Effects of using wearable devices on reducing sedentary time and prolonged sitting in healthy adults: a network meta-analysis

Abstract

Objectives

This study aimed to compare the effectiveness of different wearable intervention strategies in reducing sedentary time (ST) and prolonged sitting (PS) on healthy adults.

Design

A network meta-analysis (NMA).

Data sources

PubMed, Web of Science, SPORTDiscus, ProQuest, Opengrey, Medline and Cochrane Central Register of Controlled Trials were searched up to 1 June 2024.

Eligibility criteria for selecting studies

Randomised controlled trials (RCTs) that examined the effect of wearable device interventions on ST and PS among healthy adults were included.

Data extraction and synthesis

Two independent reviewers used standardised methods to search, screen and code included studies. Bias risks were assessed using Cochrane tools (Risk of Bias 2.0). Data were analysed using a frequentist framework NMA to directly and indirectly compare the effects of the five different intervention strategies (comparators). The results were reported as standardised mean differences (SMDs) with 95% CI and surface under cumulative ranking curve (SUCRA) was used to rank the best interventions. The five comparators were as follows: (1) wearable-only intervention (only using wearable devices for self-monitoring); (2) wearable combined with online intervention (ie, online coaching and social media support); (3) wearable combined with offline intervention (ie, face-to-face seminars and courses); (4) comparison group (ie, traditional, non-wearable interventions); (5) control group (ie, maintaining daily routine, waitlist).

Results

12 RCTs with a total of 2957 participants were included. Results of NMA showed that the ‘wearable+online’ has significantly better effects in reducing ST compared with control group, comparison group and ‘wearable only’, with moderate to large effect sizes (SMD=0.96, 95% CI 0.65 to 1.27; SMD=0.87, 95% CI 0.21 to 1.53; SMD=0.78, 95% CI 0.14 to 1.42, respectively). However, no significant differences were identified between the groups in reducing PS. The SUCRA values were ranked as wearable+online (98.1%), wearable+offline (64.4%,), ‘wearable only’ (40.5%), comparison group (25.9%) and control group (21.1%) for ST reduction. Similar rankings were observed for PS reduction, with probabilities of 69.9%, 61.1%, 59.7%, 37.1% and 22.1%, respectively.

Conclusions

Wearable+online is the best intervention strategy for reducing ST in healthy adults. Additionally, none of the wearable-based interventions effectively reduced PS in healthy adults, but as there is little research on PS, it should receive more attention in the future.

PROSPERO registration number: CRD42021290017.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This is a network meta-analysis (NMA) within a frequentist framework to compare and rank wearable-based interventions for reducing sedentary time in healthy adults.

• Strong and reliable methodological and statistical procedures were applied.

• The risk of bias was assessed using the best practice and the latest Cochrane collaboration tool.

• It was not possible to adjust for differences in statistical analysis across randomised controlled trials.

• Insufficient studies with a low risk of bias limited our ability to conduct a standalone NMA.

Introduction

Sedentary time (ST) is an independent risk factor for many health issues.1,4 According to a recent study, ST was associated with a higher risk of 12 of 45 types of non-infectious chronic disease comparing>6 hours/day to≤2 hours/day ST.⁵ A cohort study revealed a 21% reduction in mortality rates for infectious diseases (0.78 (95% CI 0.72 to 0.87)) and pneumonia (0.79 (95% CI 0.67 to 0.94)) among individuals with low ST compared with those with high ST.⁶ Conversely, breaking up prolonged sitting (PS) or replacing it with low-intensity physical activity (PA) has been shown to lower postprandial glucose and insulin response, as well as improve cardiovascular disease risk.7,9 However, the prevalence of ST among adults worldwide is concerning.¹⁰ A survey of four European countries showed that the average ST among adults was as high as 530 min/day, with 23% exceeding 10 hours/day.¹¹ Academics have researched intervention strategies for ST at the individual, environmental and policy levels.¹² These involve education, goal setting, workplace adjustments and community initiatives, often incorporating behaviour change techniques (BCTs), such as self-monitoring and problem-solving.¹³ However, evidence for effective ST interventions in adults is limited, with most studies focused on workplaces. Scaling-up strategies are challenging due to time and cost constraints, leading to suggestions for digital self-monitoring to raise awareness and reduce ST.¹⁴

