This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR mHealth and uHealth, is properly cited. The complete bibliographic information, a link to the original publication on https://mhealth.jmir.org/, as well as this copyright and license information must be included.
Intermittent fasting (IF) is an increasingly popular approach to dietary control that focuses on the timing of eating rather than the quantity and content of caloric intake. IF practitioners typically seek to improve their weight and other health factors. Millions of practitioners have turned to purpose-built mobile apps to help them track and adhere to their fasts and monitor changes in their weight and other biometrics.
This study aimed to quantify user retention, fasting patterns, and weight loss by users of 2 IF mobile apps. We also sought to describe and model starting BMI, amount of fasting, frequency of weight tracking, and other demographics as correlates of retention and weight change.
We assembled height, weight, fasting, and demographic data of adult users (ages 18-100 years) of the LIFE Fasting Tracker and LIFE Extend apps from 2018 to 2020. Retention for up to 52 weeks was quantified based on recorded fasts and correlated with user demographics. Users who provided height and at least 2 readings of weight and whose first fast and weight records were contemporaneous were included in the weight loss analysis. Fasting was quantified as extended fasting hours (EFH; hours beyond 12 in a fast) averaged per day (EFH per day). Retention was modeled using a Cox proportional hazards regression. Weight loss was analyzed using linear regression.
A total of 792,692 users were followed for retention based on 26 million recorded fasts. Of these, 132,775 (16.7%) users were retained at 13 weeks, 54,881 (6.9%) at 26 weeks, and 16,478 (2.1%) at 52 weeks, allowing 4 consecutive weeks of inactivity. The survival analysis using Cox regression indicated that retention was positively associated with age and exercise and negatively associated with stress and smoking. Weight loss in the qualifying cohort (n=161,346) was strongly correlated with starting BMI and EFH per day, which displayed a positive interaction. Users with a BMI ≥40 kg/m2 lost 13.9% of their starting weight by 52 weeks versus a slight weight gain on average for users with starting BMI <23 kg/m2
IF with LIFE mobile apps appears to be a sustainable approach to weight reduction in the overweight and obese population. Healthy weight and underweight individuals do not lose much weight on average, even with extensive fasting. Users who are obese lose substantial weight over time, with more weight loss in those who fast more.
Worldwide, 13% of adults have obesity (BMI ≥30 kg/m2) and 39% are overweight (BMI ≥25 kg/m2) [
IF is a set of dietary patterns commonly pursued for weight loss that limits the timing of eating without restricting food content. Studies have shown that various IF methods are effective for weight loss in people who are overweight [
Mobile apps for IF and weight tracking offer an opportunity to examine IF in a less-controlled setting and investigate its real-world efficacy for weight management. They are also a low-cost intervention for addressing obesity in the general population and may incentivize the adoption of healthy habits, including exercise and healthy eating [
We evaluated retention, fasting patterns, and weight change among users of 2 free IF tracking apps, collectively known as LIFE apps: LIFE Fasting Tracker (LFT), which is focused on fasting, and LIFE Extend (LX), which additionally supports tracking of physical activity, mindfulness, sleep, and healthy plant intake. From 2018 to 2020, the 2 apps acquired a combined user base of 2.5 million downloads. User accounts and backend data storage for the 2 apps are shared, such that fasts could be started in one app and stopped in the other, and all the data are interchangeable. LFT was launched over a year earlier than LX, so only a small fraction of the data in this study was generated via LX.
We followed nearly 800,000 users for retention and real-world fasting behaviors. We further analyzed weight change patterns relative to app use and demographics in a subpopulation of over 160,000 users who used apps to track their weight over time.
We showed that practicing IF with a dedicated mobile app is an effective and sustainable approach to weight loss in individuals initially classified as overweight and obese. Many users consistently used the apps to record fasts every week for months. Users with obesity lost substantial weight over time, with more weight loss in avid fasters. Weight loss in users with obese or overweight BMI was sustained, on average, for up to a year with little rebound. Conversely, users with a healthy or underweight BMI lost little or no weight, even with extensive fasting over 52 weeks.
We assembled all fasting and weight data for users who began using either of the LIFE apps between the launch of the LFT in May 2018 and December 2020. Analyses of fasting, retention, and weight are all relative to when the user began recording fasts in the apps, minimizing seasonal and other calendar effects. Other voluntary data collected were sex, age, race, height, diet, exercise frequency, stress level, smoking status, and primary health concern. For inclusion in our analyses, users had to have provided sex and date of birth and recorded at least one fast, the first of which had to have been started on or after their 18th birthday.
