A protocol was registered with the International Prospective Register for Systematic Reviews (CRD42018096932), with the review structure following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Multimedia Appendix 1). We narrowed down the review question of the protocol, focusing on long-term weight management.

Textbox 1 summarizes the inclusion and exclusion criteria for the participant, intervention, comparators, outcomes, and study types of this systematic review.

We searched the following databases: Medical Literature Analysis and Retrieval System Online, Compendex, ScienceDirect, Cochrane Central, Scopus, and EMBASE through Ovid. Multimedia Appendix 2 outlines the search terms. Data published before 2008 were not included as these data are not reflective of the rapid change in the use of mobile phones and wearables. Keywords related to participant, intervention, comparators, and outcome items were used to search for relevant papers. A librarian was consulted for advice on the searches. The search was adjusted and modified for each database.

The references found were imported into EndNote X9 (Clarivate, Pennsylvania), and duplicates were removed. Overall, 2 reviewers conducted title and abstract screening. Any differences in the chosen studies were discussed until a consensus was reached. The full texts of potentially eligible studies were retrieved and analyzed for eligibility by 2 reviewers.

A standardized data extraction sheet was used to extract data. The extracted data included the title, the research question, the data sources, how the data were analyzed, the main findings, and the conclusions.

All included studies underwent a methodological quality appraisal. Relevant appraisal tools were used for different study designs. The Cochrane risk of bias tool was used for randomized controlled trials (RCTs), Risk Of Bias In Nonrandomized Studies-of Interventions was used for nonrandomized studies of interventions, A MeaSurement Tool to Assess systematic Reviews was used for systematic reviews, and 6 questions to trigger critical thinking were used for qualitative papers [13-16].

We have provided a narrative overview and tabular summary of the findings. A meta-analysis of the studies could not be conducted because of the heterogeneity in their interventions, participants, and outcomes.

We found 1116 references, and after removing duplicates and adding six references identified through searching reference lists of included studies, 684 titles and abstracts were screened (Figure 1). Furthermore, 44 full texts were assessed for inclusion, of which 28 were excluded (Multimedia Appendix 3). We included five intervention studies: four RCTs and one nonrandomized study of an intervention as shown in Table 1. In addition, we used insights from six systematic reviews, four commentary papers, and a dissertation for additional insights (Table 2).

There was a large variation in bias in all the included papers (Multimedia Appendix 4). In most of the RCTs, there was a high risk of bias because of the lack of blinding of participants, as a blinded version of a wearable device intervention is not possible. All the included RCTs used random sequence generation to generate groups, although information about allocation concealment was missing in half of them.

Table 3 outlines the studies that contributed findings to the fulfillment of the objectives of this review: physical activity and weight outcomes, duration of wearable technology use, accuracy of wearable technology over self-reporting, and use by people with specific medical conditions.

The three studies reporting on weight loss and physical activity outcomes concluded that using wearable devices had a benefit on these outcomes, but not compared with the comparator groups. Study 1 showed a significant rise in moderate-to-vigorous physical activity over 18 months and divided participants into high or low original weight loss and high or low weight regain groups. At 6 months, the high weight loss groups had significantly higher level of moderate-to-vigorous physical activity than the low weight loss group. However, at 12 and 18 months, the high loss and high regain groups’ level of moderate-to-vigorous physical activity fell, leaving the high loss and low regain group with a significantly higher level of moderate-to-vigorous physical activity than all other groups.

Study 2 compared 27 individuals using wearable devices with 260 individuals who were not using a wearable device. A total of 8000 steps per day was recommended for the intervention group, but it is not noted whether the comparator group were given similar recommendations. A significant benefit was only found at 2 years (P=.03).

In both the standard and wearable device groups of study 3, there was an increase in the duration of moderate-to-vigorous physical activity sessions ≥10 min over a 12-month period, but there was no statistically significant difference between the two groups. The percentage of weight lost did differ, with significantly greater weight loss in the comparator group compared with the wearable devices group from 12 months onward. Study 3 did not find an association between physical activity and weight loss within its groups.

Retention was fairly high in four studies, but study 2 did not provide data (Table 4). Study 4 was the only study mentioning to offer a monetary incentive for assessments (US $10-$25). Study 4 compared a standard intervention with a stepped-care intervention, where the intensity of support (such as telephone intervention and additional individual sessions) increased if certain goals were not met.

Accuracy was reported by two studies. Study 1 found that self-reported moderate-to-vigorous physical activity was significantly higher than that recorded by the wearable device. The groups with poorer outcomes (low loss or high regain) had larger discrepancies between the two methods. The self-reported and accelerometer-derived moderate-to-vigorous physical activities were most similar to the high loss and low regain group. the moderate-to-vigorous physical activity data from self-reporting and the accelerometer were, however, collected on different weeks. However, the high loss and low regain group still overestimated moderate-to-vigorous physical activity, and the overestimation did not reduce over time. Study 1 suggested “social desirability to report physical activity adherence,” with participants inflating self-reported moderate-to-vigorous physical activity.

