An Initial Test of the Efficacy of a Digital Health Intervention for Bariatric Surgery Candidates

Abstract

Purpose:

Rigorous research on smartphone apps for individuals pursuing bariatric surgery is limited. A digital health intervention was recently developed using standard behavioral weight loss programs with specific modifications for bariatric surgery. The current study evaluated this intervention for improving diet, exercise, and psychosocial health over 8 weeks prior to surgery in an academic medical center.

Materials and Methods:

Fifty patients were randomized to receive either the digital intervention or treatment as usual prior to a surgical procedure. Measures of anxiety, depression, stress, quality of life, physical activity, and diet were administered at baseline and at 8-week follow-up. Statistical power of 80% estimated for N = 50 to detect ES = 0.68 with alpha = 0.05.

Results:

Results of intent to treat (N=50 baseline, N=36 follow-up) analyses indicated significant moderate differences in stress and anxiety (ES = −0.58 to −0.62) favoring the digital intervention. Effects of the program on total daily calories consumed, body mass index, quality of life, and eating disorder symptoms were small (ES = −0.24 to 0.33) and not significant. Given small effects for these domains, the sample size of the study likely affected the ability to detect significant differences.

Conclusion:

The digital health intervention appears to significantly impact several measures of physical activity and emotional functioning in candidates for bariatric surgery, which could augment surgical outcomes.

Introduction/Purpose

Digital health technology, including smartphone applications (apps) have expanded exponentially, and more than 10,000 target mental or behavioral health [1]. This technology may uniquely circumvent obstacles to traditional weight loss treatments, like price, access to trained professionals, and time-burden/scalability. One digital weight loss program with specialized coaches (Noom Coach) found that 77% achieved reliable weight loss (>5%) over the short term [2]. Coaching appears to facilitate behavioral health interventions more broadly [3–8] including in health care systems like Kaiser Permanente [9–10] and in countries with limited infrastructure for health care [11].

Conflicting data exist for whether pre-surgery behavioral interventions augment post-surgery weight loss [12–13], but clinically significant reductions in weight are associated with favorable changes in cardiometabolic risk factors [14–15]. Digital technology could supplement bariatric services given the efficacy of behavioral interventions for weight loss in non-surgical groups and disseminability [16]. However, empirical investigations of digital programs are limited and bariatric surgery apps focus on patient information, support forums, or self-monitoring [17], instead of more comprehensive treatment approaches. One study found utility in educating patients before bariatric surgery via mobile technology [18], but more data are needed [19].

This study aimed to evaluate a digital app-based health intervention on dietary intake, physical activity, and weight for patients planning to undergo bariatric surgery. The intervention was developed from standard behavioral weight loss treatments with modifications for bariatric surgery (Noom Bariatric Health, Noom Inc, New York, NY), and positive results would initiate a larger phase II efficacy trial. Evidence using in-person and telemedicine delivered interventions reliably demonstrate improved psychosocial and weight outcomes pre-surgery and over post-surgery follow-up [20–24]. We therefore hypothesized that changes consistent with programmatic recommendations (e.g., reducing caloric consumption, increasing physical activity, weight loss) and improved psychosocial adjustment (quality of life, depression/anxiety symptoms or eating disorder pathology) would be observed.

Materials and Methods

The study was approved by the institutional review board at the Icahn School of Medicine at Mount Sinai and registered at ClinicalTrials.gov. Participants were recruited from the Mount Sinai Health System or from phone calls to the specialty clinic. Inclusion criteria were: enrolled in a bariatric surgery program, age 18–60 years, and fluent in English. Exclusion criteria included a Wechsler Abbreviated Scale of Intelligence IQ Estimate (vocabulary and matrix reasoning subscales) < 70 or history of developmental disability, history of neurological disorder or injury (e.g., moderate/severe head injury), current/lifetime DSM-5 bipolar disorder, schizophrenia, or psychotic disorder, current DSM-5 alcohol or substance use disorder, acute suicide risk, and pregnancy. In addition to screening for inclusion/exclusion, individuals providing informed consent for participation completed a standard psychiatric evaluation and anthropometric assessments. Eligible participants were told of their randomization at the end of the visit and given a wrist-based activity tracker (Fitbit model Flex 2, FibBit Inc, San Francisco, CA).

