Interventions outside the workplace for reducing sedentary behaviour in adults under 60 years of age

Summary of findings 1. Intervention compared to Control for reducing sedentary behaviour in adults under 60

Outcomes	*Anticipated absolute effects^ (95% CI)**		№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Intervention compared to control for reducing sedentary behaviour in adults under 60
Patient or population: community‐dwelling adults under 60 years of age Setting: outside the workplace Intervention: individual‐level interventions aiming to reduce sedentary behaviour Comparison: no intervention or attention control
Outcomes	Risk with Control	Risk with Intervention	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Device‐measured sedentary time	Short‐term follow‐up (up to 4 months)
	Control group mean was 574.44 min/d	MD 8.36 lower (‐27.12 lower to 10.40 higher	262 (4 RCTs)	⊕⊕⊕⊝ MODERATE ¹
	Medium‐term follow‐up (> 4 months to 12 months)
	Control group mean was 590.67 min/day	MD 51.37 lower (126.34 lower to 23.59 higher	188 (3 RCTs)	⊕⊝⊝⊝ VERY LOW ^{1 2 3}
Self‐report TV viewing time	Medium follow‐up (> 4 months to 12 months)
	Control group mean was 99.30 min/day	MD 12.45 lower (50.40 lower to 25.49 higher)	459 (2 RCTs)	⊕⊕⊝⊝ LOW ^{1 4}
	Long‐term follow‐up (> 12 months)
	Control group mean was 111.22	MD 0.30 higher (0.63 lower to 1.23 higher)	709 (2 RCTs)	⊕⊕⊝⊝ LOW ^{5 6}
Device‐measured MVPA	Short‐term (up to 4 months)
	Control group mean was 48.76 min/day	MD 1.99 higher (4.27 lower to 8.25 higher)	296 (4 RCTs)	⊕⊕⊝⊝ LOW ^{1 7}
	Medium‐term follow‐up (> 4 months to 12 months)
	Control group mean was 62.97 min/day	MD 6.59 higher (7.35 lower to 20.53 higher)	214 (3 RCTs)	⊕⊝⊝⊝ VERY LOW ^{1 2 8}
Self‐report light PA	Short‐term follow‐up (up to 4 months)
Self‐report light PA	Control group mean was 232.86 min/day	MD 156.32 higher (34.34 higher to 278.31 higher)	115 (2 RCTs)	⊕⊝⊝⊝ VERY LOW ^{5 9 10}
Adverse events and symptoms	None reported
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI:Confidence interval; MD: mean difference; RCT: randomised controlled trial; min/day: minutes per day
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹ Concerns about imprecision due to wide confidence intervals and small sample sizes ² Low risk of bias for outcome assessment, however 2 studies have high risk for several domains ³ Large variation in effect, I² = 87% ⁴ High risk of bias for outcome assessment for this outcome, unclear risk of bias for several other domains. ⁵ High risk of bias for outcomes assessment, unclear or high risk of bias for several other domains ⁶ Large sample size however large confidence intervals lead to uncertainty ⁷ Low risk of bias for outcome assessment, however majority of studies have unclear or high risk of bias for several domains ⁸ Large variation in effect, I² = 71% ⁹ Large variation in effect, CI's slightly overlap, Chi2 P < 0.05, I² = 79% ¹⁰ Very serious concerns about precision due to large confidence internals and small sample size

Background

Description of the condition

Research into sedentary behaviour is an emerging and rapidly growing field. Sedentary behaviour is defined as waking activity characterised by an energy expenditure of 1.5 or fewer metabolic equivalents and a sitting or reclining posture (Sedentary Behaviour Research Network 2012). A recent overview of systematic reviews of observational studies concluded that there is strong evidence of a positive relationship between sedentary behaviour and all‐cause mortality, fatal and non‐fatal cardiovascular disease, and type 2 diabetes and metabolic syndrome, along with moderate evidence of increased incidence of ovarian, colon, and endometrial cancers (De Rezende 2014). Conversely, interrupting sedentary time and/or replacing it with light‐intensity activity has been shown to improve several markers of cardiovascular disease risk (Dunstan 2012; Peddie 2013; Thorp 2014). Some research suggests that sedentary behaviour may be a distinct risk factor, independent of physical activity, for multiple adverse health outcomes (Chomistek 2013; Stamatakis 2011; Thorp 2011). Indeed, even people who are physically active at or above recommended levels experience the adverse effects of sedentary behaviour (Katzmarzyk 2009). Researchers estimate that people need approximately 60 to 75 minutes per day of moderate‐intensity physical activity to eliminate the increased risk of death associated with high sitting time; however, this high activity level reduces but does not eliminate the increased risk associated with high TV‐viewing time (Ekelund 2016).

The mechanisms through which sedentary behaviours lead to cardiovascular morbidity and mortality are under‐explored in the literature, but hypotheses point to defects in lipoprotein metabolism, early atherosclerosis, insulin resistance, and development of the metabolic syndrome (Same 2016). Obesity may act as a mediator between sedentary behaviours and negative health outcomes (Same 2016). Research from the genetics field has identified a genotype that is particularly susceptible to the adverse effects of excessive sedentary periods on glycaemic regulation (Alibegovic 2010), thus suggesting a potential gene–environment interplay that determines who is most susceptible to developing diabetes when exposed to excess sedentary time (Wilmot 2012).

Sedentary behaviour in adults is characterised as TV viewing and other screen‐focused behaviours in domestic environments, prolonged sitting in the workplace, and time spent sitting in automobiles (Owen 2011). Accelerometer data from a representative sample of US adults show that over 50% of waking hours are spent sedentary (Healy 2011). Weekday self‐reported sitting time varies considerably across European countries, with adults in northwestern European countries sitting the most (means 5.6 to 6.8 hours/d) (Bennie 2013). Accelerometer data suggest that UK men and women actually spend approximately 7.5 and 7 hours per day, respectively, being sedentary (Ekelund 2009). Many interventions to reduce sitting time in adults have focused on the workplace setting (Shrestha 2016); however, workplace sitting represents only one domain of sedentary behaviour, as adults spend approximately 70% of their non‐work time being sedentary as well (Parry 2013). TV viewing is a major contributor to sedentary behaviour in the USA, with the average adult watching five hours of TV per day (Pettee 2009; The Nielsen Company 2009). In addition, inactive travel modes and other non‐occupational behaviours such as leisure‐time computer use are increasing (Brownson 2005; Chau 2012). Serial cross‐sectional US data show that from 2001 to 2016, the estimated prevalence of computer use outside school or work for at least one hour per day increased from 29% to 50% for adults (Yang 2019). There are several known individual correlates of sedentary behaviour, such as age, physical activity level, body mass index, and socioeconomic status, and evidence related to social and environmental factors is emerging (O'Donoghue 2016). A taxonomy of sedentary behaviours is currently under development to provide a structure for current and future knowledge of sedentary behavior and a basis for distinguishing different behaviours (Chastin 2013).

Although no global (e.g. World Health Organization (WHO)) guidelines on sedentary behaviour exist, several countries have made population‐based recommendations. Much of the focus thus far is related to screen time for children. For example, since 2001, the American Academy of Pediatrics has recommended that parents limit children's total entertainment media time to no more than one to two hours of quality programming per day (American Academy of Pediatrics 2001). This two‐hour limit for children is consistent with the 2004 Australian guidelines (Australian Government 2004). Canada addressed general sedentary behaviour in its 2011 guidelines by recommending that children should minimise the time that they spend being sedentary each day (Tremblay 2011). More recently, the WHO included a screen time guideline for children younger than five years of age (WHO 2019). In 2011 the UK Chief Medical Officers joined Australia (among others) in providing public health guidelines aimed specifically at highlighting the potential health risks associated with sedentary behaviour for adults (BHFNC Physical Activity and Health 2012). The UK guidelines recommend that all adults minimise the amount of time spent being sedentary (sitting) for extended periods (Department of Health 2011), without specifying a duration of time. The Australian guidelines recommend that adults minimise the amount of time spent in prolonged sitting and break up long periods of sitting as often as possible (Australian Government 2014). A recent paper led by UK researchers suggested that for predominantly desk‐based occupations, workers should aim to initially progress towards accumulating two hours per day of standing and light activity during working hours, eventually progressing to a total accumulation of four hours per day (Buckley 2015); however, this is not an official guideline from the UK Chief Medical Officers.

Public health agencies have yet to present a quantified time limit on daily or weekly volumes of sedentary behaviour. Indeed several researchers suggest that the development of quantitative public health guidelines is premature, as little is known about the independent detrimental health effects of sitting, and there are many inconsistencies in how the evidence based was developed and interpreted (Stamatakis 2019). Some evidence suggests that a reduction of one to two hours of sedentary time per day could equate to substantial reductions in cardiovascular disease risk (Healy 2011). One study estimated that beneficial effects on cardiovascular disease risk biomarkers were associated with the reallocation of 30 minutes per day of sedentary time with an equal amount of sleep, light‐intensity physical activity, or moderate to vigorous physical activity (Buman 2013). A recent review of experimental studies concluded that breaking up sitting time and replacing it with light‐intensity ambulatory physical activity and standing may be a sufficient stimulus to induce acute favourable changes in postprandial (the period after eating a meal) metabolic parameters such as glucose and insulin response in people who are physically inactive and have type 2 diabetes, whereas a higher intensity or volume seems to be more effective in rendering such positive outcomes in young, regularly active people (Benatti 2015). However Stamatkais and colleagues noted the absence of long‐term prospective epidemiological evidence from studies that use objective measures of actual sitting, as opposed to absence of ambulatory movement (Stamatakis 2019). Of note is that the Second Edition of the Physical Activity Guidelines for Americans concludes that the existing evidence base does not allow a specific healthy target for total sedentary behaviour time to be determined (PA Guidelines Advisory Committee 2018). Similarly, the UK Expert Working Group for Sedentary Behaviour (tasked with examining evidence to decide if changes to current physical activity recommendations are warranted) did not support any significant changes to existing guidance nor suggest that a time limit or minimum threshold for sedentary behaviour should be added (Cooper 2018). The Expert Group supports a recommendation to interrupt prolonged periods of sedentary behaviour with light‐intensity physical activity but does not suggest that prolonged sedentary time should be interrupted by standing (Cooper 2018).

Description of the intervention

Our review assessed effects of interventions that aim to reduce sedentary behaviour among adults in non‐occupational settings. This will include studies that incorporate any component intending to reduce sedentary time, including if this is part of a larger intervention. We define a component as any strategy that explicitly targets a reduction in sedentary behaviour and is reported as a component of the intervention. This approach allows our review to include not only studies that focus exclusively on sedentary behaviour but also those that take a combined approach to reduce sedentary behaviour and increase physical activity. We note from the literature that some studies target a specific sedentary behaviour, such as TV viewing, or a collection of behaviours like overall use of screen time.

Interventions may be delivered at the individual, environmental, or policy level and include interventions within domestic environments, transport, and the wider community. Interventions include education and counselling sessions, wherein participants develop an implementation plan for behaviour change (De Greef 2010); self‐monitoring of behaviour alongside goal‐setting, where participants are encouraged to track their sitting time and set goals to increase breaks from sitting (Adams 2013); and multi‐component lifestyle interventions. Interventions targeting the environmental level may include point‐of‐decision prompts to encourage adults to stand (Lang 2015), or they could consist of controls placed on use of screen time, for example, limiting TV viewing by installing an electronic lockout system (Otten 2009). Those delivering the interventions will include counsellors, researchers, exercise physiologists, psychologists, general practitioners (GPs), and other public health professionals. Delivery modes are likely to involve face‐to‐face individual and/or group sessions, telephone support, provision of written leaflets, and use of online platforms. Many studies incorporate specific behaviour change strategies in the design, with self‐monitoring behaviour, problem‐solving, modifying social and physical environments, and giving information on the health impact of sitting most closely associated with promising interventions (Gardner 2016).