Overall, three out of the five studies focused on populations with a specific medical condition (Table 4). This included a history of breast cancer (study 1), repair of failed Roux-en-Y gastric bypass (study 2), or prediabetes (study 5). Study 1 analyzed those who attended the visits but had invalid or missing data and found no significant difference in cancer treatment–related variables. Study 5 measured blood glucose (hemoglobin A_1c) levels and showed a significant beneficial reduction over 24 months.

This review showed some evidence that wearable devices can improve long-term physical activity and weight loss outcomes, but there was not enough evidence to show a benefit over the comparator methods. The comparator interventions differed among studies, which adds to the difficultly in determining the impact on outcomes. Although the term standard was used, there was no standardization in the comparators’ intervention, with different levels of support and procedures.

Overall physical activity levels increased from baseline, but there was no difference between wearable and comparator interventions. Study 1 found that those who sustained higher physical activity levels were more likely to maintain weight loss. Retention was fairly high in the included intervention studies. The mechanism through which wearable devices have an effect compared with other methods was not known as diet and physical activity were not different. The accuracy of wearable devices varied, which could be explained by the different features and technology of wearables. A total of three included studies focused on populations with a specific medical condition. The difference in populations added a challenge to comparing the studies as the results of a study on the weight management of patients with one medical condition may not apply to patients with another medical condition or the general population.

There were only five studies with a relatively small sample size assessing the long-term use of wearable devices. It was not possible to undertake a meta-analysis because of the heterogeneity among participants, wearables, methods, and outcomes. The included studies were likely to suffer from bias. Wearable device interventions cannot be blinded to the user. Only outcome assessors could have been blinded, which most studies did not attempt to do. The use of wearables in these studies may not be applicable to real-world scenarios as the companies selling these wearable devices do not offer the support that was offered by researchers in the studies. A limitation of this review is that we only conducted a basic search limited to a few keywords and phrases. In addition, databases such as Institute of Electrical and Electronics Engineers Xplore and Cumulative Index of Nursing and Allied Health Literature for clinical and behavioral science research were not searched.

It is important to retain participants in studies to separate the effect of the study design and intervention [31]. Study 5 compared a standard intervention with a stepped-care intervention where the intensity of support (such as telephone intervention and additional individual sessions) increased if certain goals were not met. Interventions of this kind have been shown to reduce attrition [32]. Other strategies for improving long-term data collection are offering incentives, reducing barriers by offering alternative data collection modes, and reminder calls [32]. Improving adoption and retention through methods such as monetary incentives could be counterproductive as this is not possible in real-life settings.

Consciously or subconsciously, self-reported physical activity levels are often overestimated [33]. Wearable devices are more accurate at estimating physical activity levels than self-reporting, though a truly objective method is currently not available for everyday purposes. Accelerometers, which were used in the wearable devices in the included studies, can lead to different estimates, even when using the same device [34]. The accuracy of heart rate monitors has been reported to be higher but still insufficient [35].

Wearable devices have shown benefit in managing medical conditions, eg, diabetes [36]. Studies with populations having medical conditions or risk factors could suffer from higher dropout because of the higher risk of a medical event. However, having a specific condition or medical event could be a stronger motivation than having a vaguer risk factor such as being overweight or obese [37].

Different aspects of weight loss maintenance and wearable devices have been studied, but large areas are still unknown. These include the mechanisms through which using wearable devices can lead to weight loss and studies into the usefulness of wearable devices for long-term weight management.

Those who managed to sustain raised physical activity levels had a weight maintenance benefit. Not all groups managed to sustain increased activity levels, and it would be valuable to understand why. Individuals who sustained exercise could have been more likely to commit to other lifestyle changes around weight management.

Investigating the reasons for dropout could help to understand to what extent this is caused by study design and/or flaws in wearable devices. Discovering how wearable devices are being used, and whether their use is improved through outside support, would give valuable information for designing more effective wearable devices. It could also help health care practitioners to advice and support people who are trying to lose weight and are interested in using wearable devices.

We found a small number of long-term studies showing some evidence that wearable devices can improve long-term physical activity and weight loss outcomes, but there was not enough evidence to show a benefit over the comparator methods. A major issue is the challenge to separate the effect of the decreasing use of wearable devices over time from the effect of wearable devices on the outcomes. Consistency in study methods is needed in future long-term studies on the use of wearable devices for weight loss.