For the weight change analyses, we further required the user to have provided height and an initial weight recorded within 7 days of the first fast. Heights and weights had to have been within validation ranges of 145-203 cm and 25-249 kg, respectively. Height and weight can be entered in either metric or imperial units, with subsequent conversion to metric units for storage and analysis. We identified 902 users whose weight change at weeks 1 to 52 was >5 SD from the average across all users for that week. Without knowing which value or values were presumably misentered, we simply excluded those users entirely from the weight analysis, made feasible by the study’s large sample size. Weights were subject to a 24-hour
We assembled all fasting records for the full set of nearly 800,000 users. Although the apps allowed shorter and longer fasts to be tracked, we eliminated fasts under 8 hours and truncated fasts to a maximum length of 240 hours. To reduce the effects of forgotten fasts that were ended and saved in the apps long after eating had resumed, we eliminated any fast that was 120 hours or longer but where a fasting goal of under 24 hours had been specified by the user. This yielded 25,983,817 fasts for our analyses.
We aggregated fasting statistics for each user for weeks 1 to 104 but primarily investigated weeks 1 to 52. Information regarding week 53 to 104 was used, when available and applicable, to determine retention. For each week, we totaled the number of started fasts, the average fast length, and the sum of hours beyond the first 12 in a fast, which we call extended fasting hours (EFH), and EFH per day (ie, EFH per day=sum of EFH for all fasts started in a week/7). The 12-hour time point in a fast is when the body is expected to have depleted the energy from recently consumed food and may begin a metabolic switch to deriving energy from the breakdown of fat [
We also calculated the cumulative means of these measures for all weeks, up to and including the given week.
We assessed user retention based solely on records of completed fasts and not on other user behavior such as log-ins or use of other app features. Starting with the date of each user’s first fast, we assessed their fasting activity for each week. The most restrictive definition of retention is when a user is only considered retained so long as they record a fast in each consecutive week. We refer to this definition as retention with a 0-week grace period. In contrast, the most lenient definition of retention is where the user is considered retained the entire time between their first and last recorded fast, regardless of how much activity they have in between. We refer to this as retention with an unlimited grace period. This definition is also sometimes called rolling retention [
We explored retention by varying the number of weeks in the grace period. We looked at 0, 2, 4, 8, 13, 26, and unlimited-week grace periods. After considering this spectrum of retention metrics, we decided to apply the 4-week grace period retention definition for all subsequent analyses. For example, if the user recorded no fasts in weeks 10 to 13 but did fast in week 14, the user was still considered retained in weeks 10 to 14, but if they resumed fasting in week 15 or later, their retention would have ended with week 9. Note that our univariate estimates of retention are conservative because many users start near the end of our data collection period, thus not having the opportunity to be counted as active in the app during the full 52 weeks (plus the grace period) that they might otherwise have counted toward. In the multivariate analysis, we used right censoring to account for this issue.
Users were included in the weight change analyses for all weeks for which they satisfied the 4-week grace period retention definition and in which they had a recorded weight. To account for the wide range of starting weights, weight change was analyzed as percent change from the user’s starting weight. The effect of obesity was also considered in some analyses by stratification on starting BMI. We categorized BMI using the Centers for Disease Control and Prevention definitions [
For the weekly weight aggregates, we calculated mean weight and the number of weights recorded during the week. The baseline weights were excluded from the week 1 aggregates.
We performed all data analysis using Python 3.9 libraries in a JupyterLab [
This study was exempt from institutional review board approval per Indiana University’s research guidelines [
A total of 792,692 users satisfied the inclusion requirements for the fasting and retention analysis. The detailed demographic and biometric data for this population are presented in
User retention, calculated by different grace periods of inactivity. In total, 792,692 users were tracked starting from their first recorded fast. Grace periods extended out to 104 weeks for the unlimited definition.
Intermediate grace periods corresponded naturally to intermediate retention rates (
Retention using the 4-week grace period differed substantially by several demographic criteria (
Hazard ratios with 95% CIs for failure to retain. Based on the Cox proportional hazards model over the 52-week study. HR=1 corresponds to the reference values: female, age <30 years, primary health concern as weight loss, starting BMI in the normal low category, white, typical western diet, sedentary, and never smoker. HR<1 reflects higher retention rates.
Even when retained, user fasting behavior is likely to change over time. We examined fasting patterns based mainly on the first 26 weeks among users retained that long. The 26-week period is long enough to see what long-term use of the fasting apps is like, while affording a larger sample size than looking only at users who were retained at 52 weeks. It also avoids overweighting the first few weeks of use when we had the largest sample but while users were still establishing their fasting routines.