Wearable technology that is convenient and affordable may offer a potentially efficient answer to the aforementioned problems as technology advances. Global Status Report on PA 2022 published by WHO reports that one-thirds of countries have implemented mobile health (mHealth) programmes in the past 2 years.¹⁵ mHealth involves using mobile devices to deliver health services and interventions, and it often integrates with wearable technology. Wearables, such as fitness trackers, provide real-time data on PA and ST enhancing mHealth’s effectiveness in managing health.¹⁵ Evidence-based research has demonstrated the critical role that wearable technology plays in medical treatment,¹⁶ individualised health supervision¹⁷ and other sectors.¹⁸ Similarly, ST is difficult to monitor and change in daily life due to its prevalence, and the use of wearable devices with sedentary alert and monitoring capabilities may be able to change habitual lifestyle patterns and contribute to the development of self-health management behaviours. For example, wearable devices, such as Fitbit and Jawbone, have PS reminders and can properly track ST in real time.^{19 20}

Previous systematic reviews have explored the role of wearable devices in enhancing PA. For example, Wang et al demonstrated the effectiveness of a wearable device-based PA intervention in managing obesity in children and adolescents,²¹ while Ferguson et al highlighted their benefits for increasing PA and improving health outcomes across various populations.²² Similarly, several reviews have assessed the impact of wearable devices on both PA and ST.23,26 However, these reviews primarily focused on PA, with intervention strategies predominantly targeting PA rather than ST, and often included diverse populations without specifically addressing healthy adults. However, ST, which is often an unconscious behaviour, tends to be overlooked by healthy adults and can lead to significant health issues. Addressing ST is, therefore, crucial for disease prevention in healthy individuals. Various wearable devices are currently available for healthy adults, including smartwatches, such as Fitbit and Jawbone.^{27 28} Wearable devices could also be paired with online interventions, whereby support is provided through apps or other online platforms29,34; alternatively, some interventions combine wearable devices with offline components, such as face-to-face seminars, courses or workshops, offering in-person guidance and supervision.35,38 Therefore, a comprehensive understanding of these interventions and their comparative effectiveness is crucial for identifying optimal strategies for reducing ST.

Network meta-analysis (NMA) is an emerging method of data comparison, which assesses three or more interventions simultaneously using indirect comparisons or a combination of indirect and direct comparisons.³⁹ The application of this statistical method is very developed in the medical and health fields and is now gaining attention in the field of PA. It has been suggested that the application and importance of NMA in the field of PA and health promotion are important issues to be addressed.⁴⁰ Therefore, our review aims to perform an NMA to compare the effects of different interventions based on wearable devices on reducing ST and PS in healthy adults and to test the hypothesis that different intervention methods of wearable devices are superior to the control group in reducing ST and PS.

Methods

This review was reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)⁴¹ and the PRISMA extension statement for NMAs.⁴² A completed checklist can be found in online supplemental appendix 1. The study protocol was registered in PROSPERO (registration number: CRD42021290017). The details of protocol modifications are in online supplemental appendix 2.

Patient and public involvement

Patients or the public were not involved in the design, conduct, reporting or dissemination plans of our research.

Search strategy

A systematic literature search was conducted to identify relevant studies in seven electronic databases: PubMed, Web of Science, SPORTDiscus, ProQuest, Opengrey, Medline and Cochrane Central Register of Controlled Trials. The core keywords identified include adult, wearable and “sedentary behavior”. PubMed Medical Subject Headings database and other search engines were used to find synonyms of keywords, including the following four groups

Population

adult* OR women OR men OR worker* OR employee* OR college* OR university.

Intervention

wearable OR wristband OR “smart band” OR “smart bracelet” OR smartwatch OR “Apple Watch” OR bang OR Fitbit* OR Pebble OR Jawbone OR Misfit OR Garmin OR “activity tracker*” OR “activity monitor” OR XiaoMi OR HUAWEI OR Google OR Samsung OR Polar OR samsung OR fitmeter OR Withings;

Outcomes

inactivit* OR “health behavio*” OR “sedentary behavior*” OR “sedentary behaviour*” OR “sedentary lifestyle” OR “screen time” OR “screentime” OR “screen-viewing” OR “sitting time” OR “prolonged sitting”

Study design

“randomized controlled trial” OR “clinical trial”. The search period was all-inclusive up to 1 June 2024. A ‘snowball’ search was also conducted to identify additional studies by examining the reference lists of publications eligible for full-text review and screening studies that cited them. The search strategy was adapted to the search language and rules of each database (online supplemental appendix 3).

Initially, two reviewers (ZH and GZ) searched the database and exported all studies to the online literature management tool Rayyan⁴³ to remove duplicates. Subsequently, two independent reviewers (ZH and CL) screened all titles and abstracts in duplicate to identify articles eligible for full-text assessment. Both reviewers (ZH and CL) then independently and, in duplicate, reviewed the full texts of the selected articles to determine their eligibility. Additionally, two reviewers (ZH and CL) manually reviewed the reference lists of relevant original studies and review studies. Disagreements were resolved through group discussions with a third reviewer (RW).

Inclusion and exclusion criteria

Inclusion and exclusion criteria, according to population, intervention, comparison, outcomes and study, were as follows.

Population

Participants were healthy adults (18–59 years old), excluding the elderly (≥60 years old), overweight and obese people, and other people with physical and mental diseases; additionally, excluding research settings in hospitals, health centres and chronic disease centres.

Intervention

The primary intervention strategy consisted of self-monitoring based on wearable devices (eg, smartwatch and wristband) to reduce ST or PS. We defined three intervention categories: use of the wearable device alone with self-monitoring only (‘wearable only’); use of a wearable device with online support from researchers (wearable+online) and use of a wearable device with offline support from researchers (wearable+offline); excluding the research on non-wearable devices interventions (eg, smartphone and computer) and the goal of intervention did not include reducing ST or PS. Additionally, no exclusions were made based on the length of intervention.

Comparison

Comparison group (ie, traditional, non-wearable interventions) or control group (ie, maintaining a daily routine, waitlist, where participants receive the intervention after the study).

Outcomes

Indicator includes ST, defined as the time spent for any duration (eg, minutes per day) or in any context (eg, at school or work) in sedentary behaviors,⁴⁴ and PS, defined as continuous ST for more than 30 minutes^{30 45}; both outcomes were measured either subjectively (eg, questionnaires) or objectively (eg, accelerometers), in minutes or hours per day.

Study design

Randomised controlled trials (RCT), crossover design, cluster RCT or clinical trials. Additionally, we only included studies published in English.

Data extraction and management

We designed a data extraction form. Two researchers (GZ and ZH) separately extracted data from the included studies and entered them into Microsoft Word tables. Disagreements were resolved through group discussions with a third reviewer (RW). Mean differences and their associated SDs were extracted from the included studies. Therefore, studies were converted to SD for analysis if they reported SE, CI or quartile. We only extracted the data from the baseline and postintervention, and the follow-up data were not included in the analysis. When there were several publications from the same project, the study with the longest intervention was selected. Data from all phases of the crossover trial were included in the analysis. In addition, we extracted basic information (year, country, first author, mean age, gender and sample size), intervention characteristics (types of interventions and intervention duration) and outcomes (measurement methods, statistical analysis and results). In case of missing data, the authors were contacted to obtain them a minimum of 3 times over 4 weeks.