The trial design was a parallel group permutated block randomization (blocks 6–8 participants) with 1:1 randomization ratio for the digital health intervention or standard care (clinicaltrials.gov record # NCT03386006). Participants completed baseline and 8 week (end of treatment) evaluations via digital direct entry. Randomization and direct entry were all executed via StudyTrax software (https://www.studytrax.com/).

Intervention

Participants randomized to the digital health intervention received the app program and usual care in the bariatric program (see below). The program draws from the CDC’s National Diabetes Prevention Program, including: a weight loss goal, dietary goals (e.g., fat and/or calories, eating 3 meals/day), a physical activity goal (150 minutes of moderate activity/week), stimulus control and problem-solving, talking back to negative thoughts, identifying problem social cues, and time-management. Information is provided about eating and physical activity, exercise safety, health benefits of lifestyle changes, food substitutions, portion control, lifestyle activities (e.g., taking the stairs), stress reduction, weight loss maintenance, and reading nutrition labels. The platform uses a ‘home screen,’ or landing area to prompt users to tap and log meals and weights, read articles, answer questions, and develop a workout schedule. The home screen icons are a combination of standard prompts (food logging) and components ‘released’ using algorithms programed into the app. An automated physical activity tracker and a messaging interface for group members or a coach are also accessible. Articles developed for this project addressed caloric beverages, liquid nutrition, multivitamins, bariatric specific protein consumption, binge eating, grazing, emotional eating, positive coping, behavior activation, mindfulness, thought distortions, valued activities, social support, SMART goals, and bariatric-specific triggers (e.g., adaptations for holidays, etc.). Additional features include tracking mood, overall motivation, and sleep. A support group feature asked users to post personal challenges, get feedback from other users, and respond to others’ challenges. Expectations for the time spent on the app is 5–15 minutes per day.

Virtual coaches are trained in behavioral therapy, motivational interviewing techniques, and are supervised by a licensed clinical psychologist. Coaches help each user identify individual goals, focus on diet changes, exercise, and wellness, troubleshoot weight loss barriers, individualize feedback, find ways to adapt to lifestyle changes, and increase motivation. Coaches schedule weekly check-ins with users to review weight loss and review overall progress.

Full functionality of the intervention (app program + coaching) was provided at no cost.

Standard Care

Individuals randomized to standard care completed a psychosocial evaluation and all usual visits with the bariatric team. Although total amount contact with providers on the team varied depending on insurance requirements, a typical course of care included meetings with a nutritionist, monthly weights with a primary care physician, visits with the surgeon and/or bariatric nurse practitioner, and other assessments (e.g., endoscopy, etc.). As visits occurred as part of programmatic requirements and insurance approval, total amount of contact with the team was not expected to differ by condition.

Measures.

24 Hour Dietary Recall Interview.

The 24-hour recall is a common method [25] to collect information about intake and portion sizes. The ASA24^™ is a modified version of the interviewer-administered Automated Multiple-Pass Method, and is available from the National Cancer Institute (http://appliedresearch.cancer.gov/asa24/). Data from the 24 hours prior to the baseline and end-of-treatment assessments were collected.

Activity Tracker.

As above, participants wore an activity tracker to monitor steps, distance, calories burned, and active minutes during the intervention.

International Physical Activity Questionnaire (IPAQ) [29].

The IPAQ is a common brief self-report measure of time spent in an average week completing vigorous or moderate physical activity, walking, and sitting. The IPAQ has acceptable test-retest reliability and criterion validity (comparison against accelerometer) [29].

Eating Disorder Examination Questionnaire (EDE-Q) [26].

The EDE-Q measures dietary restraint, eating, shape, and weight concerns, loss of control eating episodes [27]. The EDE-Q has excellent content validity, validity generalization, and norms [28].

Depression Anxiety Stress Scales (DASS) [30].

The DASS is a 42-item self-report measure of negative emotional states with subscales of depression, anxiety, and stress symptoms used previously with bariatric candidates [31, 32]. Good to excellent internal consistency [30], and good convergent validity with other convergent measures [33] has been observed.

Short Form-36 Health Survey (SF-36) [34].

The SF-36 is a 31-item self-report of health-related quality of life measure that generates two scores: The Mental Composite Score (MCS), and the Physical Composite Score (PCS). Reliability and validity of the SF-36 is well-documented [35–36], and the measure has been used in bariatric populations [37–38].

Power and Statistical Analysis.