The most common days to start a fast were Sunday, Monday, and Tuesday, whereas Friday and Saturday were the least popular. We also examined the distribution of fasts per user per week over the first 26 weeks for 54,811 users retained at 26 weeks using the 4-week grace period. The mean frequency was 4.25 (SD 1.91) fasts per week. Fasting frequency was approximately bimodal, with a broad peak centered on 3 fasts per week and a sharp peak at 7. Slightly more than one-quarter (13,981/54,881, 25.5%) of the users fasted 6 to 7 times per week. In
Fasting frequency statistics for users retained at 26 weeks, averaged over the first 26 weeks of use and grouped by self-reported race. Bins are half-fast width, left-inclusive, and include 7 in the highest bin.
The most common fasting length of the 26 million fasts analyzed over the entire length of the study was 16 hours. The mean and median lengths were 21.0 and 18.0 hours, respectively, while the lower and upper quartiles were 16.1 and 20.9 hours.
Histogram of fast lengths and a log scale histogram inset.
Distribution of starting and ending hour of fasts in local time.
We also examined the average fast length by user for the 54,881 users who were retained at 26 weeks under the 4-week grace period definition.
Distribution of average fast lengths per user across the first 26 weeks for users still retained at 26 weeks, broken down by weekly fasting frequency.
Combining fasting length and frequency, the cumulative mean EFH per day was 5.0 at 26 weeks, which would correspond to a daily fasting routine of 17 hours.
We analyzed weight change for the 161,346 users who met the 4-week grace period retention criteria and recorded multiple weights in the fasting apps. From the univariate perspective, weight change as an outcome varied by several factors, including age, primary health concern, starting BMI, and EFH per day (
To address the correlation and confounding among variables, we conducted an ordinary least squares regression analysis of weight change at the 26-week time point. At 26 weeks, there were 1252 users with a recorded weight and values for all input variables. The result was that the only factors with
Regression coefficients with 95% CIs for weight change at 26 weeks. The model was built using ordinary least squares linear regression for the 1252 users who had answers for all variables and a weight recorded in week 26 (
We further graphically explored the 52-week patterns of weight change relative to EFH per day and starting BMI, which emerged as the main variables explaining variability in weight change.
Weight change over time, stratified by users’ cumulative average extended fasting hours (EFH) per day. Excludes users with underweight starting BMI. Mean values are plotted with 95% CIs represented by shading.
Weight change over time, stratified by user’s starting BMI category. Mean values are plotted with 95% CIs represented by shading.
We examined the combined effects of starting BMI and fasting quantity by plotting the EFH per day strata separately for each starting BMI category (
Weight change over time, stratified by user’s starting BMI level and cumulative average extended fasting hours (EFH) per day. Mean values are plotted with 95% CIs represented by shading.
We also examined the number of users who achieved certain thresholds of weight loss.
Percentage of users with obese or overweight starting BMI (≥25 kg/m2) who achieved 5%, 10%, 15%, and 20% weight loss by week.
This study is the largest examination of IF conducted to date and is orders of magnitude larger than any previous effort [
The spectrum of retention metrics (
Retention statistics for mobile apps are not commonly available for proprietary apps. An analysis by AppsFlyer found that day 30 retention (fraction of original users active on day 30) for health and fitness apps in the United States in 2020 averaged <6% [
Age was the best predictor of retention in our study, which is consistent with other analyses of retention predictors for lifestyle interventions [
Several other variables were notable in their relationship to retention in both univariate and multivariate analyses (
The real-world spectrum of fasting behavior documented in our study shows variable and flexible adherence to IF regimens, making specific idealized fasting protocols hard to discern in the data. We did clearly see a group of 25.5% of users who practice a daily, or nearly daily, fasting routine (≥6 days per week) averaged over the first 26 weeks, whereas the rest skip multiple days per week. Among users who fast, on average, fewer than 3 times per week, the majority fast under 24 hours, suggesting that they are more sporadic in their fasting or less vigorous in tracking in the apps. A total of 7.5% of users had average fasts over 32 hours, likely corresponding to the extended paradigms of IF such as 5:2 and alternate day fasting.