Risk of bias and quality assessment

The risk of bias for RCTs was assessed independently using the Cochrane risk of bias assessment tool (Risk of Bias 2.0).⁴⁶ Two cluster-randomised trials and one crossover trial used Risk of Bias 2.0, which is appropriate to the design.⁴⁶ Detailed bias assessment can be found in online supplemental appendix 4. Each item was assessed at three levels: low, unclear and high risk of bias. Overall classification of low, unclear or high risk of bias in each study was based on the combination of the items. Bias in outcome measurement was assessed according to the measurement method used: objective measures were classified as low risk of bias, while subjective measures were deemed high risk of bias.^{47 48} Additionally, if the study’s outcomes included ST and PS, the risk of bias assessment was assessed separately. Two researchers (ZH and CL) were involved in the assessment process independently. Then, discrepancies were consulted by a third author (GZ).

Statistical analysis

Transitivity is the core assumption of indirect comparative effectiveness. We first made an epidemiological judgement to evaluate whether the distribution of cross-study effect modifiers allowed reliable indirect comparison. We a priori selected gender, age and weight status as potentially important effect modifiers. Four researchers (ZH, GZ, RW and CL) discussed the network transitivity, and we decided that gender was not an important effect modifier, but age and weight status might be important effect modifiers. Therefore, we decided to exclude any studies that included elderly or obese people from the review.

NMA was conducted separately for ST and PS outcomes to assess the effects of different interventions on each outcome. All the outcomes data were converted into minutes per day. The formulae for the mean and SD pre- to postchange values were as follows: ‘Mean_change=Mean_post − Mean_pre’ and ‘SD_change=SQRT ((SD_pre² + SD_post²) − (2×Corr×SD_pre×SD_post))’. According to the Cochrane Handbook for Systematic Reviews of Interventions version 6.4, the correlation coefficients of all the intervention groups and the control group were calculated, respectively.⁴⁶ Standardised mean differences (SMDs) with 95% CI were calculated in this study due to the differences in measurement scales.⁴⁶ A higher value measured in minutes per day indicates that ST or PS is worse. Therefore, a negative SMD would indicate an improvement in these outcomes. To help interpretability, we multiplied the SMDs and the corresponding CIs by −1 so that a positive value indicates an improvement in the outcomes. SMD values<0.2, 0.2<0.5, 0.5<0.8 and 0.8 were categorised as trivial, small, moderate and large effect sizes, respectively.⁷ To address multiple comparisons, we applied the Bonferroni correction, setting the p-value threshold to 0.05/n, where n represents the number of comparisons made.^{49 50} Thus, p≤0.005 was defined as statistically significant difference. This adjustment helps to minimise the risk of errors in our analysis. In addition, for cluster RCTs and crossover trials, we used approximate methods to adjust the relevant data.⁴⁶

To help distinguish the most effective intervention strategies, we referred to the classification of wearable devices as described by McDonough et al to define the nodes for the NMA.⁵¹ Studies were divided into five comparison groups: (1) ‘wearable only’ (only conducting self-monitoring); (2) wearable+online (ie, online coaching and social media support); (3) wearable+offline (ie, face-to-face seminars and courses); (4) comparison group (ie, traditional, non-wearable interventions); (5) control group (ie, maintaining daily routine, waitlist). NMAs were conducted by using a frequentist framework via a random-effect model in STATA (produced by StataCorp, College Station, TX, USA, https://www.stata.com/) and the mvmeta command.⁴⁶ Consistency was a statistical manifestation of transitivity and the global inconsistency test compares the fit and resolving power of the consistent and inconsistent models. Node-splitting method was used to evaluate local inconsistency. Additionally, we assessed inconsistency and loop heterogeneity separately in each closed loop of evidence using the ‘ifplot’ command in STATA. The tests for local and global inconsistency indicated that there was no inconsistency in the network for either outcome, which suggested that the consistency model was appropriate. Finally, STATA was used for direct and indirect comparisons, and forest plots and network plots were drawn. The interventions were ranked according to the overall probability of surface under the cumulative ranking curve (SUCRA), ranging from 0% to 100%, according to the outcome, with larger values indicating better intervention effects.