Mixed effects regression models estimated: (1) treatment effects for primary outcomes, and (2) Time × Treatment interactions for our hypothesis that the intervention would have a significantly greater impact after 8 weeks of treatment. All models controlled for baseline age, sex, and body mass index (BMI; kg/m²). With this design and n = 25 per group (N = 50), statistical power using GLIMMPSE [39] was estimated at p < 0.05 to be > 0.80 for effect sizes of d ≥ 0.68, a moderate to large effect size. All models were estimated in R version 3.6.3. [40]. Effect sizes were Cohen’s d with adjustment for overdispersed outcomes (variation higher than expected) when necessary using a shiny app (https://stefany.shinyapps.io/RcountD/).

Random effects regression models of treatment effects were intent-to-treat analyses (ITT) and included all available data. To account for statistical outliers and truncated or zero-inflated (excess of “0”‘s) outcomes within the sample, we use robust linear regression [41], zero-inflated negative binomial regression [42], and zero-one beta regression models respectively to reduce Type I error. R code and output are available upon request.

Results

Table 1 provides sample characteristics and Figure 1 summarizes CONSORT details for the sample size in each group and ITT models. The institutional review board only allowed data collection on individuals providing informed consent, or 55 individuals. Five participants were not randomized, including two individuals who were ineligible at screening, one individual chose not to continue participating, one individual received surgery before completing the study, and one withdrew without explanation. There was no significant difference in participants lost to follow-up (Chi Square = 0.0, p = 1.0) and all analyses report full ITT. Seven participants were lost in each group during the study, yielding 50 baseline and 36 follow-up visits for analysis.

Table 1:

Sample Characteristics

	Intervention	Control	p Value
Age (mean (SD))	39.72 (11.83)	37.40 (11.52)	0.486
Sex
Male (%)	7 (28.0)	1 (4.0)
Female (%)	18 (72.0)	24 (96.0)	0.054
Hispanic or Latino Origin
Yes (%)	16 (64.0)	12 (48.0)
No (%)	9 (36.0)	12 (48.0)
Unknown (%)	0	1 (4.0)	0.368
Race
Asian (%)	1 (4.0)	0 (0.0)
Black or African American (%)	8 (32.0)	8 (32.0)
Hispanic or Latino & Black or African American (%)	0 (0.0)	1 (4.0)
Hispanic or Latino & Unspecified Race (%)	2 (8.0)	1 (4.0)
Hispanic or Latino & White (%)	14 (56.0)	10 (40.0)
Mixed Race (%)	0 (0.0)	2 (8.0)
White (%)	0 (0.0)	2 (8.0)
Unknown (%)	0 (0.0)	1 (4.0)	0.333
Highest Level of Education
No School (%)	1 (4.0)	1 (4.0)
High School (%)	7 (28.0)	4 (16.0)
Some College (%)	11 (44.0)	13 (52.0)
Bachelor’s Degree (%)	4 (16.0)	3 (12.0)
Graduate Degree (%)	2 (8.0)	4 (16.0)	0.774
Total Annual Household Income
<$40,000 (%)	12 (48.0)	13 (52.0)
$40,000-$59,000 (%)	4 (16.0)	6 (24.0)
$60,000-$79,000 (%)	6 (24.0)	2 (8.0)
$80,000-$99,999 (%)	1 (4.0)	3 (12.0)
$100,000 or more (%)	2 (8.0)	1 (4.0)	0.438

Figure 1: CONSORT Diagram for Randomization and Intent-To-Treat Models.

Primary Outcomes

Table 2 summarizes the effects of the intervention on primary outcomes. Effects on BMI and total kcal consumed from the ASA24^® were small (ES = −0.26 and −0.12 respectively) and the Treatment × Time effects were not significant. After accounting for outliers, the intervention group lost about 0.3 BMI units more than TAU group members, or 0.66 lbs. No significant differences in kcal consumed were noted, although on average men reported consuming about 618 more kcal than women and had about 5.5 kg/m² higher BMI.

Table 2:

Primary Outcomes for Digital Intervention vs. Standard Care

Outcome	Treatment * Time	Effect Size	95% CI
BMI	−0.3 (0.5)	−0.3	−0.9, 0.4
ASA24	−101.6 (316.0)	−0.1	−0.9, 0.7
IPAQ Total	−0.50 (0.53)	−0.4	−0.8, 0.7
IPAQ Leisure	−0.7 (0.5)	−0.4	−0.7, 0.3
IPAQ Walking	0.3 (0.5)	0.3	−0.4, 2.1
IPAQ Moderate	0.4 (0.7)	−0.1	−0.6, 1.6
IPAQ Vigorous	−0.4 (0.7)	−0.6	−1.0, 1.9
FitBit Total	−0.004 (0.00001)***	−0.6	−0.8, −0.5
FitBit Sitting	−2.1 (0.6) ***	−0.1	−0.2, −0.1

Note.

^***p < .001

The IPAQ physical activity scores negatively skewed and log transformed prior to analysis. Several models were fit to account for heavy tails and overdispersion of distribution. Negative binomial models provided the best overall fit. Effect sizes ranged from 0.27 to −0.55 for the Time × Treatment interaction effects and none were significantly different than zero at p < 0.05. Intervention participants reported significantly less decline in physical activity than those in the control condition over the 8 week intervention. For continuous wearable activity data, model estimates indicated the intervention users retained about 56 min more baseline physical activity over the eight weeks [ES = −0.64, 95%CI = −0.76, −0.54] and experienced a 117 min greater decrease in sitting over the same period [ES = −0.14, 95%CI = −0.026, −0.06].

Secondary Outcomes

Table 3 summarizes psychosocial outcomes, including significant Time × Treatment interactions for Anxiety and Stress. A linear model for the DASS-depression scale scores was a poor fit due to floor effects (35% scored 0) and a truncated model yielded similar effects and effect sizes as the Stress and Anxiety scales. Figure 2 illustrates the change in DASS subscales. Effect sizes ranged from 0.58 to 0.62, suggesting moderate effects across all three scales.

Table 3:

Secondary Outcomes for Digital Intervention vs. Standard Care

Outcome	Treatment * Time	Effect Size	95% CI
DASS Stress	2.73 (1.34)*	0.58	0.03, 1.35
DASS Anxiety	0.96 (0.46)*	0.62	0.04, 1.31
DASS Depression	0.49 (0.64)	0.31	−0.41, 0.91
DASS Depression (>0 threshold)	5.57 (3.00)	0.58	−0 .03, 1.21
EDEQ Global	0.19 (0.34)	0.23	−0.46, 0.83
EDEQ Shape Concern	0.16 (0.45)	0.15	−0.54, 0.78
EDEQ Weight Concern	0.13 (0.54)	0.08	−0.54, 0.70
EDEQ Restraint	0.12 (0.68)	0.14	−0.67, 0.81
EDEQ Eating Concern	0.38 (0.28)	0.45	−1.83, 3.29
SF General Health	−5.90 (5.48)	−.033	−0.95, 0.27
SF Mental Health	−5.47 (5.46)	−.033	−0.99, 0.91
SF Vitality	−2.23 (6.38)	−0.12	−0.81, 1.54
SF Bodily Pain	−9.78 (9.68)	−0.31	−0.94, 0.28
SF Physical Functioning	−10.01 (7.38)	−0.42	−1.03, 0.58
SF Social Functioning	−12.94 (9.47)	−0.43	−1.08, 0.53
SF Role Physical	−0.35 (0.88)	--	--
SF Role Social	−1.44 (1.12)	--	--

Note.

^*p < .05

Figure 2:

Change in Depression Anxiety Stress Scale Subscales over the Eight Weeks of Assessment.

None of the Time × Treatment interactions for eating disorder symptoms by EDE-Q were significant. Point estimates of effect sizes ranged from 0.08 to 0.23, indicating small standardized differences in treatment effects. Robust regression results indicted no significant effects of treatment on quality of life from the SF-36; model estimated effect size estimates (ES = −0.33 to −0.43) were moderate. Both role limitation scales (Social and Relationship scales) had a limited response set (4 to 5 values), outcomes were therefore analyzed as ordinal response outcomes.

Utilization

Table 4 shows feature utilization frequencies of the intervention. On average, users opened the program 5 times per week with substantial variability within each day of the week (range 0–5 times opened). Coaches messaged individuals about 2× per week and users posted to the group approximately every other week, tagging each other’s posts with hearts less frequently. On average, group members entered their weights twice per week.

Table 4:

Summary of Engagement in the Digital Health Intervention

Engagement Behavior (per week)	Mean	SD	Range (days per week)
App Opens	5.46	1.72	0–5
Coaching Messages	2.27	2.53	0–27
Food Logs Entered	17.6	12.1	0–6
Group Posts	0.70	0.56	0–6
Group Posts Hearts (liked)	0.19	0.36	0–7
Group Comments	0.42	0.59	0–7
Group Comment Hearts (liked)	0.14	0.22	0–4
Weigh in	2.30	2.39	0–2

Note. Engagement activities all collected from N = 25 Intervention Users and averaged over week of trial. Ranges per day reported to provide estimate of local variability.