How the intervention might work

Several frameworks have emerged from recent research for understanding sedentary behaviour and informing intervention development (Owen 2014; Prapavessis 2015). An ecological model of sedentary behaviours highlights a behaviour‐ and context‐specific approach to understand the multiple determinants (Owen 2011). The behaviours and contexts of primary concern are TV viewing and other screen‐focused behaviours in domestic environments, prolonged sitting in the workplace, and time spent sitting in automobiles (Owen 2011). Trial authors suggest that change to sedentary behaviour in these domains may be altered by focusing on a specific setting with due consideration of the correlates of sedentary behaviour for that setting along with understanding factors related to high levels of overall sedentary time. A recent review of behaviour change strategies used in interventions for sedentary behaviour concluded that the most promising interventions were based on environmental restructuring, persuasion, or education (Gardner 2016). In addition, the following behaviour change techniques were particularly promising: self‐monitoring, problem‐solving, and restructuring of the social or physical environment.

We developed a logic model based on Baker 2015 to illustrate how the interventions might work and to describe the interactions between intervention activities and outcomes (Figure 1). We envisage several ways that interventions in non‐occupational settings may reduce sedentary behaviour in adults under 60 years of age.

Figure 1

Logic Model for interventions targeted outside of workplace settings for reducing sedentary behaviour (adapted from Baker 2015).

Individual, including education/information/counselling: adults may be willing to alter behaviour after learning about the health risks of a sedentary lifestyle. To support efforts to change behaviour, counsellors could encourage adults to track their sitting time and set goals to increase breaks. Similarly, they may receive suggestions to reduce sitting time.
Environmental: for example, removing seats from certain carriages on a train would force commuters to stand for the journey. Similarly, studies could limit recreational TV viewing by installing a lockout system that engages after a specific usage period per day, thus encouraging adults to change their usual behaviour. Placing computers at standing height would also prompt standing.
Policy, including challenges to socials norms: for example, by providing prompts and invitations to encourage standing at events, participants may be more likely to stand for some or all of the duration.

Why it is important to do this review

The evidence base reporting the health implications of sedentary behaviour and interventions to address this problem is rapidly expanding. Although studies first identified an increase in cardiovascular disease (CVD) risk experienced by people in highly sedentary jobs in the 1950s, only in recent years have the potential CVD risks from sedentary behaviour, as distinct from physical activity, come to be appreciated (Ford 2012). Recent observational and experimental evidence makes a compelling case for reducing and breaking up prolonged sitting time in both primary prevention and disease management contexts (Dempsey 2014). The scale of the problem is evidenced by the fact that the adverse health effects of sedentary behaviour are present even among those who are physically active at or above recommended levels (Katzmarzyk 2009). An estimated 5.9% of deaths may be attributable to daily total sitting time, suggesting that its reduction in the population could produce comparable benefits to those achieved by reducing smoking, inactivity, and overweight and obesity (Chau 2013). In this comparison, physical inactivity is defined as "doing no or very little physical activity at work, at home, for transport, or in discretionary time" (Bull 2004; WHO 2009). See Published notes.

Although several reviews have examined interventions to reduce sedentary time in children and young people, a paucity of systematic reviews in adults have been published. The reviews to date have often included interventions designed to increase physical activity but have also reported changes in sedentary time as unintended or secondary outcomes, rather than solely focusing on interventions that purposely aimed to reduce sedentary behaviour (Martin 2015; Prince 2014). A recent review found that the most promising interventions targeted sedentary behaviour instead of physical activity (Gardner 2016). The key difference between our review and these previous reviews is that we will examine only the effects of interventions to reduce sedentary behaviour on sedentary time and health outcomes in non‐occupational settings (Martin 2015; Prince 2014; Shrestha 2016). A recent Cochrane Review examined interventions to reduce sitting time in the workplace setting (Shrestha 2016), another Cochrane Review is examining interventions for reducing sedentary behaviour in community‐dwelling older adults (Chastin 2017), and two further Cochrane Reviews have examined workplace interventions for increasing standing or walking for preventing and decreasing musculoskeletal symptoms among sedentary workers (Parry 2017a; Parry 2017b). However, to our knowledge, there is only one published synthesis of evidence in non‐occupational settings (Thraen‐Borowski 2017), and a meta‐analysis was not conducted in that review. As adults spend approximately 60% to 70% of their non‐work time being sedentary (Clemes 2014; Parry 2013), there is great scope for intervention, and a synthesis of evidence on existing interventions will help guide this task. We believe that non‐occupational settings may offer greater scope for changing sedentary behaviour than occupational settings, where individuals may have less control over their working environments and practices.

The need that policymakers and practitioners have for this Cochrane Review is evident in the focus on sedentary behaviour at the governmental level worldwide. This is also reflected in much being written in the media about the dangers of sitting. Countries are expanding their public health guidelines to include recommendations on limiting sedentary time (e.g. see Healthy Ireland 2016 and Sedentary Behaviour and Obesity Working Group 2010). This review will also provide key evidence for countries that seek to update existing sedentary behaviour guidelines in future years (e.g. Australian Government 2014). The findings of the review will therefore aid evidence‐based decision‐making by policymakers and practitioners working to address sedentary behaviour worldwide. This rapidly growing field will inform the development of public health policy over the coming decade, and a regularly updated, robust, comprehensive review of the evidence is required to support this task.

Objectives

Primary

To assess effects on sedentary time of non‐occupational interventions for reducing sedentary behaviour in adults under 60 years of age

Secondary

To describe other health effects and adverse events or unintended consequences of these interventions
To determine whether specific components of interventions are associated with changes in sedentary behaviour
To identify if there are any differential effects of interventions based on health inequalities (e.g. age, sex, income, employment)

Methods

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs) and cluster randomised controlled trials (cluster RCTs) aimed at changing sedentary behaviour. Given the growing volume of research on interventions targeting sedentary behaviour, particularly RCTs, we believe that solely including RCTs and cluster RCTs will allow us to draw conclusions from the best available evidence.

Types of participants

We included studies involving community‐dwelling adults aged 18 to 59 years who are free from pre‐existing medical conditions that may limit participation in the intervention.

Types of interventions

We included interventions targeted outside of workplace settings. Hypothetically, these may include interventions within domestic environments, transport, and the wider community. The following are examples of interventions that may be included in the review.

Counselling/education to reduce and self‐monitor sedentary behaviour.
Limits/controls placed on screen time.
Environmental change interventions, for example, point‐of‐decision prompts to encourage standing.
Multi‐component lifestyle interventions that include a sedentary behaviour element.
Community‐level interventions that specifically aim to address sedentary behaviour.

Interventions may be delivered at the individual, environmental, or policy level. We excluded interventions provided in workplace settings, as they fall under the scope of a separate Cochrane Review (Shrestha 2016). In addition, we excluded studies with participants 60 years of age and older, as another Cochrane Review is focusing on that age group (Chastin 2017). We also excluded studies that aim to improve physical activity levels but happen to report sedentary time, as they do not specifically target sedentary behaviour in their design.

Comparison was between those receiving the intervention and those receiving no intervention or attention controls.

Types of outcome measures

We included studies that report sedentary behaviour as a primary or secondary outcome.

Primary outcomes

The primary outcome is sedentary behaviour, assessed at baseline and post intervention. There is no international consensus on a gold standard measure of sedentary behaviour. With this in mind, we included studies that utilised device‐based (e.g. accelerometer, inclinometer) or self‐report (e.g. diary, questionnaire) measures of sedentary time. We included studies that report sedentary behaviour in one domain only, for example, sitting during transport or TV viewing at home, as well as those reporting total daily sedentary behaviour. We considered both total duration of sedentary behaviour reported and breaks in sedentary behaviour as primary outcome measures.

We included the following primary outcomes.

Device‐measured sedentary time.
Self‐report sitting time.
Self‐report TV viewing.
Breaks in sedentary time.

Secondary outcomes

We included the following secondary outcome measures.

Energy expenditure.
Body composition (e.g. body mass index, waist and hip circumference, body fat percentage, body weight).
Cholesterol (e.g. total cholesterol, low‐density lipoprotein (LDL) cholesterol, high‐density lipoprotein (HDL) cholesterol).
Markers of insulin resistance (e.g. fasting blood glucose, liver transaminases, insulin levels or insulin resistance/impaired insulin sensitivity).
Inflammatory markers (e.g. C‐reactive protein (CRP), interleukin (IL)‐6, tumour necrosis factor (TNF)‐α).
Measures of carotid intima media thickness (e.g. ultrasound).
Measures of endothelial function (e.g. peripheral arterial tonometry).
Measures of mental health (e.g. stress symptoms, anxiety, depression, self‐image).
- Mood
- Wellness
Adverse events and symptoms (e.g. musculoskeletal injuries/pain, cardiovascular events).
Unintended outcomes (e.g. social approval/disapproval by others, change in overall physical activity behaviour).
- Device‐measured moderate to vigorous physical activity (MVPA)
- Self‐report MVPA
- Self‐report light, moderate, vigorous, and total physical activity (PA)
- Step count

Search methods for identification of studies

Electronic searches

We searched the following electronic databases up to 14 April 2020, using a search strategy developed by NR and EM in liaison with the Cochrane Public Health Group (CPHG) Trials Search Co‐ordinator (see Appendix 1).

CPHG Specialised Register.
Cochrane Central Register of Controlled Trials (CENTRAL), in the Cochrane Library, Wiley.
MEDLINE (Ovid MEDLINE Epub Ahead of Print, In‐Process & Other Non‐Indexed Citations, Ovid MEDLINE Daily and Ovid MEDLINE) (OvidSP) (1946 to present).
Embase (OvidSP) (1974 to present).
Cochrane Database of Systematic Reviews, in the Cochrane Library, Wiley.
Cumulative Index to Nursing and Allied Health Literature (CINAHL) (EBSCOHost) (1982 to present).
PsycINFO (OvidSP) (1806 to present).
SportDiscus (EBSCOHost).

We did not impose any language, publication status, or date restrictions. We contacted trial authors and research groups for information about unpublished or ongoing studies.

Searching other resources

We handsearched reference lists of included studies and key systematic reviews. We searched the Clinicaltrials.gov trial register (http://clinicaltrials.gov) on 14 April 2020. We contacted authors of included studies and relevant systematic reviews to identify additional studies. In addition, we contacted experts in the field and asked them to identify further articles. We also searched the websites of organisations involved in addressing and reporting research on sedentary behaviour.

Sedentary Behaviour Research Network (http://www.sedentarybehaviour.org).
World Health Organization (http://www.who.int).
US Centers for Disease Control and Prevention (http://www.cdc.gov).
International Society for Physical Activity and Health (http://www.ispah.org).
National Physical Activity Plan (http://www.physicalactivityplan.org).
The Community Guide (http://www.thecommunityguide.org).
European Commission, Public Health (https://ec.europa.eu/health/home_en).

NR and EM carried out the searches.

