Effects of Interrupting Prolonged Sitting with Physical Activity Breaks on Blood Glucose, Insulin and Triacylglycerol Measures: A Systematic Review and Meta-analysis

Increasing physical activity (PA) [1] and both decreasing and interrupting “sedentary behaviour” are emphasised in public health guidelines [2]. “Sedentary behaviour” (SB) is any seated or reclining behaviour, whilst awake, with energy expenditure (EE) at or below 1.5 metabolic equivalents (METs) [3, 4], such as sitting in the office. The UK Department of Health [2] recommends breaking up long periods of sitting during working hours and interrupting sedentary time. Australia’s Department of Health [5] recommends interrupting long sitting periods, although no quantitative threshold is specified.

A systematic review and meta-analysis of cross-sectional observational and laboratory-based experimental studies on the effects of breaks in SB [6] concluded that walking-based light-intensity physical activity (LIPA) and moderate intensity physical activity (MPA) breaks resulted in significant reductions in post-prandial glucose and insulin. Physical activity (PA) breaks in sitting were also more effective than one continuous bout of exercise on glucose. Nonetheless because this review only included five studies on glucose, published between 2011 and 2014, some relevant earlier studies [7‐12] and more recent studies [13‐22] might have been omitted or missed. There was no date restriction in Benatti et al. [23] but no meta-analysis was performed. Therefore, the magnitude and moderators of PA breaks on metabolic variables compared to sitting were not quantitatively assessed. It also remains to be established if PA breaks influence metabolic markers in a different way to structured continuous exercise, and thus confer a different benefit to structured continuous exercise. Recently, the United States of America Physical Activity Guidelines Advisory Committee in its Scientific Report to the Secretary of Health and Human Services stated a need for randomised controlled trials to test the effects of interventions to replace time spent in SB with PA [24]. Therefore, an updated meta-analysis of such existing trials, in adults, whether healthy or with type 2 diabetes, that can be used as part of the development of public health guidelines, is apposite.

Accordingly, there is scope for a new systematic review and meta-analysis of the experimental literature on the metabolic effects of interruptions of prolonged sitting with PA breaks, as an important contributor to the evidence pool used to develop, update, and refine public health guidance.

The primary aim was to systematically review and meta-analyse that studied the effects of controlled trials breaking up prolonged sitting with PA breaks throughout the day compared with prolonged sitting on glucose, insulin and TAG. A secondary aim was to systematically review and meta-analyse controlled trials that compared the effects of PA breaks against continuous exercise on glucose, insulin and TAG.

Firstly, a systematic database search of PubMed, OvidSP, Journals@Ovid and PsycINFO, Science Direct, and SPORTDiscus, was conducted on 04/03/2017. The search was subsequently updated on 03/07/2018. Search terms were collated into four broad categories, based on the PICOT (population, intervention, comparison, outcome, time) format [26, 27]: setting (“sedentary behaviour”), intervention (“physical activity”), intervention type/comparison (“breaks”), outcomes (“glucose”) [28]. Full search details terms for all databases searched are provided in Electronic Supplementary Material Appendix S1.

Additionally, a hand search of the reference lists of articles included in the final analysis that were identified via the database search was conducted, as were the first 20 “related articles”, via the “related articles” link on PubMed, of those included database search articles. A hand search of other reviews, commentaries, letters, PhD dissertations, and reference lists of original articles was also conducted.

Studies were then selected according to the following inclusion and exclusion criteria. Studies were included if they fulfilled all of the following criteria, with PICOT categories in parentheses where appropriate:

Studies were excluded if they met any of the following criteria:

Studies were independently assessed for inclusion by two reviewers, RL, DF, with disagreements resolved via discussion. The reviewers, RL, DF, were not blinded to authors, institutions or journals of publication. If a decision on whether to include or exclude a paper could not be made from the title and abstract, the full text was obtained and checked. The flow diagram for the search process is presented in Fig. 1. A complete list of excluded studies, with reasons for exclusion, is available upon request.

Data from included studies were extracted (by RL) for first author name, publication date, participant characteristics, full description of the PA and sitting intervention protocol and outcomes. Outcomes extracted for the narrative review were measures of glucose, insulin, triacyglycerol, c-peptide, non-esterified fatty acids (NEFA), cholesterol, lipoproteins from blood whether plasma, serum or whole, and blood pressure.

The Cochrane Collaboration’s risk of bias (RoB) tool [32] was used to aid in assessing the RoB in individual studies. Components were assessed independently, with no overall composite score assigned, as per PRISMA [25, 26] and Cochrane collaboration [32, 33] recommendations. Washout period for crossover studies was used for the “other” sources of bias component. Each component rated was as “high risk” or “low risk”. If details for a particular domain were insufficient, the risk of bias was assessed as “unclear”. Assessments were performed independently by two authors (RL, DF) with disagreements resolved by discussion, and then arbitration (HJM) if necessary.