We planned to conduct sensitivity analyses by repeating the NMA for each outcome in the following subgroups: short-term (less than 12 weeks) versus long-term (greater than or equal to 12 weeks) intervention duration; high risk of bias studies versus low or unclear risk of bias; objective versus subjective outcome measurement.^{46 51}

Result

Study selection

A total of 6702 potentially relevant studies were searched in databases, and 984 duplicates were removed. By screening the titles and abstracts, 5542 studies were excluded. Eligibility was assessed for 174 full-text studies, 162 of which were excluded from the NMA. Finally, the remaining 12 studies were included, and the flow of the systematic review is presented in figure 1.

Figure 1. Flow diagram of study selection. RCT, randomised controlled trials.

Study characteristics

The included studies were published between 2013 and 2022, with eight of them after 2018. Most studies were from Europe and America, including the ^{28 34}USA (n=2), the ^{31 32}UK (n=2), Belgium²⁷ (n=1), Ireland³³ (n=1) and Finland³⁶ (n=1); the rest were from Australia^{29 35} (n=2), Japan (n=2) and Taiwan³⁷ (n=1). The total sample size was 2957. Three studies used wearable devices solely for self-monitoring, with participants using the devices independently and without implementer involvement. Six studies incorporated wearable devices with online interventions, using internet platforms to provide education, training or motivation through non-contact methods, such as website courses, coaching or messaging services. Another three studies combined wearable devices with offline interventions, including face-to-face seminars and courses. These studies aimed to reduce ST or PS by altering the participants’ environments, providing social support and implementing reward systems. Five studies were comparison groups, which mainly used traditional interventions (ie, traditional, non-wearable interventions) to compare with the intervention based on wearable devices (table 1) (online supplemental appendix 5). In addition, 11 studies on ST were identified: 2 used wearables alone (n=46), 6 combined wearables with online interventions (n=713) and 3 combined wearables with offline interventions (n=215). For PS, 6 studies were identified: 2 used wearables alone (n=62), 3 combined wearables with online interventions (n=29), 1 of which included both wearable alone and wearable combined with online intervention and 2 studies combined wearables with offline interventions (n=164). Wearable devices used were the Fitbit, Garmin, Polar, ActivPAL, ActivPAL, Jawbone, SitFIT, Yamax, LUMO Bodytech, Shimmer and Gruve (table 1) (online supplemental appendix 5). The average age of the subjects was 20–51 years old. The intervention duration ranged from 2 weeks to 1 year. Regarding the outcome measures, only three studies used subjective assessments, including the Marshall Questionnaire, the Chinese Taiwan Version of the International PA Questionnaire and the 7-day Sedentary and Light Intensity PA Log, while the rest were objective measures, mainly including the accelerometers of activPAL and the accelerometers of ActiGraph (table 1) (online supplemental appendix 5).

Table 1. Characteristics of included studies.

Author(year)	Country	Sample (I/C)	Age (mean±SD)	Duration	Intervention type	Devices	Measurement	Outcome
Clemes et al (2022)32	UK	378 (182/196)	48.4±9.4	6 mo.	W+online/control	Fitbit	Objective	ST
Nicolson et al (2021)33	Ireland	36 (17/19)	42.9±11.0	8 w.	W+online/control	Garmin	Objective	ST
Renaud et al (2020)36	Finland	193 (98/95)	42.3±10.2	4 mo.	W+offline/comparison	Activator	Objective	ST; PS
Arrogi et al (2019)27	Belgium	58 (31/27)	36.5±14.1	12 w.	‘W only‘/control	ActivPAL	Objective	ST; PS
Nishimura et al (2019)30	Japan	26 (13/13)	51±9	8 w.	W+online/control	ActivPAL	Objective	ST; PS
Kitagawa (2020)	Japan	48 (16/16/16)	38.0±4.5	2 w.	‘W only‘/ W+online/control	Jawbone	Objective	PS
Wyke et al (2019)31	UK	935 (464/471)	45.7±12.5	12 w.	W+online/control	SitFIT	Subjective	ST
Pope et al (2019)34	USA	38 (19/19)	21.5±2.3	12 w.	W+online/comparison	Polar M400	Objective	ST
Lin et al (2018)37	Taiwan	99 (51/48)	49.5	3 mo.	W+offline/comparison	Yamax	Subjective	ST
Brakenridge et al (2016)35	Australia	153 (66/87)	38.9±8.0	12 mo.	W+offline/comparison	LUMO Bodytech	Objective	ST; PS
Ellingson et al (2016)28	USA	28 (15/13)	20.1±1.5	5 w.	‘W only’/ comparison	Shimmer	Objective	ST; PS
Barwais et al (2013)29	Australia	33 (18/15)	27±4	4 w.	W+online/control	Gruve	Subjective	ST

C, comparison/control; I, intervention; mo., month; PS, prolonged sitting; STsedentary timew., weekW, wearable devices

Risk of bias

The assessment of risk of bias was consistent across ST and PS outcomes within the same study. Three studies were classified as having a low risk of bias, three studies were classified as having an unclear risk of bias and six had a high risk of bias rating. Detailed bias assessment can be found in online supplemental appendix 4.

Analysis of inconsistency

The network plot showed that a line between two points in the graph indicates evidence of a direct comparison between two interventions, and no line indicates evidence of indirect comparison, with the wider line indicating a greater number of studies (figures2 3).

Figure 2. Network of treatment comparisons on ST. ST, sedentary time.

Figure 3. Network of treatment comparisons on PS. PS, prolonged sitting.

Results of global inconsistency did not show any significant differences between the consistency and inconsistency models for ST (p=0.85) or PS (p=0.52) (online supplemental file 6) and the test for node-splitting analysis also revealed no significant differences between direct and indirect estimates for ST and PS, suggesting no evidence of inconsistency in the model (online supplemental appendix 6). Similarly, the loop inconsistency results for ST and for PS showed that the CI includes 0 (p=0.83 and p=0.27), indicating that the direct and indirect comparisons were very consistent and there was no inconsistency (online supplemental appendix 6). In addition, no significance was found in fit consistency model (online supplemental appendix 6).

Results of NMA

Results of NMA showed that the ‘wearable+online’ was significantly better in reducing ST compared with control group, comparison group and ‘wearable+only’, with moderate to large effect sizes (SMD=0.96, 95% CI 0.65 to 1.27; SMD=0.87, 95% CI 0.21 to 1.53; SMD=0.78, 95 % CI 0.14 to 1.42, respectively). In contrast, no significant differences were identified between the groups in reducing PS (table 2) (online supplemental appendix 7). Additionally, the SCURA values show the cumulative probability of each intervention being the best choice, ‘wearable+online’ intervention had the highest cumulative probability of being the best choice at 98.1%, followed by ‘wearable+offline’ (64.4%), ‘wearable only’ (40.5%), comparison group (25.9%) and control group (21.1%) for ST reduction. Similar rankings were observed for PS reduction, with probabilities of 69.9%, 61.1%, 59.7%, 37.1% and 22.1%, respectively (table 3).

Table 2. Results of NMA.

Comparison(A vs B)	Direct comparison trials(N)		NMASMD (95% CI)
	ST	PS	ST	PS
W+online versus W+offline	0	0	−0.57 (−1.31, 0.17)	−0.10 (−1.05, 0.85)
W+only versus W+offline	0	0	0.21 (−0.55, 0.97)	0.00 (−0.77, 0.77)
comparison versus W+offline	2	1	0.30 (−0.02, 0.63)	0.11 (−0.11, 0.32)
control versus W+offline	0	0	0.39 (−0.37, 1.14)	0.24 (−0.63, 1.12)
W+only versus W+online	0	1	0.78 (0.14, 1.42)	0.10 (−0.46, 0.65)
comparison versus W+online	1	1	0.87 (0.21, 1.53)	0.20 (−0.72, 1.13)
control versus W+online	5	1	0.96 (0.65, 1.27)	0.34 (−0.15, 0.83)
comparison versus W+only	1	1	0.09 (−0.59, 0.78)	0.11 (−0.64, 0.85)
control versus W+only	2	3	0.18 (−0.42, 0.78)	0.24 (−0.16, 0.65)
control versus comparison	0	0	0.09 (−0.60, 0.77)	0.14 (−0.71, 0.98)

SMDs refer to A–B for the comparison of A versus B.

Nnumber of studiesNMAnetwork meta-analysisPSprolonged sittingSMDstandardised mean differenceSTsedentary timeW12 wearable devices

Table 3. Results of the surface under the cumulative ranking and probability.

Treatment	SUCRA	PrBest	MeanRank
ST
wearable+online	98.1	93.2	1.1
wearable+offline	64.4	6.2	2.4
‘wearable only’	40.5	0.06	3.4
comparison	25.9	0.0	4.0
control	21.1	0.0	4.2
PS
wearable+online	69.9	45.0	2.2
wearable+offline	61.1	32.0	2.6
‘wearable only’	59.7	17.9	2.6
comparison	37.1	4.0	3.5
control	22.1	1.1	4.2

PS, prolonged sitting; ST, sedentary time; SUCRA, surface under cumulative ranking curve; PrBest, probability of being the best; MeanRank, the mean rank of interventions based on their effectiveness

Our sensitivity analysis results of ST of studies with intervention duration≥12 weeks, non-high risk of bias and objective measurement did not materially differ from our primary analysis (online supplemental appendix 8). However, because there are not enough studies in the subgroup of short-term (duration less than 12 weeks) intervention, high risk of bias and subjective measurement, we were unable to generate a network for comparison. Similarly, no sensitivity analyses were conducted for the PS outcome based on the length of intervention and risk of bias due to a paucity of studies. All studies measured PS objectively (online supplemental appendix 8).

Discussion

Comparisons with previous studies

This study suggests that wearable technology can be an effective strategy for reducing ST in healthy adults, contrasting with four meta-analyses from 2019,²³ 2020,²⁴ 2021²⁵ and 2022.²⁶ These analyses reported no significant reduction in ST with wearable device-based interventions in adults (SMD=−0.20, 95% CI −0.43 to 0.03; MD=−10.62, 95% CI −35.50 to 14.27; SMD=0.01, 95% CI −0.46 to 0.48; SMD=−0.12, 95% CI −0.25 to 0.01, respectively). The discrepancies may arise from differences in participant characteristics and intervention strategies. The earlier reviews included adults without distinguishing between various physical health conditions, potentially influencing the outcomes. Furthermore, these studies mainly aimed to increase PA, not specifically reduce ST; hence, the interventions were not directly targeting ST. In contrast, another review on self-monitoring interventions for ST reduction showed results consistent with our findings.⁵² This suggests that interventions specifically targeting ST may be more effective than those focused solely on increasing PA.

Best strategies among different types of interventions for ST

Brickwood et al demonstrated that wearable devices combined with additional content significantly improved PA, outperforming ‘wearable only’ interventions, which aligns with our findings that showed significant effects when online support was included. This enhancement may be due to the integration of various BCTs, such as self-monitoring, problem-solving, social environment modification and social support.¹³ Studies suggest that incorporating incentives like social and environmental support alongside self-monitoring can enhance intervention effects.⁵³ For instance, Oppezzo et al used the Tweet4Wellness platform with a ‘Fitbit+online’ intervention, providing a social network for users to share and receive content, thereby facilitating real-time interaction and support.⁵⁴ This approach leverages the practicality and affordability of online interventions, offering an advantage over traditional face-to-face methods. Similarly, Mitchell et al employed a ‘wearable+online’ strategy in rural Australian adults, incorporating a website, personalised step goals and telephone support, which effectively reduced ST and maintained improvements.⁵⁵ Thus, combining wearable devices with online interventions may be an effective approach to reducing ST.