Data collection and analysis

Selection of studies

We downloaded the references retrieved through electronic searches and handsearching to the reference management software Endnote, removing duplicates (Endnote 2015). Two review authors (EM and MM) independently undertook an initial screening of titles and abstracts to exclude records outside the scope of the review. A third review author (CF) reviewed any items of disagreement to reach a consensus. We obtained full‐text papers when we deemed titles to be relevant, or when eligibility was unclear. Inclusion decisions were based on the full texts of potentially eligible studies. Two review authors, working independently, determined whether each study met the eligibility criteria (EM and MM). When any disagreements occurred, a third review author (CF) examined the paper and the three review authors reached a consensus. We kept a record of reasons for excluding studies. If we identified papers detailing study design, study protocols, or process evaluations, we contacted the study authors to locate published or unpublished findings from the study. We collated multiple reports of the same study and treated each study as the unit of interest. We did not find any potentially relevant title of a paper in a language other than English, so we did not require translation services.

We used the online software Covidence to manage the study selection process (Covidence 2016).

Data extraction and management

Two review authors (EM and KM) independently extracted study characteristics and outcome data using a modified version of the Public Health Group Data Extraction and Assessment Form. We consulted a third review author (CF) when disagreements occurred and we reached consensus among the three authors. All participating review authors piloted the Data Extraction and Assessment Form, modifying it where necessary to ensure comprehensiveness and comparability between results. We completed data extraction online using Covidence software and exported data directly to Review Manager 5 (Covidence 2016; RevMan 2014). When information was missing or when we needed clarification, we contacted the authors of included studies. We report relevant information in the Characteristics of included studies table. When we found multiple articles from the same study, we compared them for completeness and possible contradictions.

We extracted the following data.

Study objectives: for example, to decrease sedentary time or to decrease sedentary time and increase physical activity.
Study design: RCTs and cluster RCTs.
Methods: study location, study setting, dates of study, duration of intervention, and duration of follow‐up. We recorded how investigators measured sedentary behaviour, for example, by questionnaire/accelerometer.
Participants: number randomised to each group, age, withdrawals. We extracted sociodemographic characteristics at baseline and at endpoint using the PROGRESS framework (Place, Race, Occupation, Gender, Religion, Education, Socioeconomic status, Social status).
Intervention: content of intervention, description of comparison. We noted whether or not interventions included particular strategies to address diversity or disadvantage. We also noted the theoretical basis for the intervention when reported.
Outcomes: outcome measures post intervention and at follow‐up when available. We noted whether clustering was taken into account in cluster RCTs. When data on multiple measures of the same or similar outcomes were available, for example, body composition measures of body mass index (BMI) and body fat percentage, we recorded both.
Notes: funding received and conflicts of interest as declared by study authors.

In addition to study characteristics and outcomes data, we collected from included studies any available information about context, implementation factors, equity, cost, and sustainability and reported it in the Characteristics of included studies table (CPHG 2011). We view sustainability of the interventions as a combination of intervention components (dose) and magnitude of effect over time. We collected any available data related to sustainability (e.g. follow‐up measures) and assessed the data using an adapted version of the approach adopted by Müller‐Riemenschneider 2008. We included potential moderators and confounders of study outcomes, such as age, race, and gender, on the Data Extraction and Assessment Form.

For several studies, it was necessary to process data in preparation for analysis. For example, in four studies, sedentary time was converted from hours per day to minutes per day (Barwais 2013; Biddle 2015; French 2011; Laska 2016). To calculate the mean device‐measured MVPA in Ellingson 2016, we summed the mean values reported for moderate PA and vigorous PA. For Arrogi 2017, we calculated the mean value for device‐measured sedentary time by obtaining the average of weekday and weekend day results reported in the paper. We utilised the same methods to calculate mean device‐measured MVPA in Jago 2013.

In Finni 2011, three of the time points for data collection (6, 9, and 12 months) would all be considered as medium‐term follow‐up in the present review (i.e. > 4 months to 12 months). We used the data collected at 12 months as the medium‐term value for subgroup analysis. Similarly, Williams 2019 reported data at 17 weeks and 6 months; we used the data collected at 6 months in our analysis.

Assessment of risk of bias in included studies

Two review authors (EM and KM) independently assessed risk of bias using the Cochrane 'Risk of bias' tool (Higgins 2011a). When disagreements occurred, a third review author (MM) reviewed the studies, and review authors together reached consensus by discussion. This tool assesses:

selection bias (sequence generation and allocation concealment);
performance bias (blinding of participants and personnel);
detection bias (blinding of outcome assessment);
attrition bias (incomplete outcome data); and
reporting bias (selective reporting).

We graded each domain as being at 'low', 'high', or 'unclear' risk of bias.

We considered blinding separately for different key outcomes when necessary, for example, the risk of bias for sitting measured by means of an inclinometer may be very different from that for a self‐reported reduction in sitting time (Shrestha 2016). We did not consider blinding of participants and personnel for risk of bias assessment, as it is not possible to blind these individuals in studies examining attempts to modify activity behaviour (Shrestha 2014). We considered the following additional criteria for cluster RCTs, as recommended in Section 16.3.2 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011b): recruitment bias; baseline imbalance; loss of clusters; incorrect analysis; and comparability with individually randomised trials.

We summarised risk of bias at the outcome level and judged each outcome as being at 'low', 'medium', or 'high' overall risk, given the study design and the potential impact of identified risks noted in the table for each study that contributed results for that outcome (CPHG 2011).

Measures of treatment effect

For studies with continuous outcome measures, we reported mean scores and standard deviations. We used the mean difference between post‐intervention values of intervention and control groups to analyse the size of the effects of interventions. For cluster RCTs we used the adjusted MD between groups.

Unit of analysis issues

We identified one study with multiple intervention groups (Kitagawa 2020). We pooled the intervention arms into one group to create a single pair‐wise comparison, as recommended in Section 16.5.4 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011b). This method avoids including a group of participants twice in the same meta‐analysis. All of the included cluster RCTs made allowance for the design effect of clustering; therefore it was not necessary to re‐analyse data.

Dealing with missing data

We contacted study authors via email when data were missing or unclear (to request e.g. missing information on methods, missing participants due to dropout, and missing statistics). We retrieved email addresses from author information provided in the study publication and, when necessary, accessed contact directories from the authors' documented affiliated organisations. We noted missing data on the data extraction form and reported this in the 'Risk of bias' table. If numerical outcome data such as standard deviations or correlation coefficients were missing, and we could not obtain them from the study authors, we calculated these values from other available statistics such as P values, according to the methods described in Chapter 16 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011b; Shrestha 2014). We used the RevMan calculator to determine standard deviation (SD) from standard error (SE) for several studies (Biddle 2015; Laska 2016; Williams 2019). Similarly, for French 2011, we used the SE of the mean difference to calculate the SD in intervention and control groups post intervention. For outcomes that are reported narratively, we used the RevMan calculator to determine MD for several studies (Barwais 2013; Cotten 2016; Ellingson 2016; French 2011; Sui 2018).

Assessment of heterogeneity

We considered methodological heterogeneity by assessing differences between included studies in terms of study design. We considered clinical heterogeneity by assessing variability among participants, interventions, and outcomes, as recorded in the Characteristics of included studies table. We visually inspected forest plots to assess statistical heterogeneity and used the I² statistic to quantify the level of heterogeneity present (P < 0.10). This describes the percentage of variability in effect estimates due to heterogeneity rather than to sampling error (chance) (Deeks 2011). We planned to perform sensitivity analyses to investigate heterogeneous results; however due to the number of included studies, we did not conduct these analyses.

Assessment of reporting biases

As we included fewer than 10 studies per outcome, we could not use funnel plots to assess reporting bias, as the power of these tests would be too low to distinguish chance from real asymmetry (Sterne 2011).

Data synthesis

Given that participants, interventions, and comparisons were sufficiently similar, we conducted a meta‐analysis using RevMan 5. We used the random‐effects model, as it allows for a greater level of natural heterogeneity between studies. The appropriate method of meta‐analysis depends on the nature of the data, and we followed the guidelines presented in Chapter 9 ("Analysing data and undertaking meta‐analyses") of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2011). We included data from cluster randomised trials in meta‐analyses, as trial authors had taken clustering into account. We were unable to examine the effects of interventions according to types of intervention, as all were provided at the individual level.

When it was not possible to conduct a meta‐analysis, we reported results narratively. We grouped the data by outcome, as this makes the most sense for the reader. We presented data in tables and summarised results narratively.

We created Summary of findings table 1 for the main comparisons. Summary of findings table 1 includes the numbers of participants and studies for the primary outcomes (device‐based and self‐report measures of sedentary behaviour), summarises the intervention effects, and includes a measure of the certainty of evidence (see Quality of the evidence section below). We also reported the following secondary outcomes as they were deemed most relevant: device‐measured MVPA, self‐report light PA, steps.

We identified the theoretical frameworks and models identified in the primary studies. We considered costs and sustainability of studies in preparing the synthesis.

Subgroup analysis and investigation of heterogeneity

Given the small number of common outcomes across studies, we were unable to conduct the planned range of subgroup analyses for our primary outcome to see if there was any evidence of differential responses to the intervention. We included subgroup analysis for length of follow‐up. When studies reported several follow‐up points, we included data from each relevant time point. In addition, we assessed subgroup heterogeneity by examining forest plots and quantification by using the I² statistic.

Had sufficient data been available, we planned to carry out the following subgroup analyses for our primary outcome to see if there was any evidence of differential responses to interventions.

Intervention type: e.g. personal monitoring device, information/education, counselling, text messages or combinations of these categories.
Gender: given the unique sedentary behaviour profiles of men and women (Bennie 2013; Matthews 2008), and the fact that interventions to reduce sedentary behaviour seem to have limited effects when targeting women only (Martin 2015), outcomes by gender should be examined (men, women, men and women).
Socioeconomic group (education or income): because variations in response to public health interventions according to socioeconomic status are frequent (White 2009), outcomes by socioeconomic group should be compared. It has been noted that high levels of education are associated with higher levels of sitting (Bennie 2013).
Age: subgroup analysis to consider the influence of the age of participants.
Intensity of the intervention: using an adapted version of the approach used by Baker 2015.
Category of study setting: as interventions may be setting‐specific, the influence of study setting should be considred (e.g. schools/universities, transport, home).
BMIor another measure of overweight/obesity: to consider the influence of body composition given the evidence that associations between prolonged sitting and risk of CVD are stronger in overweight than in normal weight adults (Chomistek 2013).
Study aim: as previous reviews have demonstrated differential effects between interventions that solely aim to reduce sedentary behaviour or that take a combined approach to reducing sedentary behaviour and increasing physical activity (Gardner 2016; Martin 2015), subgroup analysis to compare outcomes by study intention are warranted.
Baseline sedentary status: as daily sedentary time for adults varies across studies (Bennie 2013), whether baseline sedentary level has an influence on outcomes should be investigated.
Baseline physical activity: influence of baseline physical activity level should be considered.

Sensitivity analysis

Given the small number of studies included in the review, we did not undertake sensitivity analysis. Had sufficient data been available, we planned to use sensitivity analysis for primary outcomes to explore the impact of risk of bias on study findings, while excluding studies at high or unclear risk of bias.

Summary of findings and assessment of the certainty of the evidence

We used the GRADE system to assess the certainty of the body of evidence for each outcome, and to draw conclusions about this within the text of the review. The certainty of a body of evidence as assessed by GRADE is understood as the extent to which one can be confident in the estimate of effect (Guyatt 2008). We summarised the assessment in a 'Summary of findings' table created with GRADEpro software (GRADEpro GDT). Two review authors independently assessed outcomes across studies (EM and KM). We resolved disagreements by consensus.

We rated evidence as very low, low, moderate, or high certainty by considering the GRADE domains. Table 1 presents definitions for these ratings (Balshem 2011). The GRADE approach to rating the quality of evidence begins with the study design (randomised trials start as high quality) and then addresses five reasons to possibly downgrade the quality of evidence (Balshem 2011). The five factors that may lead to downgrading the certainty of evidence are:

See Characteristics of included studies, Characteristics of excluded studies, Characteristics of ongoing studies.