A narrative overview provided in text and tables summarises study characteristics. The narrative synthesis includes studies in which PA break or sitting protocols were not strictly controlled to provide a broader summary of the literature, whereas only controlled laboratory studies were statistically meta-analysed.

C-peptide, blood pressure, NEFA, cholesterol and lipoprotein outcomes were not meta-analysed because few studies had these variables as outcomes. Studies with glucose, insulin and TAG measures were meta-analysed. Interstitial glucose data via continuous glucose monitoring system (CGMS), if available, were extracted for the meta-analysis as a first preference over post-prandial measures of venous or capillary blood glucose, as continuous glucose data, as opposed to the snapshot nature of venous or capillary blood draw, provides a more comprehensive view of glucose responses, that is not dependent on the blood draw schedule. Incremental area under the curve (iAUC) for glucose, insulin, TAG was meta-analysed in preference to total area under the curve (tAUC), as iAUC is the recommended measure for detecting differences in post-prandial responses [34‐36]. Data from prior studies that were reanalysed, combined for reanalysis, and reported in a later study were not extracted. If a later publication reported a new measure of, for example glucose, obtained from the same experimental conditions as a prior publication, CGMS glucose was used as the first preference, if available. If this was not available, post-prandial iAUC was used, followed by tAUC.

Means, standard deviations or standard errors or 95% confidence intervals (CI) were extracted from individual studies and used to calculate standardised mean differences (SMD) using DerSimonian and Laird random-effects models [33]. Continuous outcomes were analysed using SMD to account for different measurement scales [37], tAUC or iAUC over different time scales. When multiple exercise conditions were used in a study, data for all relevant conditions were synthesised and reported separately in the appropriate meta-analysis.

If a study contained more than 2 trial arms, and a control comparison condition was used twice in the meta-analysis, the sample size for the control condition was divided by the number of times the control condition was used [33]. If means were not reported, and medians were reported instead, the study was not meta-analysed. Pooled continuous data were expressed as SMD with 95% CI. SMDs were interpreted according to Cohen [38]: 0.2 represents a small effect, 0.5 a moderate effect, and 0.8 a large effect.

When required outcome data for glucose, insulin and TAG were not available in the full text, but data were presented graphically, an attempt was made to digitise the graph. If this was not possible, the original authors were contacted. If data still could not be obtained successfully, the affected study was omitted from the meta-analysis, and the results summarised in the narrative review.

Statistical heterogeneity was tested with the Chi-square test (p < 0.05) and I² statistic (0–40%: might not be important; 30–60%: may represent moderate heterogeneity; 50–90%: may represent substantial heterogeneity; 75–100%: considerable heterogeneity) [33].

Subgroup analysis for TAG was pre-specified [33] according to whether the experimental condition was performed on 1 day, or over multiple days, as there is considerable evidence that the effects of exercise on TAG peak approximately 18 h post-exercise [39, 40]. Usual PA, body mass index, cardio-respiratory fitness (CRF) or insulin resistance status of participants was selected as another subgroup characteristic, given that metabolic responses to exercise might be affected by CRF or insulin sensitivity status [41‐44], with one subgroup consisting of studies that assessed participants who were physically inactive, or sedentary, or were overweight/obese or had type 2 diabetes or impaired fasting glucose, and the other subgroup containing physically active participants. “Physically active” was defined as either exceeding the recommended 150 min of moderate to vigorous physical activity (MVPA) per week, or reported as “recreationally active” [7, 8, 45]. “Sedentary” was defined as not working in a non-sedentary job [46], exceeding 5 h of sitting time per day [15, 22, 47], or any study that defined participants as sedentary. If a study did not report the PA, body mass or health status of participants, it was omitted from the subgroup analysis. Subgroup analysis was also performed for sex, as sex might affect metabolic responses to exercise, feeding, and metabolic health [48‐51], possibly due to the effects of sex hormones such as oestrogen [52]. As EE of exercise might affect results, subgroup analysis was also performed to determine whether EE between conditions was matched when comparing PA breaks with continuous exercise.

Meta-regression was only performed, to explore the possible effects of any explanatory variable on differences in post-prandial glucose, if at least ten studies were included in the meta-analysis, as there should be at least ten studies in a meta-regression for each explanatory variable modelled [33]. If there were sufficient studies, a random-effects model was used to assess whether body mass index (BMI) moderates the effect, as evaluated by SMD, of PA breaks compared with sitting, and of PA breaks vs continuous exercise.

Funnel plots, Begg and Mazumdar’s rank correlation test [53], Egger’s regression test [54] and Rosenthal’s fail-safe N [55] were used to assess publication bias if more than ten studies were included in the meta-analysis [26, 56]. The trim and fill method, with L0 as the estimator [57], was used to estimate “missing” studies, if any, in the funnel plots. The method of Vevea and Woods [58] was used to calculate the modified SMD in the event of severe 2-tailed selection bias.

Graphical representations of potential bias within and across studies are presented using Review Manager 5.3 (RevMan5.3) (Cochrane Collaboration, Copenhagen, Denmark). All statistical calculations for summary measures were analysed in RevMan 5.3 and presented as SMD and 95% CI. Meta-regression and publication bias analyses were performed in R (The R Project for Statistical Computing). Statistical adjustment of SMD for publication bias was performed in SPSS version 23 (IBM Corporation, Armonk, NY, USA) and R, using the macros developed by Field and Gillett [59].

The initial database search was performed on 04/03/2017. Subsequently, the search was updated on 03/07/2018. There were 897 studies in the initial search results after removal of duplicates. 28 studies met the inclusion criteria. In the updated search results, there were 174 studies after removal of duplicates, of which 14 met the inclusion criteria. Therefore, a total of 42 studies were included in the final systematic review, of which 37 were included in the meta-analysis. The results of the systematic search are presented in Fig. 1.

Other than a few studies [15, 17, 21, 46, 47] most did not utilise or report any form of blinding. All studies included in the meta-analysis, except one [77], were randomised, but only a few reported the randomisation methods clearly [17, 19, 21, 22, 46, 47, 62]. Additionally, with the exception of a few studies [15, 17, 19, 46, 47, 62], most did not report how any possible missing data were handled. Notably, studies with the most rigorous design or reporting [15, 17, 46, 47] appeared to report larger effects, for example, on glucose and insulin (Figs. 4, 7) (Table 5).

Several short-term experimental studies have shown that PA breaks attenuate post-prandial increases in glucose (Fig. 4) and insulin (Fig. 7) on the same day, compared to no-exercise sitting. Additionally, these effects persisted overnight [14, 15]. The sustained effects of PA breaks warrant further research, especially with the increasing use and availability of CGMS. The effects of breaks on TAG were weaker, but PA breaks still appear to attenuate TAG somewhat (Fig. 2). Physically inactive or sedentary participants or those with IFG or T2D experienced greater benefits in glycaemic attenuation (Fig. 5), as did those with higher BMI, as revealed by meta-regression (Sect. 3.3.2) (Fig. 13).

To place these results in the wider context of the effects of exercise on markers of metabolic health, in a meta-analysis of non-laboratory based randomised controlled trials of PA interventions lasting from 2 to 6 months in people with type 2 diabetes aged 35–71 years, walking, yoga, tai chi and qigong had a cumulative SMD of − 0.60 (95% CI − 0.83, − 0.37) compared with no exercise on glycaemic control, as indicated by glycated haemoglobin (HbA1c) [88]. High intensity interval training (HIIT) interventions lasting more than 2 weeks, compared to no exercise, reduced insulin resistance by an SMD of − 0.49 (95% CI − 0.87, − 0.12) in all groups, by − 0.38 (95% CI − 1.39, 0.63) in overweight/obese, and by − 0.62 (95% CI − 1.10, − 0.14) in people with type 2 diabetes [89]. Similarly, short term HIIT, lasting less than 12 weeks, reduced fasting glucose by an SMD of − 0.35 (95% CI − 0.62, 0.0.09) in overweight or obese people [90]. Additionally, in people with non-alcoholic fatty liver disease, exercise interventions, whether aerobic, resistance, or combined, lasting more than 1 month, reduced the glucose parameters HbA1c and homeostatic model of assessment of insulin resistance (HOMA-IR) by SMDs of − 0.76 (95% CI − 0.78, − 0.42) and − 0.50 (95% CI − 0.85, − 0.15), respectively [91], compared to normal care. Similarly, exercise reduced postprandial total TAG by Cohen’s d of − 0.60 (95% CI − 0.69, − 0.50), and iAUC TAG by − 0.59 (95% CI − 0.76, − 0.42) in all participants [92]. Since all but one of these meta-analyses [92] were not laboratory based and evaluated acute or longer-term protocols and adherence to the exercise protocols was less easy to confirm, they should be compared with our findings only generally and cautiously. Similarly, a previous meta-analysis [6] of 5 studies [46, 60, 62, 68, 83] reported PA breaks resulted in lower glucose measures than sitting. The effect sizes reported in our current meta-analysis, whether for measures of glucose, insulin, or TAG, can be seen to be generally similar to the effect sizes of diverse exercise modalities in various populations reported in the literature.

Recently, a meta-analysis [93] reported that activity breaks compared to sitting lowered post-prandial glucose by Cohen’s d of − 0.36 (95% CI − 0.50, − 0.21), and postprandial insulin by Cohen’s d of − 0.37 (95% CI − 0.53, − 0.20). The mean postprandial TAG response with breaks was reduced by 0.06 (95% CI − 0.15, 0.26) compared with sitting. The findings of the meta-analysis by Saunders et al. [93] for glucose and insulin outcomes were broadly similar to our findings, but with smaller effect sizes. However, we found that PA breaks lowered post-prandial TAG outcomes, in contrast to Saunders et al. [93]. The differences in results could be explained by differences in inclusion criteria. Saunders et al. included adolescents and teenagers [94‐98] in their meta-analysis [93], whereas we did not. Furthermore, whereas we included studies with people with type 2 diabetes in our meta-analysis, Saunders et al. [93] did not. It is possible that we found that PA breaks compared to sitting had greater benefits on glucose, insulin and TAG outcomes than Saunders et al. [93] because participants in their meta-analysis were healthier and younger. This is supported generally by our meta-regression and subgroup analyses, which suggested that people with higher BMI, lower cardiovascular fitness, impaired fasting glucose or type 2 diabetes, experienced greater reductions in post-prandial glucose and insulin, compared to those with lower BMI or who were healthier. However, Saunders et al. [93] reported that neither glucose nor insulin outcomes were associated with BMI. This discrepancy between their findings and ours might again be due to the younger and healthier participants in their analyses, as transport, uptake and metabolism of glucose might be greater in the insulin sensitive compared to the insulin resistant [44]. Additionally, whereas Saunders et al. only included studies involving bouts of light to moderate activity, we did not limit studies based on exercise intensity. Moreover, we also performed meta-analyses of PA breaks in comparison to one continuous bout of exercise, reporting a small effect in favour of PA breaks on post-prandial glucose outcomes.

These post-prandial effects of PA breaks on measures of glucose, insulin and TAG could be relevant to the prevention of type 2 diabetes and atherosclerosis. The post-prandial [99] state is the more common metabolic state during non-sleeping hours for many people in modern society, who consume three large meals a day in addition to snacks and drinks [100]. Post-prandial and nocturnal hyperglycaemic excursions might be an early and undetected aspect of an insulin-resistant state [101]. Hyperglycaemic spikes are more strongly associated with, and might be more predictive of cardiovascular complications, risk and all-cause mortality than fasting plasma glucose or HbA1c levels [102] and should be targeted [103] since HbA1c is an integrative measure of blood glucose and post-prandial hyperglycaemia occurs even when HbA1c control is adequate [104]. Notably, post-load glucose-predicted cardiovascular mortality and diabetes, whereas neither fasting glucose nor HbA1c did [105]. Additionally, elevated 30 min post-load glycaemia is associated with increased risk of type 2 diabetes and all-cause mortality, independent of both fasting and 2 h post-load glucose [106]. Similarly, post-load insulin levels during a glucose tolerance test predict the development of type 2 diabetes [107], as insulin release is pulsatile, resulting in oscillating ultradian periodicity [108, 109]. Similarly, post-prandial excursions in TAG also increase CVD risk [110‐114], via atherogenesis [115]. Therefore, the moderate decreases in post-prandial glucose and insulin, and the small decrease in post-prandial TAG, as a result of PA breaks in sitting, if confirmed in longer-term studies, may have implications for the prevention of metabolic disease, at least in comparison with only sitting.

This meta-analysis suggests any differences in metabolic effects between regular PA breaks and one continuous bout of exercise are non-existent for TAG (Fig. 9) and insulin (Fig. 12), or statistically significant but small for glucose (Fig. 10). In a previous meta-analysis [6], MPA breaks were more effective than a single prolonged bout of MPA at regulating glycaemia, even when the study in which the continuous bout resulted in double the amount of energy expended compared to the intermittent bout was included [6]. However, only three studies [60, 62, 83], two of which were energy matched [60, 62], were meta-analysed [6]. In our current meta-analysis, when EE was matched, there was a small and statistically significant effect in favour of regular PA breaks on post-prandial glycaemia (Fig. 10). In the largest meta-analysis of observational studies to date, the increased risk of all-cause and CVD mortality associated with high sitting time, specifically sitting for more than 8 h daily, was entirely eliminated by approximately 60-75 min daily, and reduced by approximately 30 min daily, of self-reported PA [116]. PA breaks in the current meta-analysis totalled approximately 30 min of PA daily, with a small but statistically significant advantage for PA breaks over continuous exercise. Taken together, the observational and experimental research suggest that PA breaks might have a small advantage over continuous exercise, but any such advantages are abolished with high amounts of daily exercise. However, such comparisons between cross-sectional controlled laboratory studies and observational studies need to be interpreted cautiously, as the results of Ekelund et al. cannot rule out possible effects resulting from patterns of accumulated sitting.

The evidence on the effects of sitting on metabolic health generated in our review is supported modestly by epidemiological evidence. