Types of studies

We wanted our conclusions to be based on the best available evidence, so we chose to include only randomised controlled trials (RCTs) and cluster‐RCTs in the review.

Types of participants

Studies had to report data on participants with the following characteristics.

• Inclusion of a sample of adults of mean age of 60 years and over with no participants younger than 55 years old.

• Older adults living independently in the community, either at home or in sheltered housing or a residential complex that does not provide daily nursing or social care. We excluded older adults living in care home or nursing homes.

• Participants not recruited for belonging to specific clinical populations (e.g. stroke survivors) or having a specific condition or pathology. Participants may have comorbidities, but may not be recruited as a result of these.

Types of interventions

Studies had to report population‐, community‐, or individual‐based interventions specifically designed to reduce sedentary time, shorten lengths of prolonged sitting, and/or reduce particular sedentary behaviours, for example sitting time or watching television.

• Interventions targeting the raising of awareness and provision of information. This might include interventions which use technology (e.g. digital, mHealth) and social networks in addition to peer support networks.

• Interventions which use prompting on multiple occasions (vibration monitors, phones, emails) to raise awareness and provide feedback on sedentary behaviour.

• Interventions which attempt to alter the environment, e.g. modifying the layout of indoor and outdoor spaces and furniture.

• Interventions that aim to change culture, policy, and practice in people's work with older people, e.g. motivating staff to encourage older people to move more frequently.

We excluded studies reporting interventions aimed solely at increasing physical activity, but included interventions targeting both an increase in physical activity and a reduction in sedentary behaviour. The intervention could be delivered in community settings and within primary care. The interventions described above were compared with no intervention or with standard care or with interventions without sedentary behaviour change components.

Types of outcome measures

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

Electronic searches

We created a comprehensive search strategy derived from terms related to sedentary behaviour, study design type, and population of study participants. The search criteria were informed by previous reviews (e.g. Martin 2015 ; Prince 2014 ). We searched the following electronic databases up to 18 January 2021.

• Cochrane Central Register of Controlled Trials (CENTRAL) in the Cochrane Library. ( Appendix 1 )

• MEDLINE (PubMed). ( Appendix 2 )

• Embase (Embase.com). ( Appendix 3 )

• CINAHL EBSCO (Cumulative Index to Nursing and Allied Health Literature). ( Appendix 4 )

• PsycINFO (ProQuest). ( Appendix 5 )

• PEDro (Physiotherapy Evidence Database; www.pedro.org.au ). ( Appendix 6 )

• EPPI‐Centre databases ( eppi.ioe.ac.uk/cms/Default.aspx?tabid=185 ): Trials Register of Promoting Health Interventions (TRoPHI) and The database on obesity and sedentary behaviour studies. ( Appendix 7 )

• World Health Organisation (WHO) International Clinical Trials Registry Platform (ICTRP) ( apps.who.int/trialsearch ). ( Appendix 8 )

• US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov ( clinicaltrials.gov ). ( Appendix 9 )

We also screened the reference lists of included articles and contacted authors to identify additional studies.

Searching other resources

We screened the reference lists of included articles and identified systematic reviews for additional studies. We also contacted experts in the field through the International Physical and Environment Network ( www.ipenproject.org/index.html ), International Society for Physical Activity and Health ( www.ispah.org ), Sedentary Behaviour Research Network ( www.sedentarybehaviour.org ), and other leading international research networks to identify any additional work which is unpublished.

Selection of studies

We downloaded the references retrieved by the electronic and handsearches into Covidence ( Covidence ). Two review authors (out of a pool of nine) independently screened studies identified by the searches through two stages: 1) title and abstract screening and 2) full‐text screening using Covidence. Review authors coded the studies at each stage of the review process as 'included' or 'excluded'. The two review authors resolved any discrepancies regarding inclusion or exclusion by discussion or by consulting a third review author who acted as arbitrator. Review authors were not allowed to screen any studies on which they were co‐author. Duplicates of identified studies were excluded. We recorded the study selection process in a PRISMA flow diagram ( Moher 2009 ). We did not find any potentially relevant papers in a language other than English, so we did not require translation services.

Data extraction and management

We extracted the following data from the included studies using Covidence extraction templates.

• Methodological information: study design, randomisation, intervention duration, follow‐up duration, study date, context.

• Participant information: inclusion and exclusion criteria, sample size, age range, pre‐intervention and postintervention sitting time, health status, socioeconomic status.

• Sedentary behaviour monitoring method.

• Intervention information: intervention description, length, and comparison condition. We categorised interventions using the SOS framework as individual psychology and behaviour, socio‐cultural settings, environmental, home settings, policy change intervention ( Chastin 2016 ).

• Outcome information: reporting both primary and secondary outcomes.

• Additional information: missing data; conflicts of interest; intervention fidelity (whether the intervention was delivered as planned); information on the nature and extent of any additional actions given as part of the intervention (co‐interventions); intervention costs; source of study funding; adverse events.

Two review authors (from a pool of nine) independently extracted data from each study. Any disagreements regarding data extraction were resolved by discussion or by consulting a third review author if necessary. Review authors were not allowed to extract data from studies on which they were a co‐author.

Assessment of risk of bias in included studies

Two review authors (from a pool of nine) independently assessed risk of bias of the included studies using the Cochrane risk of bias tool ( Higgins 2021 ). Any disagreements in risk of bias assessment were resolved by discussion or through consensus with a third review author. Review authors were not allowed to assess the risk of bias of any studies on which they were a co‐author. The risk of bias tool assesses the following five domains:

• selection bias (sequence generation and allocation concealment);

• performance and detection bias (blinding);

• attrition bias (incomplete outcome data, withdrawals, dropouts, protocol deviations);

• reporting bias;

• and an open 'other bias' category (e.g. baseline comparability for age, gender, and occupation).

