How effective and how expensive are interventions to reduce sedentary behavior? An umbrella review and meta‐analysis

A reduction in sedentary behavior (SB) can contribute to the prevention of chronic diseases. This is the first umbrella review that summarizes the effectiveness and monetary costs of different types of interventions to reduce SB across all age groups and populations in different settings. We comprehensively searched seven databases for systematic reviews and meta‐analyses and conducted an umbrella review of the effects of interventions to reduce SB. Additionally, we performed a meta‐analysis of primary studies included in the umbrella review. Furthermore, we analyzed health economic aspects of interventions to reduce SB. We included 40 systematic reviews in our umbrella review, with 136 primary studies suitable for further meta‐analyses. We found that interventions targeting the physical environment reduce SB most effectively in the majority of populations and settings. Workplace interventions reduced SB by −89.83 min/day (95% CI −124.58 to −55.09; p ≤ 0.0001). Twenty‐two of 169 primary studies (13.0%) contained health economic information. The intervention costs per participant ranged from €0 to €3587. Our findings demonstrate that physical environment interventions most effectively reduce SB in a majority of populations and settings. Health economic information was reported in few studies and was mostly restricted to acquisition costs.


| INTRODUCTION
Sedentary behavior is defined as "any waking behaviour characterized by an energy expenditure ≤1.5 METs while in a sitting or reclining posture." 1 Sedentary behavior has increased worldwide in the past decades. 2,3 The estimated prevalence of sedentary behavior is 6.4 h per day in adults, about 9.3 h in teenagers, and 4.4 h in children, and it differs by age, sex, socioeconomic status, and other factors. 4 Current research suggests that sedentary behavior increases the risk of developing chronic conditions such as coronary heart disease, type 2 diabetes, [5][6][7][8] and several types of cancer. 9 In addition, large amounts of time spent sedentary increase the risk of obesity and overweight, particularly during childhood. 10 Furthermore, prolonged sitting increases the risk of premature mortality. 11 Recent meta-analyses of sedentary behavior interventions show improvements in cardiovascular risk factors (e.g., systolic blood pressure, plasma glucose and insulin level) [12][13][14] and a positive impact on body anthropometry (e.g., body mass, waist circumference, percentage of body fat). 14 Some of these factors also play a role in the development of cancer. [15][16][17] Globally, chronic diseases are the principal causes of mortality, 18 contributing considerably to the overall burden of disease 19 and increasing the direct and indirect costs of health care systems. 20,21 From a public health perspective, reducing sedentary behavior as a risk factor can-at least partly-contribute to the prevention of several chronic diseases and reduce health care costs.
The latest WHO Guidelines on physical activity and sedentary behavior 22 recommend decreasing sedentary behavior for all age groups (children, adolescents, adults, older adults) and for people with chronic conditions, pregnant and postpartum women, and people living with disability.
A large number of trials have investigated the effectiveness of interventions to reduce sedentary behavior. Findings have been summarized by several systematic reviews and meta-analyses and they suggest that sedentary behavior interventions are effective for reducing sedentary behavior in specific populations (e.g., children and teenagers, office workers, the elderly, people with overweight, and obesity) and settings (e.g., the workplace, schools, during leisure time). [23][24][25][26][27] Furthermore, two umbrella reviews summarized findings from existing systematic reviews and meta-analyses. 28,29 However, those umbrella reviews were limited to certain age groups and settings, with one umbrella review 28 focusing on sitting and screen time in children and adolescents and the other 29

| Study inclusion and exclusion criteria
The main outcome of the umbrella review was sedentary behavior. 1 All systematic reviews that included primary studies with this outcome were eligible. We included all systematic reviews that contained intervention studies (e.g., randomized control trials [RCT], cluster randomized studies, cross-over studies, or quasi-randomized studies as well as single arm intervention studies with before/after comparisons as primary studies) that analyzed the effect of interventions on sedentary behavior. All age groups were included. Studies that included individuals with chronic diseases (including for example overweight, obesity, and cancer) were included. We considered no intervention or alternative interventions as control groups. Both objective measurement methods such as objective devices (e.g., accelerometer and video) and subjective measurement methods (e.g., questionnaire, logbook, and parent reports) of sedentary behavior were eligible. No language or date restriction were applied. We excluded systematic reviews that contained only observational studies as well as primary studies that only measured screen time as a sedentary behavior outcome. Health economic evaluation studies were eligible if they contained health economic information (e.g., acquisition costs, costs of implementation, evaluation costs, and cost-effectiveness) or referenced health economic publications.

| Study selection
Two researchers (KL, CJ) independently screened titles and abstracts with no date or language restrictions, read the full texts of potentially eligible systematic reviews, and decided independently if the systematic reviews should be included in the umbrella review. Disagreements were resolved by discussion. The study selection of the systematic reviews was documented by KL according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. 30 We removed duplicates manually. Table S1 shows the reasons for exclusions of systematic reviews and meta-analyses. In addition, we searched the reference lists of included systematic reviews to identify potentially eligible primary studies for the meta-analyses. Furthermore, we searched the gray literature (non-peer-reviewed articles) such as dissertations and conference presentations for potentially eligible studies mentioned in the reference lists of the included systematic reviews.
In addition, potential eligible primary studies were independently screened by AN and JBS for health economic information.

| Data extraction
CW and HB independently extracted the following information from systematic reviews and meta-analyses: author's name, publication year, target population, assessment of sitting time, type of intervention, type of control, and main findings. KL and CJ independently extracted the following information from eligible primary studies of the selected systematic reviews and meta-analyses: title of the manuscript, author's name, publication year, study name, study design (e.g., RCT and cluster RCT), target population (e.g., office workers, older adults, children), sex and average age of participants, number of participants included in the analysis, country, region, ethnicity, intervention focus (e.g., sedentary behavior, physical activity, nutrition),  We meta-analyzed the results of primary studies if the standard error (SE) was given or could be calculated from the standard deviation (SD) (SE = SD/ √ n; n = number of participants), the 95% confidence interval (CI), or the 25th/75th percentile (assuming a normal distribution). We applied the restricted maximum likelihood method (REML) 31 to estimate the between-study variance in order to assess heterogeneity across types of interventions, study designs, and outcome definitions. Furthermore, we determined study heterogeneity using the Q statistic and its derivatives. 32 In the primary studies, sedentary behavior was measured as change in sedentary behavior (within-group change in sedentary behavior) or sedentary behavior preintervention/and postintervention. Assuming a normal distribution, we approximated change in sedentary behavior by the differences of the averages of sedentary behavior preintervention/postintervention and performed two meta-analyses with the two distinct outcomes. The REM meta-analyses only included data measured as minutes per day or data that could be converted to minutes per day to ensure comparability and to permit meta-analyses of the data across primary studies.
We applied REM meta-analyses if at least 10 primary studies were available for a specific combination of target population and setting to obtain meaningful results.

| Subgroup analyses
We performed subgroup analyses according to the main target of the intervention, that is, interventions aiming to reduce sedentary behavior of participants, interventions aiming to increase physical activity of participants, and interventions targeting both outcomes.
Furthermore, we performed a subgroup analysis that compared the effect on sedentary behavior reduction <6 months and ≥6 months after the intervention. In addition, we performed a subgroup analysis that compared objective and subjective assessments of sedentary behavior.
If studies reported multiple settings of sedentary behavior for the same participant, we included the total sedentary behavior measure in the subgroup analyses.

| Publication bias
We investigated potential publication bias using funnel plots, 33

| Sensitivity analyses
We performed sensitivity analyses using fixed effect models (FEM) and REMs with alternative between-study heterogeneity estimators (e.g., DerSimonian and Laird [DL] 35 or Sidik & Jonkman 36 ). In addition, we conducted outlier and influence diagnostics. 37

| Methodical quality assessment of systematic reviews and meta-analyses
Two researchers (NM, CW) evaluated in duplicate the methodical quality of the included systematic reviews and meta-analyses with the A MeaSurement Tool to Assess systematic Reviews (AMSTAR-2). 40 This tool was specifically developed for the assessment of the methodological quality of systematic reviews of randomized and nonrandomized studies. 40 It contains 16 items such as risk of bias assessment or availability of a study protocol. The results of the AMSTAR-2 evaluation are summarized in Table S2.

| Health economic consideration
If total costs were not mentioned, they were calculated from the product of resource use and price per unit. If year of intervention was not specified in the publication, we searched for the clinical trials registration number to determine the year of first participant enrolment.
Publication year was approximated if year of intervention or year of cost determination was not specified. Costs incurred in the control group were also extracted. For comparability, the cost data were currency-converted using purchasing power parities for gross domestic product given by OECD. Stat 41 and adjusted for inflation to 2020 using the harmonized consumer price index 42 -following the procedure described by Scholz et al. 43 If more than one cost component of an intervention was described, the total sum was calculated over the individual components. For further comparability, the total costs were allocated to the number of participants.

| Literature search results
In total, 6583 records were obtained from the bibliographic databases.
After removing all duplicates (N = 1900), we screened 4683 potential studies for titles and abstracts and excluded 4365 studies due to their irrelevance for the investigation. We read the full texts and evaluated the eligibility of the remaining 318 studies. We excluded 278 studies and categorized those by exclusion reason. The remaining 40 systematic reviews were deemed eligible for the umbrella review. After excluding overlapping primary studies, the included systematic reviews contained a total of 169 eligible primary studies. Both the characteristics of the 40 systematic reviews and meta-analyses and of the 169 primary studies are described in the narrative synthesis.
Of the 169 primary studies, 33 articles could not be included in the meta-analyses because information on post-intervention sedentary behavior was missing or the standard error was not available or could not be calculated. The results of remaining 136 primary studies were included in the meta-analysis. Twenty-two of the 169 eligible primary studies (13.0%) contained health economic information. In two cases, separate health economic publications were referenced, so these were used for further analysis instead of the primary study. 44,45 One study provided a detailed list of intervention costs upon request. 44 Figure 1 documents the literature search according to the PRISMA statement. 30

| Description of included systematic reviews, meta-analyses and individual studies
Of the 40 systematic reviews and meta-analyses, 20 studies reported quantitatively pooled data from meta-analyses. In total, 15 systematic reviews and meta-analyses focused on children and/or adolescents, F I G U R E 1 PRISMA flow diagram on database search and identification of systematic reviews and primary studies. Abbreviations: n = number; SR = systematic review 1.2%). Obesity/overweight was an inclusion criterion in 30 (17.8%) primary studies. Twenty studies (11.8%) examined patients with diverse chronic diseases (e.g., cancer, stroke, and rheumatoid arthritis).
None of these studies included people who were unable to stand (such as wheelchair users). Table S4 provides   In older adults, studies that assessed the effectiveness of physical environment interventions were not available for meta-analyses. In that target group, total sedentary behavior ( Figure S4

| Sensitivity analyses
In a sensitivity analysis, we excluded one primary study 48 due to its wide 95% confidence interval (95% CI À1893.52 to 1803.52), which was 241 times larger than the variance of the pooled results of all F I G U R E 4 Meta-analysis of the effectiveness of different types of interventions on total sedentary behavior among people with overweight/ obesity. Abbreviations: CI = confidence interval; RE model = random effects model, SB difference = preintervention/postintervention difference in sedentary behavior interventions (95% CI À 34.85 to À19.52). Overall results did not substantially change (compare Figure S10 with Figure S1).
Influence and outlier diagnostics 37 identified two primary studies 49,50 as potential outliers. After exclusion of both primary studies, the order of intervention category effectiveness did not change.

| Health economic consideration
In most cases, primary studies with health economic information  Due to the heterogeneity of the interventions implemented, even within intervention categories, a comparison of studies was only meaningful for physical environment and personal behavior interventions. Studies with physical environment interventions were mostly classified as the most effective intervention category, with statistically significant results. Depending on the desk used (sit-stand workstation, electrically or manually adjustable, treadmill desk), the intervention costs amounted to €334 to €3587 per participant, whereby the cost of a treadmill desk was significantly higher than the cost of a sit-stand workstation. Personal behavior interventions included in the health economic analysis and classified in the meta-analysis as the second most effective intervention category used booklets and manuals. One study used further intervention materials like pedometers and a resistance band. In these studies, the costs amounted to €5 to €57 per participant.
As described above, two studies referenced separate health economic publications, 44,45 one study using a cost and cost-benefit analysis, and the second a cost-effectiveness analysis. Munir et al. 44 measured the increase in productivity in monetary units. After subtracting the cost of the intervention and costs for lost work time due to the implementation of the intervention, they reported net cost savings of £1770 (currency-converted and inflation-adjusted: €2039).
The incremental cost-effectiveness ratio (ICER) was between £8 and £17 (€10 and €19) per minute per workday. Sevick et al. 45 Table S2 shows the results of the AMSTAR-2 evaluation. The methodologic quality of 10 of the 40 included systematic reviews was rated critically low according to the AMSTAR-2 criteria. The methodical quality of 11 systematic reviews was rated low and the quality of 16 reviews was rated moderate. The AMSTAR-2 scores of three systematic reviews were rated high. Very few systematic reviews Activity-permissive workstations have been shown to reduce sedentary behavior in the workplace. 25 A reduction of workplace sedentary behavior results in a sizeable reduction of total sedentary behavior since up to 48.5% of total weekly sedentary behavior can be attributed to the workplace. 54 Interestingly, our umbrella review shows that multicomponent interventions do not represent the most effective intervention category. This contrasts with ecological models of health behavior that imply that multicomponent interventions reduce sedentary behavior more effectively that single level interventions. 55 Our meta-analysis shows that physical environment interventions are more effective than personal behavior and social environment interventions. Based on ecological models of health and behavior, we theorize that multicomponent interventions that include a physical environment intervention are more effective than physical environment interventions. However, we defined multicomponent interventions as any combination of intervention categories. To investigate our theory, additional analysis that further differentiate multicomponent interventions are required. and social environment (parents, teachers, carers). 56,57 The observation that physical and social environment interventions are the most effective intervention for school children suggests that health behavioral change for children works better with strategies targeting unconscious processes. Based on our findings, we assume that interventions that work with conscious decision-making processes like personal behavior interventions and multicomponent interventions, which often include personal behavior elements, might be more effective for adults than for children.