Owing to the multidimensional gradations of fasting patterns, we proposed the concept of EFH per day, as a metric to quantify fasting across all users. EFH per day combines fasting frequency and fasting length into a single measure and serves to unify the various fasting regimens for analysis. EFH per day was predictive of weight loss in a nearly linear fashion (
To explain these results, we hypothesize that the correlation between EFH per day and weight loss and the interaction with starting BMI can be attributed primarily to differences in caloric restriction. In previous studies, those who practiced alternate day fasting, the 5:2 diet, or time-restricted eating reduced their daily calorie intake by 10% to 30% [
The weight loss effects of longer fasts may additionally be driven by the metabolic switch from glucose to ketones derived from fat tissue and free fatty acids [
Analysis of weight loss threshold achievement facilitates the comparison of IF with other interventions. By 26 weeks, 67.2% (1475/2194) of users with overweight and obese starting BMI lost at least 5% of their starting weight. This is comparable to the results achieved by users of the Diabetes Prevention Program through the Noom platform, 64% of whom lost over 5% of their body weight in 24 weeks [
While it is common for people who lose weight to be subject to weight regain [
Age was positively associated with greater weight loss, consistent with previous findings [
As of 2016, close to 50% of adults in the United States had tried to lose weight within the preceding 12 months according to Centers for Disease Control and Prevention data [
Self-reported data in this mobile setting offered many intriguing correlates of retention and weight loss. Meaningful factors included diet, exercise, stress, and smoking, all of which lend themselves well to mobile tracking, including integrated wearables, for reliable measurement and analysis.
To better understand the mechanisms and residual variability of weight loss by app users, we would like to study the caloric input and expenditure of users directly. This can be achieved by asking users to record their daily dietary intake and exercise. Tracking exercise is amenable to passive tracking with wearable technology, and many users of the LIFE apps (specifically LX) already have regular data ingestion established with the most common fitness trackers. Such an additional study could help resolve the somewhat surprising finding that the diet and exercise habits self-reported at the beginning of the study did not correlate with weight change.
Although weight management is a clinically important objective, other clinically relevant outcomes could be measured and correlated with fasting behavior. These include assessments of mental and physical health, disease incidence, insulin resistance, medical costs, and professional and educational absenteeism. Advocates of IF point to studies in animal models and humans that suggest many of these benefits [
Finally, studies show that social support improves health and well-being, and that people who have strong support networks are more likely to lose weight than those who do not [
The primary limitation of this study was that most data were self-reported, except for some weight values that were entered by smart scales. This limitation is compensated for by the large sample size of the study.
The observed weight change averages may be potentially confounded by users who stopped recording weights or even stopped using the app because of a lack of progress. Conversely, users who achieved success might have been less motivated to continue recording fasts and weights. Similarly, users may have been more likely to weigh themselves and record their weight in the app if they had lost weight, which could then exaggerate the weight loss estimates in this study. These effects may be challenging to untangle, but the trends and correlates of weight change identified should be robust.
Owing to the limited observational nature of this study, users who fasted longer may have adopted other diet-related practices more than users who fasted less without our knowledge. Similarly, we did not have information about users’ previous experience with IF. Given that the largest retention losses and greatest rates of weight change occurred in the earliest weeks of app use, previous fasting experience or even recent weight changes could skew the reported progress. In an IF-naive population, we expect weight loss to be relatively larger.
Another limitation of the study is that, due to being observational, it lacked explicit controls. However, this limitation was offset by the wide range of fasting behaviors among users. We used this natural variability as a form of self-directed intervention, which allowed us to contrast and quantify the effects of different levels of fasting on a much broader scale than would be feasible for a randomized controlled trial.
Baseline demographics and retention at 13-week intervals using the 4-week grace period retention definition. Hazard ratios and corresponding
Weight change at weeks 13, 26, 39, and 52 using the 4-week grace period retention definition relative to user demographics. Starting BMI (mean, SD) for each demographic category is included for context. Sample sizes reflect the maximum eligible users at each time point. Regression coefficients and
extended fasting hours
intermittent fasting
LIFE Fasting Tracker
LIFE Extend
The authors would like to thank Steven Bray for his close reading of the manuscript. The authors would like to appreciate the detailed and constructive feedback from the reviewers. RCD was affiliated with LifeOmic at the time of the study and is currently affiliated with Infinia ML.
BAS performed most of the data collation and generation of figures and tables with substantial assistance from RCD and JPB. All major decisions were discussed and agreed upon by all authors. All authors contributed to writing the manuscript. Statistical approaches were guided by SP and executed by BAS. LT conducted most of the literature review.
BAS, LT, RCD, JPB, and SF are or were employees of LifeOmic, the makers of the apps studied in this paper, and they have ownership rights in the company. No one at LifeOmic other than the authors had any editorial oversight in performing this study or writing this paper.