We graded each domain as being at 'low', 'high', or 'unclear' risk of bias and provided a justification for our judgement in the risk of bias tables. We considered blinding of outcome assessment differently when sedentary time was assessed objectively or by self‐report, as participants cannot be blinded to self‐reported measures ( Shrestha 2014 ). For cluster‐RCTs, we considered:

• recruitment bias;

• baseline imbalance;

• loss of clusters;

• incorrect analysis; and

• comparability with individually randomised trials.

We summarised risk of bias for each outcome as being as 'low' if none of studies contributing to the outcome had any domains assessed as at high risk of bias; unclear if fewer than the majority studies contributing to the outcome had more than three domains at high risk of bias, or 'high' ( CPHG 2011 ).

Measures of treatment effect

To calculate treatment effects, we entered outcome data from all included studies into Review Manager 5 ( Review Manager 2020 ). For studies with continuous outcome measures, we reported mean scores and standard deviations. We used the mean difference (MD) between intervention and control groups postintervention for each continuous outcome to analyse the size of the effects of interventions. We used the adjusted MD between groups for cluster‐RCTs.

Unit of analysis issues

We did not identify any studies with multiple intervention arms. For cluster‐RCTs, we assessed whether the reported results were properly adjusted to account for clustering effects.

We adjusted outcomes for cluster‐RCTs for clustering effect using the effective sample sizes method ( Higgins 2021 ). We used an intraclass correlation coefficient of 0.07 based on previous research in activity behaviour research ( Kerr 2018 ).

Dealing with missing data

We noted missing data on the data extraction form and reported this in the risk of bias table. Where we encountered missing numerical outcome data, we contacted study authors to obtain this information. Where outcome data such as standard deviations were missing and were not obtainable from the study authors, we derived them from other available statistics following the methods described in the Cochrane Handbook for Systematic Reviews of Interventions ( Higgins 2021 ).

Assessment of heterogeneity

We visually inspected the forest plots to assess statistical heterogeneity. In addition, we analysed heterogeneity using the I ² and Chi ² statistics and corresponding P value. We reported heterogeneity as follows:

• low degree of heterogeneity (25% to 50%);

• moderate degree of heterogeneity (50% to 75%);

• high degree of heterogeneity (75% or higher).

We considered heterogeneity in design, intervention, participants, and outcomes, as recorded in the Characteristics of included studies table.

Assessment of reporting biases

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

Data synthesis

The findings of the studies were systematically examined and integrated across studies. The included studies were tabulated and grouped by study design, population, outcomes, and setting, to explore relationships within and between included studies in a narrative summary. We conducted meta‐analyses using Review Manager 5 ( Review Manager 2020 ) employing a random‐effects model, which allows for a greater level of natural heterogeneity between studies ( Deeks 2011 ). We included data from cluster‐RCTs in meta‐analyses, as clustering was already taken into account in the analysis reported in included studies. We compared the effect of interventions aimed at reducing sedentary behaviour against control. When the control condition was a physical activity condition, we treated it the same as usual care or any intervention not aimed at changing sedentary behaviour that was used as control. Our rationale for this was that evidence shows that physical activity interventions do not change sedentary behaviour in older adults ( Gomersall 2015 ; Martin 2015 ). For some outcomes (primary outcome of breaks in sedentary time, secondary outcomes of depression, health status, and quality of life), it was not possible to conduct a meta‐analysis because insufficient data were available. We reported results for these outcomes narratively grouped by outcome.

Subgroup analysis and investigation of heterogeneity

We conducted a subgroup analysis based on the method of measurement of sedentary time as either objectively measured using body‐worn devices or measured using self‐report tools. We also conducted subgroup analysis according to the intervention duration, classified as interventions delivered during a single point of contact or interventions delivered longitudinally. This classification differs from that in our protocol, as the classification in our protocol did not fit the type of interventions we found. We planned further subgroup analysis based on sex, geographical location, intervention setting or 'deliverer', socioeconomic status, and health status, but due to the small number of included studies this was not possible.

Sensitivity analysis

We planned to conduct a sensitivity analysis to understand the impact of risk of bias on the findings by excluding from the analysis studies rated as at high risk of bias. However, overall risk of bias was uniform across studies, and we did not rank any studies as at high risk of bias.

Summary of findings and assessment of the certainty of the evidence

We summarised our findings according to the guidelines in Chapter 14 of the Cochrane Handbook for Systematic Reviews of Interventions ( Schünemann 2021 ). Two review authors (from a pool of nine) independently assessed the certainty of evidence for each of the primary outcome measures using the GRADE methodology ( Guyatt 2011 ). Results were tabulated using a template adapted from GRADEpro GDT ( GRADEpro GDT ). The certainty of a body of evidence as assessed by GRADE is the extent to which one can be confident in the estimate of effect. We assessed the certainty of the evidence for each outcome as high, moderate, low, or very low. In the GRADE approach, RCTs are assessed as high certainty at the start. Five criteria are considered for possible downgrading of the certainty of evidence: study quality (risk of bias); consistency (consistency between studies); directness (the same study participants, intervention, and outcome measures in included studies is for the people, measures, and outcomes we wanted to study); precision of results; and reporting biases. Three criteria are considered for possible upgrading of results: strong or very strong associations between intervention and outcome; large or very large dose‐response effects; and where all plausible confounders would have reduced the effect. We reported the certainty of evidence in summary of findings Table 1 alongside the synthesis of outcomes.