| QUALITY ASSESSMENT
The pooled results of our analyses suggest that personal behavior interventions reduce total sedentary behavior in people with overweight/obesity more effectively than multicomponent and social environment interventions. This finding is in contrast to our findings from other population groups like adults, where multicomponent interventions were more effective than personal behavior interventions.
According to qualitative research, 58  While multicomponent interventions provide similar information and should be more effective than single level personal behavior interventions according to the ecological models of health and behavior, 55 we observed that personal behavior intervention were more effective than multicomponent interventions for people with overweight/obesity. One possible explanation is that multicomponent interventions often include a social environment intervention element, which may reduce the overall effectiveness of the intervention. Reporting on intervention costs was very heterogeneous. While in two studies reference was made to a separate health economic publication with a detailed list of all costs incurred and a cost-benefit or cost-effectiveness analysis, in the majority of cases only acquisition costs were reported. In this respect, it is questionable whether all relevant components of an intervention have been monetized.
Against this background, only the direct investment costs could be considered in the current review. An overview of the total intervention costs (including implementation and evaluation costs) for an overall assessment of the cost-benefit ratio was therefore not possible.
Although the cost data were currency-converted and adjusted for inflation to improve comparability, the costs of interventions are not readily applicable to real world settings. In one case, 45 the year of publication was unknown which could have produced a slight overestimation of costs. However, this had only a marginal effect because the end of the intervention and the publication date were close in time. Furthermore, it should be noted that in most cases, the costs were reported from the perspective of the party who has to pay the acquisition costs, for example, the employer or the school.

| Strengths and limitations
To our knowledge, the current umbrella review is the first comprehen- behavior and settings such as the workplace, schools, and leisure time.
A further strength of our study is the a priori published protocol (PROSPERO registration number: CRD42020197529) that followed the PRIMSA-P statement. We determined the search strategy with inclusion and exclusion criteria and the analytic plans before conducting the umbrella review to ensure a high quality of our umbrella review. During our literature search, we included all unpublished gray literature deemed eligible. We ensured the high quality of our umbrella review by assessing the methodical quality of the included systematic reviews with the AMSTAR-2 criteria. Additionally, we investigated the heterogeneity of our results with the I 2 statistic. We investigated potential publication bias of the included primary studies via funnel plots, rank correlation analyses and linear regression analyses.
However, our umbrella review has some limitations. In the absence of reported individual changes in sedentary behavior, we determined sedentary behavior change by quantifying the difference between preintervention/and postintervention measurements. Therefore, the results of the analyses of the calculated sedentary behavior change should be interpreted with caution. Furthermore, we identified some asymmetry in the included primary studies, indicating possible publication bias. 70 In addition, the large number of relatively heterogeneous primary studies may weaken the overall value of the results.
Furthermore, the quality of our umbrella review was influenced by the methodological quality of the included systematic reviews. Very few systematic reviews included a list of excluded studies and justifications for their exclusion or listed the sources of funding of the included primary studies. A consideration of the risk of bias when interpreting the results is critically important for the methodical quality of a systematic review. 71 All reviews that failed to account for risk of bias when discussing its possible influence on their results were ranked low or critically low. For almost half of the reviews, a prior published protocol was not available. Due to these features of some systematic reviews, some of the included systematic reviews have a low methodical quality.
In addition, the heterogeneity of our key findings was considerable. This may be due to the fact that the primary studies used different target groups, settings and methods of measuring sedentary behavior (objective and subjective) and interventions with varying lengths of time.

| Conclusion
In conclusion, our umbrella review shows that physical environment interventions constitute the most effective means to reduce sedentary behavior in most populations and settings. Certain populations (older adults, teenagers, preschoolers, and toddlers) as well as certain settings (transportation, day-care facilities, and preschools) are less researched and should be addressed in future research. Additionally, our umbrella review presents an insight into the costs of measures to reduce sedentary behavior.

ACKNOWLEDGMENTS
We would like to thank the German federal ministry of education and research for its funding of this project. Open Access funding enabled and organized by Projekt DEAL.

CONFLICT OF INTEREST
No conflict of interest statement.