Table 1. Definitions for quality ratings in GRADE

Quality level	Definition
High	Further research is very unlikely to change our confidence in the estimate of effect
Moderate	Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate
Low	Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate
Very low	Any estimate of effect is very uncertain

study limitations ‐ risk of bias;
publication bias ‐ available evidence derived from a number of small studies;
imprecision ‐ random error;
inconsistency ‐ inconsistency in the magnitude of effect in studies of alternative management strategies (Guyatt 2011a); and
indirectness ‐ indirect participants, interventions, outcomes, or comparisons.

If one of these factors is found to exist, it is classified either as serious (rating down by one level) or as very serious (rating down by two levels). We followed guidance from Ryan 2016 when incorporating the GRADE ratings into the results of the review, so that the certainty of evidence is clear.

Results

Description of studies

Results of the search

Our searches to 14 April 2020 yielded 21,100 hits from electronic databases and 79 from other studies. This resulted in 10,976 hits following removal of duplicates (Figure 2). The full search strategies and the number of hits for the eight electronic databases and the clinical trials registry can be found in the Appendices. After reading titles and abstracts, we excluded 10,799 records and selected 177 reports for full‐text review. Of these, we excluded 144 reports. We collated multiple reports of the same study, with the paper that reported the outcomes of particular note to this review chosen as the main source of study results. We identified 24 reports, representing 13 unique studies, for inclusion in this review (Arrogi 2017; Barwais 2013; Biddle 2015; Cotten 2016; Ellingson 2016; Finni 2011; French 2011; Jago 2013; Kitagawa 2020 Laska 2016; Lioret 2012; Sui 2018; Williams 2019). In addition, we identified eight studies that are classified as ongoing (NCT02909725; ISRCTN58484767; NCT03698903; Latomme; Martin Borras 2014; NCT04257539; Pinto 2017; Schroe 2019). For one study, we could not locate a full‐text version (Marcinkevage 2012), and we categorised this as a study awaiting classification (see Characteristics of studies awaiting classification).

Figure 2

Study flow diagram.

We sent requests to the authors of several studies to obtain additional data or to clarify data (Biddle 2015; Finni 2011; French 2011; Marcinkevage 2012; Martin Borras 2014). We received unpublished data from the authors of two studies (Biddle 2015; Finni 2011).

Included studies

Design

Ten studies are RCTs (Arrogi 2017; Barwais 2013; Biddle 2015; Cotten 2016; Ellingson 2016; Jago 2013; Kitagawa 2020; Laska 2016; Sui 2018; Williams 2019), and three are cluster RCTs (Finni 2011; French 2011; Lioret 2012). See the Characteristics of included studies table for further information. All of the cluster RCTs reported adjusted results; therefore re‐analysis of data was not required. One study included multiple intervention groups (Kitagawa 2020).

Setting

Settings for all studies were outside the workplace and included home/community (Arrogi 2017; Finni 2011; Jago 2013; Kitagawa 2020; Lioret 2012), online (Barwais 2013), primary care (Biddle 2015; Williams 2019), university (Cotten 2016; Ellingson 2016; French 2011; Sui 2018), and community college (Laska 2016).

Three studies were undertaken in the United States (Ellingson 2016; French 2011; Laska 2016), three in the United Kingdom (Biddle 2015; Jago 2013; Williams 2019), two in Australia (Barwais 2013; Lioret 2012), and two in Canada (Cotten 2016; Sui 2018). Finally, one study each was undertaken in Belgium, Finland and Japan (Arrogi 2017; Finni 2011, Kitagawa 2020 respectively),

Participants

The included studies involved 1770 participants. Sample sizes ranged from 30 participants in Ellingson 2016 to 542 participants in Lioret 2012. In nine studies, a majority of participants were female (Arrogi 2017; Biddle 2015; Cotten 2016; Finni 2011; French 2011; Jago 2013; Laska 2016; Lioret 2012; Sui 2018). One study reported that there was not a considerable difference in the proportions of males and females (Ellingson 2016). Another two studies included only females (Kitagawa 2020; Lioret 2012). The mean age of study participants ranged from 20 years in Ellingson 2016 to 41 years in French 2011. Five of the studies involved participants in their 20's (Barwais 2013; Cotten 2016; Ellingson 2016; Laska 2016; Sui 2018). Several studies targeted the family (Finni 2011; Jago 2013; Lioret 2012).

Few studies provided sociodemographic characteristics, apart from gender, using the PROGRESS framework (Place, Race, Occupation, Gender, Religion, Education, Socioeconomic status, Social status). Six studies noted the occupation of participants (97% officer workers in Arrogi 2017; 63% officer workers in Barwais 2013; 100% housewives in Kitagawa 2020; 100% university students in Cotten 2016, Ellingson 2016, and Sui 2018). Six studies reported race/ethnicity (Biddle 2015; Ellingson 2016; French 2011; Jago 2013; Laska 2016; Williams 2019). Biddle 2015 reported that participants were recruited from areas with a diverse ethnic and socioeconomic makeup. Two studies reported income data of participants (French 2011; Laska 2016). Education level was reported in one study (Lioret 2012).One study purposively recruited adults with a diagnosis of a serious mental illness (Williams 2019)

Interventions

Nine studies aimed to reduce sedentary behaviour (Arrogi 2017; Biddle 2015; Cotten 2016; Ellingson 2016; Finni 2011; French 2011; Jago 2013; Kitagawa 2020; Sui 2018), and four sought to both reduce sedentary behaviour and increase physical activity levels (Barwais 2013; Laska 2016; Lioret 2012; Williams 2019). All interventions were delivered at the individual level (i.e. none were considered environmental or policy activities) (see Figure 1). Table 2 provides a summary of the interventions. Three studies used a personal monitoring device (Arrogi 2017; Barwais 2013; Kitagawa 2020). Several studies included some form of information or education (French 2011; Laska 2016; Lioret 2012; Jago 2013), or some type of counselling (Finni 2011; Sui 2018). Two studies included both information/education and a personal monitoring device (Biddle 2015; Ellingson 2016). One study incorporated information/education, a personal monitoring device and counselling (Williams 2019). One study sent daily text messages to participants (Cotten 2016).

See: Summary of findings 1 Intervention compared to Control for reducing sedentary behaviour in adults under 60

Table 2. Summary of the interventions

Study	Target group	SB/PA‐related aim	Intervention components	Last follow‐up	Theory
Personal monitoring device
Arrogi 2017	Adults aged 18 to 55 years with sedentary jobs and/or predominantly sedentary leisure time	SB	Motion sensor and smartphone app to provide feedback in relation to targets, including a warning signal to modify behaviour	2 weeks	BCT principles
Barwais 2013	People who reported high total sitting time	SB + PA	Online personal activity monitor to provide feedback on progress towards their daily goal	4 weeks	NR
Kitagawa 2020	Housewives	SB	Motion sensor and smartphone application to provide feedback on sitting time. The tailored feedback group also received individual suggestions to shorten sitting time.	2 weeks	NR
Information/Education
French 2011	At least one child aged at least 5 years and two household members aged at least 12 years, with weekly average household TV viewing of at least 10 hours per person	SB	6 monthly face‐to‐face group sessions, monthly newsletters, 12 home‐based activities, a TV‐limiting device, and monthly telephone support calls	12 months	NR
Laska 2016	Adults aged 18 to 35 years, with BMI of 20 to 34.9 kg/m²	SB + PA	1 credit course at university plus social networking and a support website	24 months	Ecological theories of health behaviour, social‐cognitive theory, and social network theory
Lioret 2012	First‐time mothers	SB + PA	6 quarterly 2‐hour sessions at parent group and a newsletter	18 months	Theory of anticipatory guidance
Jago 2013	Parents with at least one child aged 6 to 8 years	SB	8 weekly 2‐hour education sessions	16 weeks	Self‐determination theory
Information/Education plus personal monitoring device
Biddle 2015	Overweight or obese adults aged 18 to 40 years with one or more additional risk factors for diabetes	SB	1 × 3‐hour group workshop and a self‐monitoring device to view and track progress against personal goals and provide feedback via vibration notifications	12 months	Bandura’s social‐cognitive theory, Gollwitzer’s implementation intentions concept, Behavioural Choice Theory, and Leventhal’s Common Sense Model
Ellingson 2016	Full‐time students aged 18 to 26 years, reporting more than 3 hours of daily leisure‐time sedentary behaviour	SB	10 to 15‐minute information session and personal monitor with vibration feedback to modify behaviour towards a set target	5 weeks	Habit theory
Counselling
Finni 2011	Healthy men and women with children aged 3 to 8 years and an occupation where they sit for more than 50% of their work time	SB	Tailored counselling, 2 telephone calls at 1 and 5 months, and motivational emails	12 months	Theory of planned behaviour
Sui 2018	Full‐time university students aged 18+ years with good physical and mental health	SB	Counselling sessions with an action plan	6 weeks	Health action process approach
Information/Education plus personal monitoring device plus counselling
Williams 2019	Community dwelling adults with serious mental illness	SB + PA	Initial education session, fortnightly one‐to‐one health coaching, provision of pedometers and access to a weekly walking group	17 weeks	COM‐B model of behaviour change
Daily text messages
Cotten 2016	Adults aged 18 to 65 years	SB	Daily text messages	6 weeks	NR

BCT: behaviour change technique.

BMI: body mass index.

NR: not reported.

PA: physical activity.

SB: sedentary behaviour.

In terms of sustainability of the interventions, five included a follow‐up measure of at least 12 months (Biddle 2015; Finni 2011; French 2011; Laska 2016; Lioret 2012).

Control group

The comparison group in five studies was a no intervention control (Arrogi 2017; Barwais 2013; Finni 2011; French 2011; Jago 2013), with one of these specifically instructing participants to follow their normal daily lifestyle patterns (Barwais 2013). Other studies provided information leaflets to the control group regarding sedentary behaviour (Biddle 2015; Kitagawa 2020), or consisting of basic health promotion information (Laska 2016; Lioret 2012; Williams 2019). Three studies used attention control. In one study, participants received daily text messages about random health facts (Cotten 2016), and in another study, participants were given an accelerometer to wear without feedback (Ellingson 2016). Finally in one study, the control group was given strategies to achieve Canada's Food Guide weekly food group servings (Sui 2018).

Outcomes

Primary outcomes

We found 13 studies that reported using two principal forms of continuous outcomes ‐ device‐derived and self‐reported. These were categorised into three groups: device‐measured sedentary time, self‐reported TV viewing, and self‐reported sitting time.

Self‐report measures of sedentary behaviour were utilised in nine studies, device‐based measures were used in four studies, and a further two studies used both. Seven studies used a questionnaire (Biddle 2015; Ellingson 2016; French 2011; Jago 2013; Laska 2016; Lioret 2012; Sui 2018). Two studies used the 7‐Day Sedentary and Light Intensity Physical Activity Log (7‐Day SLIPA Log) (Barwais 2013; Cotten 2016); however in the study by Cotten and colleagues, participants were asked to fill out the items based on a typical weekday and a typical weekend day, rather than on a daily basis. Six studies used a device‐based measure of sedentary time (Arrogi 2017; Biddle 2015; Ellingson 2016; Finni 2011; Kitagawa 2020; Williams 2019). Three of these studies used a thigh‐worn device that measured posture (Arrogi 2017; Biddle 2015; Ellingson 2016) and two studies used a waist‐worn accelerometer (Biddle 2015; Finni 2011), with both using < 100 counts per minute (cpm) as the definition of sedentary. Two further studies used a wrist‐worn accelerometer (Kitagawa 2020; Williams 2019). Note that two of the aforementioned studies included both a device‐based and a self‐report measure of sedentary behaviour (Biddle 2015; Ellingson 2016).