Conclusion

This study is the first randomized controlled trial of a digital health intervention for diet and physical activity, weight loss, and psychosocial adjustment prior to bariatric surgery. No significant intervention effects were found for BMI, total kcal consumed by 24 hour recall, or IPAQ with small effect sizes. Activity showed an unexpected pattern with intervention participants demonstrating significantly less decline in physical activity by IPAQ and more retention of baseline activity by wearable monitor than usual care participants. This finding may be the result of improved measurement precision as standard errors also shrank over time, suggesting an attenuation of physical activity and a slope reflecting decreases in both conditions, or a ‘last hurrah’ phenomenon prior to surgery. Moderate effects were noted for the secondary outcomes of anxiety and stress, which are consistent with in-person and telemedicine delivered interventions showing improved psychosocial outcomes pre-surgery and over post-surgery follow-up [20–22, 24]. However, significant differences were not observed for eating disorder symptoms and quality of life. Thus, the Phase I findings suggest preliminary efficacy for anxiety and stress symptoms among individuals pursuing bariatric surgery. Significant variability in use of digital intervention by group members may have affected these results, as some users engaged little and others participated often.

Limited data are available from randomized-controlled trials of eHealth specific to pre-operative bariatric surgery candidates for comparison of outcomes or effect sizes [43]. One randomized study of phone-based cognitive-behavioral therapy found significant improvements in binge eating, emotional eating, and overall health in the intervention group compared to standard care [44], suggesting that pre-operative eHealth interventions are useful for psychosocial symptoms but less potent for diet or exercise. Significantly more weight loss was observed at 12 and 24 months post-surgery with an iPad mini/MyFitnessPal eHealth strategy versus a comparison group [45]. Interventions altering behaviors that directly impact weight may therefore be most potent in the post-operative period.

A number of factors likely influenced the results including sample size and the duration and timing of the intervention. As a pilot study, only 25 individuals received the digital health intervention, and groups (n=5) were smaller than with commercially available products from the same company with coaching and group participation. Individuals who were expecting to undergo surgery sooner could have been more motivated to make changes to diet or exercise or engaged differently with the intervention. Further, the intervention was brief, and additional time might have been required to observe significant weight loss, similar to other programs (e.g., Diabetes Prevention Program). Limiting the digital intervention to 8 weeks was way for to avoid delaying the surgical process for participants, and to maximize cost-effectiveness and disseminablity. The intervention intensity is similar to a prior pre-operative eHealth trial [44] where the intervention group received 6 total sessions of 55 minutes over 7 weeks. In addition, smartphone comfort is an important part of evaluating any digital intervention, and while effects of age were not observed in this pilot study, there are well-documented moderating effects (favoring younger individuals) on acceptance and utilization of mobile health platforms [46]. It is therefore possible that including a wide range of ages in this study could have impacted the intervention outcomes. As the study was not blinded, participants could have been biased because they knew they were receiving a commercially-available smartphone app, and a digital control condition might have helped to disentangle any possible effects of this unblinded intervention. Only one 24-hour recall was administered to minimize cost and burden on participants; however, there are advantages of 3 days of records [47] and capturing only one day was a limitation. There are also several strengths of the study, including a sample of individuals from diverse ethnic and socioeconomic backgrounds and the sophistication of analytic techniques applied to the data.

In conclusion, this study provides useful data about a coach-led group-based smartphone app intervention for individuals seeking bariatric surgery. Additional research is needed to better delineate the optimal time and duration of the intervention. In addition, digital interventions provide an opportunity to scale specialty treatments typically only available from professionals with advanced degrees at a high cost per unit. Understanding whether treatments can be applied before or after surgery to augment weight loss or psychological outcomes is critical for individuals who do not achieve optimal results with bariatric metabolic surgery, as it will answer a critical question about when smartphone interventions are most valuable to patients.

Key Points.

• This randomized controlled trial tested a digital health program before bariatric surgery.

• No significant effects were found for body mass index or total kcal by 24-hr recall.

• Anxiety, stress, and depressive symptoms improved in intervention recipients.

• The intervention group did not show significant changes in eating disorder symptoms.