Analysis of PS results

This study found that none of the three types of wearable device-based interventions significantly decreased PS, consistent with the findings of Compernolle et al.⁵² One reason may be the intervention strategies used, as most studies aimed to reduce ST through general knowledge education rather than specific strategies to reduce or interrupt PS. There is a need for developing wearable devices designed to reduce ST and assist in setting goals for increasing PA. Arrogi et al developed a device that targeted reducing ST by providing feedback and alerts after 25 and 30 min of inactivity, which led to significant improvements in ST and interruptions in PS.²⁷ Similarly, the Gruve Solutions monitor used in other studies vibrated to alert users when ST became prolonged. Devices specifically designed for ST monitoring are expected to be more effective in reducing total ST and improving other metrics, such as reducing and interrupting PS. For example, the Apple Watch’s hourly stand reminder offers immediate feedback, helping interrupt prolonged inactivity and significantly reducing PS. By offering real-time alerts and allowing users to set personalised goals, these devices can motivate users to increase PA and reduce ST. Integrating advanced sensors, artificial intelligence and machine learning could further enhance the personalisation of these interventions. Additionally, although there were differences in the SUCRA rankings of the various wearable device interventions, no significant differences were found in their effects, so interpretation should be cautious.

Strengths and limitations

The strengths of this study, which were reported strictly in accordance with NMA PRISMA guidelines, are as follows. (1) It is the first to use NMA to quantitatively compare the effectiveness of different wearable device interventions for reducing ST or PS in healthy adults. This design facilitates indirect comparisons among interventions and helps identify the most effective strategies. (2) All included studies are RCTs focused on reducing ST or PS, which enhances the reliability of the evidence. However, there are limitations to consider. (1) Some interventions in the literature target both ST and PA, making it difficult to fully isolate and quantify the effects specific to ST. (2) The included studies are primarily from developed countries, with limited representation from developing regions, where innovative methods are particularly needed to improve public health. (3) While this study provides valuable insights into the effectiveness of wearable devices in reducing ST and PS, it is crucial to consider the variation in the risk of bias across the included studies. Of the 12 studies assessed for risk of bias, 3 were classified as having a low risk of bias, 3 had an unclear risk and 6 were rated as having a high risk of bias. This variation may introduce uncertainty into the findings, particularly as studies with a high risk of bias could overestimate the effectiveness of the interventions, while those with an unclear risk may introduce ambiguity to the results. Due to the limited number of low-risk studies (N=3), we were unable to conduct a standalone NMA based solely on these studies. Therefore, we combined studies with low and unclear risk of bias to perform the analysis. While including studies with an unclear risk of bias may influence the results, this approach enables us to maximise the available data and offers a better alternative than not being able to conduct any analysis due to an insufficient number of low-risk studies. Additionally, studies with an unclear risk of bias can still provide valuable insights, as they may offer a broader perspective on the effects of wearable devices. Importantly, a sensitivity analysis that excluded studies with high risk of bias showed no material differences from our primary analysis, suggesting that the conclusions remain robust despite the variation in study quality.

Conclusions

Wearable devices have shown potential as effective tools for reducing ST in healthy adults. Our NMA indicates that the combination of wearable devices with online interventions is currently the best strategy for reducing ST. However, the effectiveness of other interventions requires further investigation. Evidence to date suggests that wearables do not significantly impact PS; however, this evidence is limited, and the SUCRA results may be misleading. Therefore, future research should focus on identifying optimal wearable device interventions tailored to different age groups and physical health conditions, and implementing high-quality RCT studies. Additionally, interventions should either be specifically designed to address ST or enhance the existing PA tracking devices to maximise benefits. Future studies should also consider other health-related indicators beyond total ST, including the reduction and interruption of PS.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.