Six studies reported total sedentary time (Arrogi 2017; Barwais 2013; Biddle 2015; Ellingson 2016; Finni 2011; Williams 2019). Several other studies reported sedentary behaviour in one domain only: TV viewing (French 2011; Jago 2013; Laska 2016), or TV, video, and DVD viewing (Lioret 2012). Biddle 2015 reported sedentary behaviour separately for multiple domains: daily sitting time, sitting while travelling, sitting at work, sitting while watching TV, sitting while using a computer at home, and sitting in leisure time. One study reported longest prolonged sitting time (Kitagawa 2020). Five studies presented findings in relation to breaks in sedentary time. Two of these studies reported frequency of breaks in sedentary time (i.e. breaks every X minutes) (Cotten 2016; Sui 2018), and three studies reported number of breaks per hour or day in sedentary time (Arrogi 2017; Biddle 2015; Finni 2011). Arrogi 2017 defined breaks as the number of sit‐to‐stand transitions. Biddle 2015 reported bouts of light to vigorous physical activity as breaks in sedentary time. Finni 2011 defined a break as "an interruption in sedentary time when accelerometer counts rose up to or above 100 counts/min for a minimum of one minute".

Secondary outcomes

We found one study that reported energy expenditure; this was reported specifically in relation to leisure‐time physical activity and was assessed using the Paffenbarger Questionnaire (Laska 2016).

Several studies reported measures of body composition. Two of these reported body fat data; one used total fat mass percentage measured using dual‐energy X‐ray absorptiometry (Finni 2011), and the other reported body fat percentage measured used bioelectrical impedance analysis (Biddle 2015). Three studies reported BMI (Biddle 2015; Finni 2011; French 2011), two studies provided data on weight (Biddle 2015; Finni 2011) and two studies reported waist circumference (Biddle 2015; Williams 2019).

Only one study reported cholesterol as an outcome measure (Biddle 2015). Two studies provided data on glucose control and insulin sensitivity (Biddle 2015; Finni 2011). Both reported fasting glucose and fasting insulin. In addition, Biddle 2015 reported two‐hour post‐challenge glucose and glycosylated haemoglobin (HbA1c). Finni 2011 included homeostatic model assessment of insulin resistance (HOMA‐IR) and homeostatic model assessment of β‐cell function (HOMA‐%B).

One study included mood state as an outcome variable and used a Profle of Mood State (POMS) to assess mood over the course of the previous week (Ellingson 2016). Another study assessed wellness using an online version of the Wellness Evaluation of Lifestyle (WEL) inventory (Barwais 2013).

Five studies reported MVPA as measured by an accelerometer (Biddle 2015; Ellingson 2016; Finni 2011; Jago 2013; Williams 2019). Different methods of classifying moderate and vigorous physical activity were used across studies. Two studies used widely known cut‐points developed by Freedson and Troiano and colleagues (Biddle 2015; Freedson 1998; Jago 2013; Troiano 2008). Finni 2011 used cut‐points calibrated from Troiano 2008. Ellingson 2016 employed the sojourn method (Lyden 2014) and Williams used the thresholds developed by Esliger 2011.

Self‐report physical activity was measured in several studies using a variety of assessment tools including the 7‐Day Sedentary and Light‐Intensity Physical Activity (SLIPA) Log (Barwais 2013; Cotten 2016), the seven‐day physical activity recall questionnaire (Cotten 2016), an unspecified physical activity questionnaire (Lioret 2012), the short form of the International Physical Activity Questionnaire (IPAQ) (Barwais 2013; Biddle 2015), and a modified version of the long‐form IPAQ (French 2011). Ellingson 2016 did not report which version of IPAQ was used. Note that some studies used several self‐report instruments, with each assessing activities of different intensity.

Three studies reported step counts per day. One of these studies measured steps using a thigh‐worn ActivPAL accelerometer (Arrogi 2017), another measured steps using the waist‐worn Actigraph GT3X accelerometer (Biddle 2015) and the third measured steps using a wrist‐worn UP24 accelerometer Kitagawa 2020.

Duration of follow‐up varied across studies. In seven studies, the duration of longest follow‐up was four months or less (Arrogi 2017; Barwais 2013; Cotten 2016; Ellingson 2016; Jago 2013; Kitagawa 2020; Sui 2018); these were defined as providing short‐term follow‐up. Four studies conducted a follow‐up measure between 4 and 12 months (Biddle 2015; Finni 2011; French 2011; Williams 2019); these were considered to provide medium‐term follow‐up. We categorised studies with a follow‐up measure longer than 12 months as providing long‐term follow‐up. Only two studies met this criteria (Laska 2016; Lioret 2012). If studies reported multiple follow‐up time points, the data for each time point were pooled as appropriate with short‐, medium‐ or long‐term studies.

Excluded studies

See Characteristics of excluded studies.

Risk of bias in included studies

Risk of bias varied across studies (Figure 3). No study was judged to be at low risk of bias across all domains (Figure 4). Only one study was assessed as having low risk of bias in six of the seven domains (Biddle 2015).

Figure 3

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 4

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Allocation

Most studies described a random component in the sequence generation process (Barwais 2013; Biddle 2015; Cotten 2016; Ellingson 2016; Finni 2011 Jago 2013; Kitagawa 2020; Laska 2016; Lioret 2012; Sui 2018; Williams 2019); we therefore judged them to be at low risk of bias. Methods included utilising a random number generator (Barwais 2013; Cotten 2016; Ellingson 2016; Lioret 2012), a computer‐generated list or sequence (Biddle 2015; Jago 2013; Kitagawa 2020; Laska 2016), a web service (Sui 2018; Williams 2019), and coin flipping (Finni 2011).

The remaining two studies did not provide sufficient information to permit a judgement (Arrogi 2017; French 2011).

Selection bias due to inadequate concealment of allocations before group assignment was judged as having unclear risk in nine studies due to failure to provide sufficient information to assign a judgement of low or high risk (Arrogi 2017; Barwais 2013; Cotten 2016; Ellingson 2016; Finni 2011; French 2011; Kitagawa 2020; Laska 2016; Sui 2018). Four studies reported appropriate methods used to conceal allocation to intervention or control and were therefore judged as having low risk of bias (Biddle 2015; Jago 2013; Lioret 2012; Williams 2019). Allocation was determined by an independent statistician (Biddle 2015; Jago 2013; Lioret 2012) or researcher (Williams 2019).

Blinding

Due to the nature of the intervention, performance bias was not considered for this review. For detection bias, we considered outcome assessment and outcome assessors separately.

For outcome assessment, six studies included a device‐based measure of sedentary behaviour (Arrogi 2017; Biddle 2015; Ellingson 2016; Finni 2011; Kitagawa 2020; Williams 2019); these were judged to be at low risk of bias. Seven studies used only self‐report measures of sedentary behaviour, and we judged these to have high risk of bias for outcome assessment (Barwais 2013; Cotten 2016; French 2011; Jago 2013; Laska 2016 Lioret 2012; Sui 2018).

Regarding blinding of outcome assessors, three studies reported any level of blinding to group assignment among those responsible for data entry or analysis. We therefore judged Ellingson 2016, Kitagawa 2020 and Williams 2019 to be at low risk of bias. Sui 2018 stated that assessors were not blinded to group allocation, and we judged this study to be at high risk of bias. As the remaining studies did not report on blinding, we judged them to have unclear risk of bias (Arrogi 2017; Barwais 2013; Biddle 2015; Cotten 2016; Finni 2011; French 2011; Jago 2013; Laska 2016; Lioret 2012).

Incomplete outcome data

Ten studies were judged to be at low risk for attrition bias. Missing data were similar across groups in seven of these studies (Arrogi 2017; Biddle 2015; Cotten 2016; Finni 2011; French 2011; Lioret 2012; Sui 2018). In Barwais 2013 and Kitagawa 2020 there did not appear to be any missing data. In Ellingson 2016, the reasons provided for missing data were unlikely to be related to the outcome.

Two studies were judged to have high risk of attrition bias, as dropout was not balanced across intervention and control groups (Jago 2013; Williams 2019).

Laska 2016 was considered to be at unclear risk of bias, as reporting of attrition was insufficient to permit a judgement.

Selective reporting

We judged the majority of studies to be at unclear risk for reporting bias as information was insufficient to permit a judgement of high or low risk (Barwais 2013; Ellingson 2016; French 2011; Jago 2013; Kitagawa 2020; Laska 2016; Sui 2018). Three studies were deemed to be at high risk of bias (Cotten 2016; Lioret 2012; Williams 2019). In the case of Lioret 2012 and Williams 2019, some variables outlined in the protocol were not reported. Another study reported that the self‐reported sitting measure was not included in the analysis, and this decision was made after data collection (Cotten 2016).

Other potential sources of bias

We judged the majority of studies to have low risk of other potential sources of bias (Arrogi 2017; Barwais 2013; Biddle 2015; Ellingson 2016; Finni 2011; French 2011; Jago 2013; Laska 2016; Lioret 2012; Williams 2019). Cotten 2016 was deemed to be at high risk of bias as not all instruments used were established validated tools of sedentary behaviour. The remaining two studies were deemed as having unclear risk of bias due to insufficient information to assess whether an important risk of bias exists (Kitagawa 2020; Sui 2018).

Additional risk of bias domains for cluster RCTs

Five additional items were considered for cluster RCTs: recruitment bias; baseline imbalance; loss of clusters; incorrect analysis; and comparability with individually randomised trials.

We judged two studies to be at low risk of recruitment bias (French 2011; Lioret 2012). In Finni 2011, recruitment occurred after randomisation of clusters, and this study was considered to be at high risk of bias.

For baseline imbalance, information was insufficient to permit a judgement of high or low risk; therefore all studies were deemed to be at unclear risk of bias.

French 2011 reported similar dropout across groups; we judged this study to have low risk of bias in relation to loss of clusters. For Finni 2011 and Lioret 2012, information was insufficient to permit a judgement of high or low risk; we therefore judged these studies as having unclear risk of bias.

We judged all three cluster RCTs to have low risk of bias for incorrect analysis (Finni 2011; French 2011; Lioret 2012), as all reported adjusted results.

Regarding the risk of bias associated with comparability with individually randomised trials, one study was considered as having low risk of bias because clustering took place at the level of parent groups, and therefore contamination between groups was unlikely (Lioret 2012). In French 2011, households were recruited from various sites in the community; therefore the possibility of contamination existed. We judged this study as having high risk of bias. Finally, for Finni 2011, information was insufficient to permit a judgement of high or low risk, and this study was therefore deemed to be at unclear risk of bias.

Effects of interventions

See summary of findings Table 1 for the main comparison. We present results using only outcomes for which data were available. We were unable to address all of the secondary objectives set for this review due to lack of available data.

Primary outcomes

We pooled studies according to outcome measures.

Device‐measured sedentary time

We pooled four studies that compared effects of the intervention versus control on device‐measured sedentary time with short‐term follow‐up (up to four months) (Arrogi 2017; Biddle 2015; Ellingson 2016; Finni 2011). The interventions probably made little or no difference in device‐measured sedentary time in adults under 60 years of age (mean difference (MD) ‐8.36 min/d, 95% confidence interval (CI) ‐27.12 to 0.40; I² = 0%; Analysis 1.1; Figure 5). Results were imprecise due to wide confidence intervals and small sample sizes. Overall the certainty of evidence was moderate; therefore further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.

