Use of an mHealth Ketogenic Diet App Intervention and User Behaviors Associated With Weight Loss in Adults With Overweight or Obesity: Secondary Analysis of a Randomized Clinical Trial

Abstract

Background: Low-carbohydrate ketogenic diets are a viable method to lose weight that have regained popularity in recent years. Technology in the form of mobile health (mHealth) apps allows for scalable and remote delivery of such dietary interventions and are increasingly being used by the general population without direct medical supervision. However, it is currently unknown which factors related to app use and user behavior are associated with successful weight loss.

Objective: First, to describe and characterize user behavior, we aim to examine characteristics and user behaviors over time of participants who were enrolled in a remotely delivered clinical weight loss trial that tested an mHealth ketogenic diet app paired with a breath acetone biofeedback device. Second, to identify variables of importance to weight loss at 12 weeks that may offer insight for future development of dietary mHealth interventions, we aim to explore which app- and adherence-related user behaviors characterized successful weight loss.

Methods: We analyzed app use and self-reported questionnaire data from 75 adults with overweight or obesity who participated in the intervention arm of a previous weight loss study. We examined data patterns over time through linear mixed models and performed correlation, linear regression, and causal mediation analyses to characterize diet-, weight-, and app-related user behavior associated with weight loss.

Results: In the context of a low-carbohydrate ketogenic diet intervention delivered remotely through an mHealth app paired with a breath acetone biofeedback device, self-reported dietary adherence seemed to be the most important factor to predict weight loss (β=-.31; t54=-2.366; P=.02). Furthermore, self-reported adherence mediated the relationship between greater app engagement (from c=-0.008, 95% CI -0.014 to -0.0019 to c'=-0.0035, 95% CI -0.0094 to 0.0024) or higher breath acetone levels (from c=-1.34, 95% CI -2.28 to -0.40 to c'=-0.40, 95% CI -1.42 to 0.62) and greater weight loss, explaining a total of 27.8% and 28.8% of the variance in weight loss, respectively. User behavior (compliance with weight measurements and app engagement) and adherence-related aspects (breath acetone values and self-reported dietary adherence) over time differed between individuals who achieved a clinically significant weight loss of >5% and those who did not.

Conclusions: Our in-depth examination of app- and adherence-related user behaviors offers insight into factors associated with successful weight loss in the context of mHealth interventions. In particular, our finding that self-reported dietary adherence was the most important metric predicting weight loss may aid in the development of future mHealth dietary interventions.

Trial registration: ClinicalTrials.gov NCT04165707; https://ichgcp.net/clinical-trials-registry/NCT04165707.

International registered report identifier (irrid): RR2-10.2196/19053.

Keywords: acetone; biofeedback; diet; ketogenic; mobile apps; mobile phone; overweight; psychology; technology; telemedicine; weight loss.

Conflict of interest statement

Conflicts of Interest: JPL is volunteer chief scientific officer for the not-for-profit Institute for Personalized Therapeutic Nutrition and holds founder shares in Metabolic Insights Inc, a for-profit company that developed noninvasive metabolic monitoring devices. EJW is an equity holder at Keyto Inc and Virta Health. DAL is employed as a consultant for Keyto Inc.

©Kaja Falkenhain, Sean R Locke, Dylan A Lowe, Terry Lee, Joel Singer, Ethan J Weiss, Jonathan P Little. Originally published in JMIR mHealth and uHealth (https://mhealth.jmir.org), 14.03.2022.

Figures

Figure 1

Individual change in body weight (calculated as daily percentage change from baseline based on measurements recorded from an at-home Bluetooth scale) are shown for each participant over time throughout the duration of the study. Daily mean values over time for each group based on end-of-study weight loss at 12 weeks are represented in solid lines (dark orange, >10% weight loss; light orange, >5% to

Figure 2

App use metrics (yellow, average Keyto Level obtained through breath acetone biofeedback device; dark orange, average daily number of engagements with the Keyto app), self-reported dietary adherence (light orange), and average number of weekly weight measurements (brown) averaged across all participants throughout the intervention. Mean values are shown.

Figure 3

Aggregated responses to questionnaires sent through email, asking participants, “How does the following affect your ability to stick to your diet?” on a 4-point Likert scale, ranging from “Not at all” to “Every day.”.

Figure 4

Pairwise Spearman rank correlation matrix of adherence-, app use–, and diet-related variables. %BBW: percentage of baseline body weight; %EI: percentage of energy intake.

Figure 5

Average z-scores (the mean of the sample subtracted from the observed value divided by the SD of the sample at each time point) comparing time-series patterns of variable means between participants who lost >5% of initial body weight (light orange) and those who lost Keyto Levels, and (F) self-reported caloric intake.

Figure 6

Mean weight loss at the primary intervention end point of 12 weeks of participants in the lowest (left), medium (center), or highest (right) tertile of average self-reported dietary adherence assessed weekly through a questionnaire. One-way analysis of variance with Tukey post hoc tests comparing the differences in group means was conducted. Mean (dark orange) and SD (light orange) are shown. %BBW: percentage of baseline body weight.

Figure 7

Unstandardized effects with 95% CIs of the direct and indirect mediation effects of self-reported dietary adherence on (A) average Keyto Levels and (B) total engagement with the Keyto app on body weight loss at the end of the primary intervention phase at 12 weeks.

Figure 8

Response to the questionnaires sent through email, asking participants how (A) the COVID-19 pandemic has influenced their ability to stick to their diet on a scale of –5 (more difficult) to +5 (less difficult) and (B) how the listed items affected their ability to stick to their diet with respect to the COVID-19 pandemic on a 5-point Likert scale ranging from “None of the time” to “All of the time.”.