBCT generally more markedly and CBT vs Fitbit only over 8 weeks, though the impact of baseline PA was not as clear or convincing over the full 16-week follow-up period. On the other hand, MBCT performed comparably worse among participants who reported being unemployed, especially compared to CBT (eg, model-based difference in slopes MBCT vs CBT of −27.5 [95% CI, −44.0 to −11.0] and −15.4 [95% CI, −22.1 to −8.6] among unemployed across 8 and 16 weeks, respectively).
In sum, these exploratory moderator analyses suggest that the 2 active interventions in this study (MBCT and CBT) may have been more effective for those individuals without comorbid anxiety, with higher PA, and those who were employed, whereas they were less effective for individuals with anxiety, lower PA, and who were unemployed at baseline. Moreover, these data suggest that CBT+Fitbit was potentially more favorable than was MBCT+Fitbit for participants who were unemployed.
We also found that those exercising more at baseline, relative to those exercising less (via the IPAQ), increased their daily steps by week 8 in the MBCT+Fitbit group, whereas those exercising less at baseline, relative to those exercising more, increased their daily steps by week 8 in the CBT+Fitbit group. In contrast, participants in the Fitbit-only group decreased their daily steps regardless of their exercise levels at baseline. By week 16, participants decreased their daily steps in all treatment groups, regardless of whether they were exercising more or less at baseline (P < .01). However, participants in the CBT+Fitbit and Fitbit-only groups who exercised more at baseline had larger decreases in daily steps over all 16 weeks relative to those who exercised less at baseline, whereas in the MBCT+Fitbit group the pattern was reversed, such that those who exercised more at baseline had smaller decreases in daily steps over all 16 weeks relative to those who exercised less at baseline.
We found that, on average, the CBT+Fitbit and MBCT+Fitbit online interventions increased participants' daily steps relative to the Fitbit-only group over the 8-week intervention period, but this difference was not maintained across the entire 16-week follow-up. We did not find a difference between the CBT+Fitbit and MBCT+Fitbit groups across either analysis period. These results should be interpreted cautiously, as changes in daily steps over both 8- and 16-week periods—regardless of intervention group—were small. Overall, these data suggest that for both of our intervention groups (CBT+Fitbit and MBCT+Fitbit), it is not clear that the relative increases in daily steps represent sufficient change in daily steps to suggest that these interventions have a clinically meaningful impact on an important risk factor for CVD. As for HTE, we found that compared with the study population as a whole, participants with lifetime anxiety and greater baseline exercise were significantly different in step counts at weeks 8 and 16. The directionality of the effects (larger or smaller decreases) of greater baseline exercises varied by treatment group. The inconsistency in how lifetime anxiety and baseline exercise impacted the efficacy of the interventions for PA requires further research.
We examined the HTE of each treatment group (ie, CBT+Fitbit, MBCT+Fitbit, Fitbit only) to see if demographic, psychiatric, or medical characteristics moderated the results. Specifically, we evaluated age, sex, weight, education level, employment status, comorbid anxiety disorders, cigarette smoking, baseline depression severity, baseline (hypo)mania severity, baseline well-being (as assessed via the World Health Organization-5 Well-Being Index70), baseline perceived stress (as assessed via the Perceived Stress Scale71), baseline self-efficacy for exercise (as assessed via the Self-Efficacy for Exercise72), and baseline PA (as assessed via the IPAQ73 and DASI66). Several moderators had interaction P < .05 at postintervention and 16-week follow-up (ie, comorbid anxiety disorders, PA, and employment status). Age, comorbid anxiety disorders, baseline PA, education, and employment status had interaction P < .05 only at 8 weeks, and sex, comorbid anxiety disorders, baseline PA, mania symptoms, and employment status had interaction P < .05 only at 16 weeks.