Figure 5

Forest plot of comparison: 1 Intervention vs control, outcome: 1.1 Device‐measured sedentary time (min/day)

Three studies provided data for medium‐term follow‐up (longer than 4 to 12 months) (Biddle 2015; Finni 2011; Williams 2019). We are uncertain whether interventions reduce device‐measured sedentary time in the medium term (MD ‐51.37 min/d, 95% CI ‐126.34 to 23.59; I² = 84%; Analysis 1.1; Figure 5). Results were imprecise due to wide confidence intervals and small sample sizes. In addition we had concerns about risk of bias, with two studies having high risk of bias for several domains. There was also a large variation in effect. The certainty of evidence was very low; therefore any estimate of effect is very uncertain.

Self‐report sitting time

We are uncertain whether interventions outside the workplace reduces self‐reported sitting time in the short term. Evidence was drawn from two studies with 61 participants (Barwais 2013; Ellingson 2016) (MD ‐64.12 min/d, 95% CI ‐260.91 to 132.67; I² = 86%; Analysis 1.2; Figure 6). We had concerns about risk of bias, as several domains were judged to be unclear and outcomes assessment was at high risk of bias. Results were inconsistent due to large variation in effect and high levels of heterogeneity. Both studies recruited participants who reported high levels of sedentary behaviour and employed interventions of similar duration using personal activity monitors. It is unclear why the direct effect was different. In addition, imprecision was evident due to the wide confidence intervals and small sample sizes. Overall the certainty of evidence was very low; therefore any estimate of effect is very uncertain.

Figure 6

Forest plot of comparison: 1 Intervention vs control, outcome: 1.2 Self‐report sitting time (min/d).

Self‐report TV viewing time

Interventions outside the workplace may make little or no difference in self‐reported TV viewing in the medium term (MD ‐12.45 min/d, 95% CI ‐50.40 to 25.49; I² = 86%; Analysis 1.3; Figure 7). We pooled two studies with 459 participants that recorded data at medium‐term follow‐up (French 2011; Laska 2016). The width of confidence intervals raises concerns about imprecision. We determined that studies were at high risk of bias for outcome assessment, and we judged several other domains to be unclear.

Figure 7

Forest plot of comparison: 1 Intervention vs control, outcome: 1.3 Self‐report TV viewing time (min/day)

We pooled two studies that provided data for the comparison at long‐term follow‐up (> 12 months) (Laska 2016; Lioret 2012). Interventions may make little or no difference in self‐report TV viewing in the long term (MD 0.30 min/d, 95% CI ‐0.63 to 1.23; I² = 0%; Analysis 1.7; Figure 7). Sample sizes were large, but the large confidence intervals led to uncertainty. There was high risk of bias for outcome assessment, and we judged several other domains to be at unclear risk.

Overall the certainty of evidence for self‐report TV viewing time was low; therefore further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.

Breaks in sedentary time

It was not possible to pool all studies that reported breaks in sedentary time given the variation in the definition used for a "break" (Arrogi 2017; Biddle 2015; Cotten 2016; Finni 2011; Sui 2018). Table 3 provides a summary of data collection methods, break definitions, and study findings.

Table 3. Summary of studies reporting breaks in sedentary time

Study	Definition of break in sedentary time	Measure	Findings
Arrogi 2017	Number of sit‐to‐stand transitions per day	ActivPAL Monitor	Short‐term follow‐up: MD 5.7 per day (94% CI 1.0 to 10.4)
Biddle 2015	Bouts of light to vigorous physical activity per day	Actigraph Accelerometer	Short‐term follow‐up: MD ‐29.6 (95% CI ‐97.0 to 37.9) Medium‐term follow‐up: MD ‐2.96 (95% CI ‐73.0 to 67.0)
Cotten 2016	Frequency of breaks to "get up and move around every day" (break every X minutes). Options were provided every 30 minutes or less, 45 minutes, 60 minutes, 75 minutes, 90 minutes, 120 minutes, 180 minutes, or 240 minutes or longer	Questionnaire	Short‐term follow‐up: MD ‐10.25 (‐25.58 to 5.08)
Finni 2011	An interruption in sedentary time when accelerometer counts rose up to or above 100 counts/min for a minimum of 1 minute. Breaks per sedentary hour	Alive Technologies Accelerometer	Short‐term follow‐up: MD 1.0 (95% CI ‐0.2 to 2.2) Medium‐term follow‐up: MD 0.6 (95% CI ‐0.6 to 1.8)
Sui 2018	Frequency of interrupting sitting time (break every X minutes). Options included less than every 30 min, every 30 to 45 min, every 45 minutes to 1 hour, every 1 to 1.5 hours, every 1.5 to 2 hours, every 2 to 3 hours, every 3 to 4 hours, every 4 to 5 hours, over every 5 hours, no interruption	Questionnaire	Short‐term follow‐up: MD ‐53.12 (‐96.98 to ‐9.28)

CI: confidence interval.

MD: mean difference.

In Arrogi 2017 a mean difference in sit‐to‐stand transitions of 5.7 per day (95% CI 1.0 to 10.4 ) was reported at short‐term follow‐up. Biddle 2015 reported the number of breaks in sedentary behaviour per day (i.e. light to vigorous PA bouts) at short‐term (MD ‐29.6, 95% 95% CI ‐97.0 to 37.9) and medium‐term follow‐up (MD ‐2.96, 95% CI ‐73.0 to 67.0). Cotten 2016 reported self‐reported frequency of breaks at short‐term follow‐up (MD ‐10.25, 95% CI ‐25.58 to 5.08). Finni 2011 reported device‐determined breaks per sedentary hour during leisure time at short‐term (MD 1.0, 95% CI ‐0.2 to 2.2) and medium‐term follow‐up (MD 0.6, 95% CI ‐0.6 to 1.8). In Sui 2018, a MD of ‐53.12 (‐96.98 to ‐9.28) in self‐reported break frequency was found at short‐term.

Secondary outcomes

Energy expenditure

Laska 2016 reported that the intervention did not change energy expenditure in leisure‐time physical activity at 24 months (MD ‐66.5 weekly kcal).

Body composition

We pooled three studies that reported medium‐term follow‐up points for body mass index (Biddle 2015; Finni 2011; French 2011) (MD ‐0.25 kg/m², 95% CI ‐0.48 to ‐0.01; I² = 0%; moderate‐certainty evidence; Analysis 1.4). Most studies had high risk of bias for two domains. Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate. Biddle 2015 also reported data for short‐term follow‐up (MD ‐13.3 kg/m², 95% CI ‐15.0 to 41.6).

Biddle 2015 reported no changes in waist circumference at 3 months (MD 13.4 cm, 95% CI ‐14.7 to 41.5). We pooled two studies that report waist circumference at medium‐term follow‐up (Biddle 2015; Williams 2019) (MD ‐2.04 cm 95% CI ‐9.06 to 4.98, I² = 65%; low‐certainty evidence; Analysis 1.5). There were concerns about imprecision due to small sample size. In addition, one study had high risk of bias for two domains. Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.

Biddle 2015 reported no change in body fat percentage at 3 months (MD 0.54, 95% CI ‐38.4 to 39.5) or at 6 months (MD ‐0.25, 95% CI ‐2.72 to 2.22). In Finni 2011, there was no change in total fat mass percentage at 6 months (MD ‐0.54, 95% CI ‐1.2 to 0.11) or at 12 months (MD ‐0.60, 95% CI ‐1.26 to 0.07); however percentage leg lean mass changed between groups at 12 months (MD 0.48, 95% CI 0.18 to 0.77), and a decrease within the control group was observed.

Biddle 2015 reported no changes in body weight at 3 months (MD 14.3 kg, 95% CI ‐16.1 to 44.6) or at 6 months (MD 0.46 kg, 95% CI ‐5.06 to 5.97), whereas Finni 2011 reported changes at 6 months (MD ‐0.83 kg, 95% CI ‐1.64 to ‐0.02) and at 12 months (MD ‐0.95 kg, 95% CI ‐1.76 to ‐0.13).

Cholesterol

Only one study reported the effects of the intervention on cholesterol (Biddle 2015). No changes were noted for serum total cholesterol (MD 0.21 mmol/L, 95% CI ‐0.15 to 0.40), HDL cholesterol (MD 0.06 mmol/L, 95% CI ‐0.02 to 0.14; P= 0.122), triglycerides (MD ‐0.17 mmol/L, 95% CI, ‐0.42 to 0.08), or LDL cholesterol (MD 0.13 mmol/L, 95% CI ‐0.10 to 0.36) at 3 months and at 12 months (findings for 12 months are shown).

Markers of insulin resistance

We pooled two studies (Biddle 2015; Finni 2011 that reported glucose (Biddle 2015; Finni 2011) at short‐term follow‐up (MD ‐0.18 mmol/L, 95% CI ‐0.30 to ‐0.06, I² = 0%; moderate‐certainty evidence; Analysis 1.6) and medium‐term follow‐up (MD ‐0.08 mmol/L, 95% CI ‐0.21 to 0.05, I² = 0%; moderate‐certainty evidence; Analysis 1.6). There were concerns in relation to risk of bias as unclear or high for several domains. Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate.

Finni 2011 noted no change in insulin at 3 months (MD ‐0.55 pM, 95% CI ‐9.15 to 8.05) or at 6, 9, or 12 months (MD ‐1.47 pM, 95% CI ‐10.3 to 7.37).

Biddle 2015 reported two‐hour glucose challenge at 3 months (MD ‐0.08 mmol/L, 95% CI ‐0.63 to 0.46) and 12 months (MD 0.10 mmol/L, 95% CI ‐0.65 to 0.84). No changes were seen in HbA1c at 3 months (MD 0.01%, 95% CI ‐0.08 to 0.11) or at 12 months (MD 0.06%, 95% CI ‐0.04 to 0.16).

Finni 2011 reported HOMA‐%B at 3 months (MD 12.62, 95% CI ‐6.81 to 32.05) and 12 months (MD 7.93, 95% CI ‐12.1 to 27.96).

Inflammatory markers

No data were reported.

Meaures of carotid intima media thickness

No data were reported.

Measures of endothelial function

No data were reported.

Meaures of mental health

Mood

Ellingson 2016 reported a group by time interaction for mood state, favouring the intervention group (F_1,27 = 4.17; P = 0.05).

Wellness

Barwais 2013 revealed a time by treatment effect on total wellness (F_1,31 = 9.5; P < 0.001), with increases in wellness scores seen in the intervention group (t₁₇ = ‐6.5; P < 0.001).

Adverse events and symptoms

No data were reported.

Unintended outcomes

Device‐measured moderate to vigorous physical activity

Interventions outside the workplace may make little or no difference in device‐measured MVPA in the short term (MD 1.99 min/d, 95% CI ‐4.27 to 8.25; I² = 23%; low‐certainty evidence; Analysis 1.7; Figure 8). Four studies with 296 participants provided data for this outcome (Biddle 2015; Ellingson 2016; Finni 2011; Jago 2013). Most studies had unclear or high risk of bias in several domains. Small sample sizes and wide confidence intervals raised concerns about imprecision. Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.

Figure 8

Forest plot of comparison: 1 Intervention vs control, outcome: 1.6 Device‐measured MVPA (min/day)

We are uncertain whether interventions improve device‐measured MVPA in the medium term (MD 6.59 min/d, 95% CI ‐7.35 to 20.53; I² = 70%; very low‐certainty evidence; Analysis 1.7). Three studies with 214 participants were pooled (Biddle 2015; Finni 2011;Williams 2019). Results were imprecise due to wide confidence intervals and small sample sizes. Two studies had unclear or high risk of bias for several domains. In addition there was a large variation in effect. The certainty of evidence was very‐low and any estimate of effect is very uncertain.

Self‐report moderate to vigorous physical activity

French 2011 reported self‐report MPVA at medium‐term follow‐up (MD 29.6331 min/d; 95%CI 4.60 ‐ 54.66).

Self‐report light‐intensity physical activity

Two studies with 115 participants reported data on light‐intensity PA, assessed using the 7‐Day SLIPA Log (Barwais 2013; Cotten 2016). We are uncertain whether interventions outside the workplace improve light‐intensity PA in the short‐term (MD 156.32 min/d, 95% CI 34.34 to 278.31; I² = 79%; Analysis 1.8). Results were inconsistent due to large variation in effect and high levels of heterogeneity. We had very serious concerns about precision due to the very large confidence intervals. In addition, risk of bias was high for outcome assessment and unclear for several domains. The certainty of evidence was very low, meaning that any estimate of effect is very uncertain.

Self‐report moderate‐intensity physical activity

Three studies reported self‐report moderate‐intensity PA with short‐term follow‐up. Cotten 2016 reported moderate‐intensity activity as minutes per week with a MD of 50.39 (95% CI ‐76.27 to 177.05). Barwais 2013 found a MD of 457.0 MET/min per week (95%CI 202.43 to 711.58).The third study found a MD of ‐22.10 min/d (95%CI ‐87.59 to 43.39) (Ellingson 2016).

Self‐report vigorous‐intensity physical activity

Barwais 2013 reported vigorous‐intensity physical activity in MET‐minutes per week and found a MD of 540.00 (95% CI 129.53 to 950.47) at short‐term follow‐up. Ellingson 2016 reported minutes per day of vigorous activity at short‐term (MD ‐15.60, 95% CI ‐33.37 to 2.17). Biddle 2015 found a MD of ‐2.5 vigorous METs at short‐term (95% CI ‐536.9 to 531.8) and ‐242.0 (95% CI ‐849.9 to 365.8) at long‐term follow‐up.

Self‐report total physical activity

Lioret 2012 found no change in self‐report total physical activity at long‐term follow‐up (MD 0.75 min/week, 95% CI ‐0.90 to 2.40).

Steps

We pooled three studies that reported step count at short‐term follow‐up (Arrogi 2017; Biddle 2015; Kitagawa 2020) (MD 226.90 steps/day, 95% CI ‐519.78 to 973.59, I² = 0%; low‐certainty evidence; Analysis 1.9). There were concerns about imprecision due to large confidence internals and small sample sizes. Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.

Biddle 2015 also reported data for medium‐term follow‐up and found no change in average steps per day (MD 402.7, 95% CI ‐807.9 to 1613.4).

Discussion

Summary of main results

We included 13 studies in this review. A synopsis of findings for the primary outcomes of device‐measured sedentary time, self‐report sitting time, and self‐report TV viewing time can be seen in Summary of findings table 1. Interventions outside the workplace to reduce sedentary behaviour probably make little or no difference in device‐measured sedentary time in adults under 60 years of age in the short term (moderate‐certainty evidence). We are uncertain whether such interventions reduce device‐measured sedentary time in the medium term (very low‐certainty evidence). We are uncertain whether interventions outside the workplace reduce self‐reported sitting time in the short term (very low‐certainty evidence). Interventions may make little or no difference in self‐reported TV viewing time (low‐certainty evidence). We were unable to pool studies that reported breaks in sedentary time due to the multiple ways that studies defined a break.

We were able to complete meta‐analyses for only five of the secondary outcomes: body mass index, waist circumference, device‐measured moderate to vigorous physical activity (MVPA), self‐report light‐intensity physical activity (PA) and step count. Interventions outside the workplace to reduce sedentary behaviour probably make little or no difference in BMI among adults under 60 years of age in the medium term (moderate‐certainty evidence). Interventions outside the workplace may make little or no difference in device‐measured MVPA in the short term (low‐certainty evidence). We are uncertain whether interventions increase device‐measured MVPA in the medium term (very low‐certainty evidence) . We are uncertain whether interventions outside the workplace improve self‐report light‐intensity PA (very low‐certainty evidence). Interventions outside the workplace may make little or no difference in daily step count in the short term (low‐certainty evidence)

A majority of study participants were female and were aged 20 to 41 years. Two studies reported follow‐up measures undertaken more than 12 months post baseline. Despite six studies assessing outcomes using device‐based measures, the overall certainty of evidence was determined as moderate to very low.

All interventions were delivered at the individual level, and none were considered environmental or policy activities. Interventions at an individual level included information/education, counselling, wearable technology, apps, SMS prompts, web‐based interventions and phone calls. Interventions were delivered in home or community settings, within primary care or educational settings. Details of the frequency and intensity of these different elements and their fidelity were missing. No interventions yielded data on cost‐effectiveness, quality of life, or adverse events.

Overall completeness and applicability of evidence

A majority of studies recruited female participants and younger participants (aged 20 to 41) in high‐income countries. Therefore it is not clear to what extent these types of interventions might be effective among other population groups.

The identified studies were insufficient to address all of the objectives of the review. We could not investigate most of the secondary outcomes (related to health effects) due to absence of eligible studies reporting these outcomes. BMI status was reported in only three studies. No studies reported data on adverse events or cost‐effectiveness.

We were unable to determine whether specific components of interventions are associated with changes in sedentary behaviour. In addition, we could not examine if there were any differential effects of interventions based on health inequalities. Very few studies reported sociodemographic characteristics at baseline and endpoint using the PROGRESS framework (Place, Race, Occupation, Gender, Religion, Education, Socioeconomic status, Social status). No studies were conducted in low‐ or middle‐income countries, and we found a dearth of quality research in marginalised and poorer populations.

This review was limited to trials involving a randomised controlled trial (RCT) or cluster RCT study design. It is possible that other evidence is available from other, less robust, studies.

Overall the evidence presented may not hold true for middle‐aged adults and those from low‐middle‐income countries (for which trials were not available) or for adults over 60 years of age (outside the scope of the review).

Quality of the evidence

The body of evidence identified does not allow a robust conclusion regarding the objectives of this review. We included data from 13 studies and 1770 participants. We found certainty of evidence, according to GRADE, to range from very low to moderate. Uncertainty was mainly due to concerns about imprecision, risk of bias, and inconsistency. Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate.

Available studies demonstrate that it is possible to conduct RCTs of non‐workplace interventions to reduce sedentary behaviour; however no studies were considered to have low risk of bias. In addition to the limitations of self‐report tools, a common challenge to undertaking high‐quality trials relates to blinding of outcome assessments. Blinding was possible in studies that used device‐based measures. Most studies used self‐report measures of sedentary behaviour. These measures are subject to recall bias and may lack precision; however any misclassification is non‐differential (as both intervention and control groups complete the measure) and attenuates the effects of the intervention (Richards 2013). We also note that the use of self‐report measures and the wide ranges of values for sedentary behaviours could contribute to uncertainty around intervention effects. Although none of our studies blinded participants to their group allocation, we believe that this criterion was not appropriate because it is very difficult to do this for any intervention that requires movement.

Potential biases in the review process

Assessing risk of bias and certainty of the body of evidence (GRADE) involves personal judgement and the potential for some degree of subjectivity (Hombali 2019). We tried to minimise the potential for bias by ensuring that two review authors conducted these assessments independently. To avoid language bias, we did not impose any language restrictions in our searches. To increase the likelihood that all relevant studies were identified, we (1) emailed trial authors and research groups for information about unpublished or ongoing studies, (2) contacted experts in the field and asked them to identify further articles, (3) handsearched reference lists of included studies and key systematic reviews, and (4) searched a clinical trial registry and the websites of organisations involved in addressing and reporting research on sedentary behaviour.

One limitation of this review is the potential for publication bias. Unpublished studies may exist but have not been submitted or accepted for publication, or only those with positive results may have been published. However we were unable to examine a funnel plot due to the small numbers of studies with common outcome measures. One principal investigator declared receiving in‐kind support through provision of a sit‐to‐stand desk by Ergotron from 2012‐2014. This study reported no intervention effects (Biddle 2015).

Agreements and disagreements with other studies or reviews

This is the first review of interventions intended to change sedentary behaviour delivered in contexts outside of workplaces. This review is relevant, as adults spend approximately 70% of their non‐work time being sedentary, plus not all adults work.

A recent systematic review reported that 10 of the 13 studies observed a reduction in objectively measured sitting time (Thraen‐Borowski 2017). However this review included non‐randomised and uncontrolled studies. Another review ‐ including only RCTs ‐ noted that effects reported between 7 and 12 months were not sustained beyond 12 months, with high heterogeneity between studies (Martin 2015). However only eight non‐workplace studies were identified for the review. In addition, in the subgroup analysis by study setting (workplace vs home/community), no attempt was made to discriminate between interventions that were designed to increase physical activity and those that purposely aimed to reduce sedentary behaviour. This is important, as the same review and another demonstrated differential effects between interventions that solely aim to reduce sedentary behaviour or take a combined approach of reducing sedentary behaviour and increasing physical activity (Gardner 2016).

We were concerned that most of our studies recruited younger adults outside of workplace settings; this has the potential to increase health inequalities (i.e. differential responses in recruitment and focus on effects on the younger and potentially more active). We support the conclusions of these reviews that it remains a priority to identify what is a clinically meaningful change in sedentary behaviour.

Figure 1

Logic Model for interventions targeted outside of workplace settings for reducing sedentary behaviour (adapted from Baker 2015).

Figure 2

Study flow diagram.

Figure 3

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 4

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Figure 5

Forest plot of comparison: 1 Intervention vs control, outcome: 1.1 Device‐measured sedentary time (min/day)

Figure 6

Forest plot of comparison: 1 Intervention vs control, outcome: 1.2 Self‐report sitting time (min/d).

Figure 7

Forest plot of comparison: 1 Intervention vs control, outcome: 1.3 Self‐report TV viewing time (min/day)

Figure 8

Forest plot of comparison: 1 Intervention vs control, outcome: 1.6 Device‐measured MVPA (min/day)

Analysis 1.1

Comparison 1: Intervention vs control, Outcome 1: Device‐measured sedentary time

Analysis 1.2

Comparison 1: Intervention vs control, Outcome 2: Self‐report sitting time

Analysis 1.3

Comparison 1: Intervention vs control, Outcome 3: Self‐report TV viewing time

Analysis 1.4

Comparison 1: Intervention vs control, Outcome 4: Body mass index

Analysis 1.5

Comparison 1: Intervention vs control, Outcome 5: Waist circumference

Analysis 1.6

Comparison 1: Intervention vs control, Outcome 6: Glucose

Analysis 1.7

Comparison 1: Intervention vs control, Outcome 7: Device‐measured MVPA

Analysis 1.8

Comparison 1: Intervention vs control, Outcome 8: Self‐report light PA

Analysis 1.9

Comparison 1: Intervention vs control, Outcome 9: Steps

Summary of findings 1. Intervention compared to Control for reducing sedentary behaviour in adults under 60

Outcomes	*Anticipated absolute effects^ (95% CI)**		№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Intervention compared to control for reducing sedentary behaviour in adults under 60
Patient or population: community‐dwelling adults under 60 years of age Setting: outside the workplace Intervention: individual‐level interventions aiming to reduce sedentary behaviour Comparison: no intervention or attention control
Outcomes	Risk with Control	Risk with Intervention	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
Device‐measured sedentary time	Short‐term follow‐up (up to 4 months)
	Control group mean was 574.44 min/d	MD 8.36 lower (‐27.12 lower to 10.40 higher	262 (4 RCTs)	⊕⊕⊕⊝ MODERATE ¹
	Medium‐term follow‐up (> 4 months to 12 months)
	Control group mean was 590.67 min/day	MD 51.37 lower (126.34 lower to 23.59 higher	188 (3 RCTs)	⊕⊝⊝⊝ VERY LOW ^{1 2 3}
Self‐report TV viewing time	Medium follow‐up (> 4 months to 12 months)
	Control group mean was 99.30 min/day	MD 12.45 lower (50.40 lower to 25.49 higher)	459 (2 RCTs)	⊕⊕⊝⊝ LOW ^{1 4}
	Long‐term follow‐up (> 12 months)
	Control group mean was 111.22	MD 0.30 higher (0.63 lower to 1.23 higher)	709 (2 RCTs)	⊕⊕⊝⊝ LOW ^{5 6}
Device‐measured MVPA	Short‐term (up to 4 months)
	Control group mean was 48.76 min/day	MD 1.99 higher (4.27 lower to 8.25 higher)	296 (4 RCTs)	⊕⊕⊝⊝ LOW ^{1 7}
	Medium‐term follow‐up (> 4 months to 12 months)
	Control group mean was 62.97 min/day	MD 6.59 higher (7.35 lower to 20.53 higher)	214 (3 RCTs)	⊕⊝⊝⊝ VERY LOW ^{1 2 8}
Self‐report light PA	Short‐term follow‐up (up to 4 months)
Self‐report light PA	Control group mean was 232.86 min/day	MD 156.32 higher (34.34 higher to 278.31 higher)	115 (2 RCTs)	⊕⊝⊝⊝ VERY LOW ^{5 9 10}
Adverse events and symptoms	None reported
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI:Confidence interval; MD: mean difference; RCT: randomised controlled trial; min/day: minutes per day
GRADE Working Group grades of evidence High certainty: We are very confident that the true effect lies close to that of the estimate of the effect Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect
¹ Concerns about imprecision due to wide confidence intervals and small sample sizes ² Low risk of bias for outcome assessment, however 2 studies have high risk for several domains ³ Large variation in effect, I² = 87% ⁴ High risk of bias for outcome assessment for this outcome, unclear risk of bias for several other domains. ⁵ High risk of bias for outcomes assessment, unclear or high risk of bias for several other domains ⁶ Large sample size however large confidence intervals lead to uncertainty ⁷ Low risk of bias for outcome assessment, however majority of studies have unclear or high risk of bias for several domains ⁸ Large variation in effect, I² = 71% ⁹ Large variation in effect, CI's slightly overlap, Chi2 P < 0.05, I² = 79% ¹⁰ Very serious concerns about precision due to large confidence internals and small sample size

Summary of findings 1. Intervention compared to Control for reducing sedentary behaviour in adults under 60

Table 1. Definitions for quality ratings in GRADE

Quality level	Definition
High	Further research is very unlikely to change our confidence in the estimate of effect
Moderate	Further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate
Low	Further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate
Very low	Any estimate of effect is very uncertain

Table 1. Definitions for quality ratings in GRADE

Table 2. Summary of the interventions

Study	Target group	SB/PA‐related aim	Intervention components	Last follow‐up	Theory
Personal monitoring device
Arrogi 2017	Adults aged 18 to 55 years with sedentary jobs and/or predominantly sedentary leisure time	SB	Motion sensor and smartphone app to provide feedback in relation to targets, including a warning signal to modify behaviour	2 weeks	BCT principles
Barwais 2013	People who reported high total sitting time	SB + PA	Online personal activity monitor to provide feedback on progress towards their daily goal	4 weeks	NR
Kitagawa 2020	Housewives	SB	Motion sensor and smartphone application to provide feedback on sitting time. The tailored feedback group also received individual suggestions to shorten sitting time.	2 weeks	NR
Information/Education
French 2011	At least one child aged at least 5 years and two household members aged at least 12 years, with weekly average household TV viewing of at least 10 hours per person	SB	6 monthly face‐to‐face group sessions, monthly newsletters, 12 home‐based activities, a TV‐limiting device, and monthly telephone support calls	12 months	NR
Laska 2016	Adults aged 18 to 35 years, with BMI of 20 to 34.9 kg/m²	SB + PA	1 credit course at university plus social networking and a support website	24 months	Ecological theories of health behaviour, social‐cognitive theory, and social network theory
Lioret 2012	First‐time mothers	SB + PA	6 quarterly 2‐hour sessions at parent group and a newsletter	18 months	Theory of anticipatory guidance
Jago 2013	Parents with at least one child aged 6 to 8 years	SB	8 weekly 2‐hour education sessions	16 weeks	Self‐determination theory
Information/Education plus personal monitoring device
Biddle 2015	Overweight or obese adults aged 18 to 40 years with one or more additional risk factors for diabetes	SB	1 × 3‐hour group workshop and a self‐monitoring device to view and track progress against personal goals and provide feedback via vibration notifications	12 months	Bandura’s social‐cognitive theory, Gollwitzer’s implementation intentions concept, Behavioural Choice Theory, and Leventhal’s Common Sense Model
Ellingson 2016	Full‐time students aged 18 to 26 years, reporting more than 3 hours of daily leisure‐time sedentary behaviour	SB	10 to 15‐minute information session and personal monitor with vibration feedback to modify behaviour towards a set target	5 weeks	Habit theory
Counselling
Finni 2011	Healthy men and women with children aged 3 to 8 years and an occupation where they sit for more than 50% of their work time	SB	Tailored counselling, 2 telephone calls at 1 and 5 months, and motivational emails	12 months	Theory of planned behaviour
Sui 2018	Full‐time university students aged 18+ years with good physical and mental health	SB	Counselling sessions with an action plan	6 weeks	Health action process approach
Information/Education plus personal monitoring device plus counselling
Williams 2019	Community dwelling adults with serious mental illness	SB + PA	Initial education session, fortnightly one‐to‐one health coaching, provision of pedometers and access to a weekly walking group	17 weeks	COM‐B model of behaviour change
Daily text messages
Cotten 2016	Adults aged 18 to 65 years	SB	Daily text messages	6 weeks	NR
BCT: behaviour change technique. BMI: body mass index. NR: not reported. PA: physical activity. SB: sedentary behaviour.

Table 2. Summary of the interventions

Table 3. Summary of studies reporting breaks in sedentary time

Study	Definition of break in sedentary time	Measure	Findings
Arrogi 2017	Number of sit‐to‐stand transitions per day	ActivPAL Monitor	Short‐term follow‐up: MD 5.7 per day (94% CI 1.0 to 10.4)
Biddle 2015	Bouts of light to vigorous physical activity per day	Actigraph Accelerometer	Short‐term follow‐up: MD ‐29.6 (95% CI ‐97.0 to 37.9) Medium‐term follow‐up: MD ‐2.96 (95% CI ‐73.0 to 67.0)
Cotten 2016	Frequency of breaks to "get up and move around every day" (break every X minutes). Options were provided every 30 minutes or less, 45 minutes, 60 minutes, 75 minutes, 90 minutes, 120 minutes, 180 minutes, or 240 minutes or longer	Questionnaire	Short‐term follow‐up: MD ‐10.25 (‐25.58 to 5.08)
Finni 2011	An interruption in sedentary time when accelerometer counts rose up to or above 100 counts/min for a minimum of 1 minute. Breaks per sedentary hour	Alive Technologies Accelerometer	Short‐term follow‐up: MD 1.0 (95% CI ‐0.2 to 2.2) Medium‐term follow‐up: MD 0.6 (95% CI ‐0.6 to 1.8)
Sui 2018	Frequency of interrupting sitting time (break every X minutes). Options included less than every 30 min, every 30 to 45 min, every 45 minutes to 1 hour, every 1 to 1.5 hours, every 1.5 to 2 hours, every 2 to 3 hours, every 3 to 4 hours, every 4 to 5 hours, over every 5 hours, no interruption	Questionnaire	Short‐term follow‐up: MD ‐53.12 (‐96.98 to ‐9.28)
CI: confidence interval. MD: mean difference.

Table 3. Summary of studies reporting breaks in sedentary time

Comparison 1. Intervention vs control

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Device‐measured sedentary time Show forest plot	5		Mean Difference (IV, Random, 95% CI)	Subtotals only

1.1.1 Short‐term follow‐up (up to 4 months)	4	262	Mean Difference (IV, Random, 95% CI)	‐8.36 [‐27.12, 10.40]
1.1.2 Medium‐term follow‐up (> 4 months to 12 months)	3	188	Mean Difference (IV, Random, 95% CI)	‐51.37 [‐126.34, 23.59]
1.2 Self‐report sitting time Show forest plot	2		Mean Difference (IV, Random, 95% CI)	Subtotals only

1.2.1 Short‐term follow‐up (up to 4 months)	2	61	Mean Difference (IV, Random, 95% CI)	‐64.12 [‐260.91, 132.67]
1.3 Self‐report TV viewing time Show forest plot	3		Mean Difference (IV, Random, 95% CI)	Subtotals only

1.3.1 Medium‐term follow‐up (> 4 months to 12 months)	2	459	Mean Difference (IV, Random, 95% CI)	‐12.45 [‐50.40, 25.49]
1.3.2 Long‐term follow‐up (> 12 months)	2	709	Mean Difference (IV, Random, 95% CI)	0.30 [‐0.63, 1.23]
1.4 Body mass index Show forest plot	3		Mean Difference (IV, Random, 95% CI)	Subtotals only

1.4.1 Medium‐term follow‐up (> 4 months to 12 months)	3	326	Mean Difference (IV, Random, 95% CI)	‐0.25 [‐0.48, ‐0.01]
1.5 Waist circumference Show forest plot	2		Mean Difference (IV, Random, 95% CI)	Subtotals only

1.5.1 Medium‐term follow‐up (> 4 months to 12 months)	2	160	Mean Difference (IV, Random, 95% CI)	‐2.04 [‐9.06, 4.98]
1.6 Glucose Show forest plot	2		Mean Difference (IV, Random, 95% CI)	Subtotals only

1.6.1 Short‐term follow‐up (up to 4 months)	2	263	Mean Difference (IV, Random, 95% CI)	‐0.18 [‐0.30, ‐0.06]
1.6.2 Medium‐term follow‐up (> 4 months to 12 months)	2	238	Mean Difference (IV, Random, 95% CI)	‐0.08 [‐0.21, 0.05]
1.7 Device‐measured MVPA Show forest plot	5		Mean Difference (IV, Random, 95% CI)	Subtotals only

1.7.1 Short‐term follow‐up (up to 4 months)	4	296	Mean Difference (IV, Random, 95% CI)	1.99 [‐4.27, 8.25]
1.7.2 Medium‐term follow‐up (> 4 months to 12 months)	3	214	Mean Difference (IV, Random, 95% CI)	6.59 [‐7.35, 20.53]
1.8 Self‐report light PA Show forest plot	2	115	Mean Difference (IV, Random, 95% CI)	156.32 [34.34, 278.31]

1.8.1 Short‐term follow‐up (up to 4 months)	2	115	Mean Difference (IV, Random, 95% CI)	156.32 [34.34, 278.31]
1.9 Steps Show forest plot	3		Mean Difference (IV, Random, 95% CI)	Subtotals only

1.9.1 Short‐term follow‐up (up to 4 months)	3	204	Mean Difference (IV, Random, 95% CI)	226.90 [‐519.78, 973.59]

Comparison 1. Intervention vs control