Our results build on mixed evidence in the literature on the efficacy of online interventions for depression, CVD, and behavioral modification. According to a review of online lifestyle interventions in depressed populations, only 4 of 7 studies reported a significant change in lifestyle behavior (reduction in alcohol use, improvement in sleep, and enhanced PA), and 5 of 7 studies demonstrated moderate to high attrition rates (50%-80% missing completed measures).82 This experience is consistent with our data: we found that online interventions for exercise yielded only a modest, short-term benefit relative to the control group, a result that was not maintained at follow-up. Relative to baseline, the active interventions did not increase step counts. It is possible that outcomes could have been improved with coaching, or a synchronous, virtual component between participants and trained study staff to enhance engagement and adherence to the online program, as we noted that participants in the CBT+Fitbit and MBCT+Fitbit groups completed roughly 6 of 8 full treatment sessions. Other studies of internet-based interventions for depression and/or PA also concluded that increased contact with a coach or therapist would likely increase the efficacy of these interventions compared with that of self-guided programs.83-86 Future research should explore the impact of adjunctive coaching to online programs for improving PA.
This study has a few notable limitations. First, our sample was disproportionately White (83%), female (82%), and highly educated (ie, 92% had a college education or more), consistent with the online communities from which we recruited study participants (ie, MoodNetwork and Health eHeart). This population is highly skewed, limiting both internal validity (eg, low rates of CVD end points) and external validity (eg, exclusion of older adult and youth populations, gender). Moreover, the demographics of our study population are consistent with prior data which suggest that study populations in research studies that recruit online tend to be White, female, and well educated more broadly.87,88 Further, all study data (eg, medical history, psychiatric diagnoses, mood symptoms, level of functioning, overall wellness), with the exception of daily PA, were based on self-report alone. To allow for scalability, this study did not provide any way to make personal contact with study staff to clarify study procedures, support the intervention material, and/or troubleshoot technical issues with the Fitbit or online platform unless participants proactively emailed us. Thus, participants did not receive any formal support by study staff or accountability for adhering to study procedures and interventions. Last, we recognize that many comparisons were made with the HTE analyses (including dichotomized analyses); however, many of these exploratory analyses are meant to be hypothesis generating.
Future research should examine the potential benefit of adding a live supportive component to the online interventions, such as a coach or therapist, who could provide tailored feedback and further personalize such programs.
Despite its limitations, this study is the first to examine online CBT and MBCT tailored specifically to sedentary patients with a history of depression to increase their PA. We found that CBT+Fitbit, MBCT+Fitbit, or a Fitbit alone had minimal, clinically unimportant, unsustained effects on the number of daily steps. Participants had high rates of adherence with the online questionnaires (ie, participants [N = 316] completed 87% of online questionnaires) as well as adherence with Fitbit use (ie, participants wore their Fitbit devices for 76% of all study days). In sum, this study highlights the challenges in supporting long-term lifestyle changes via a self-guided online platform among generally sedentary participants with a history of depression.
Research reported in this report was funded through a Patient-Centered Outcomes Research Institute® (PCORI®) Award (PPRND-1507-31449). Further information available at: https://www.pcori.org/research-results/2016/comparing-three-ways-increase-physical-activity-patients-depression-and-cardiovascular-disease-healthy-hearts-healthy-minds-study
Schedule of Assessments (PDF, 96K)
Baseline Demographics Among 314 Participants (PDF, 118K)
Baseline Demographics Among 340 Participants (PDF, 119K)
Analysis of Primary Study End Points Adjusted for Postrandomization Intervention Sessions (PDF, 72K)
Select Pairwise Group Differences for Exploratory Moderator Analyses (PDF, 92K)
Sylvia LG, Albury E, Rabideau D, Stephan N, Dohse H, Nierenberg AA. (2022) Comparing Three Ways to Increase Physical Activity in Patients with Depression and Cardiovascular Disease—The Healthy Hearts Healthy Minds Study. Patient-Centered Outcomes Research Institute (PCORI). https://doi.org/10.25302/02.2022.PPRND.150731449
The [views, statements, opinions] presented in this report are solely the responsibility of the author(s) and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute® (PCORI®), its Board of Governors or Methodology Committee.
This book is distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License which permits noncommercial use and distribution provided the original author(s) and source are credited. (See https://creativecommons.org/licenses/by-nc-nd/4.0/
Your browsing activity is empty.
Activity recording is turned off.
See more...
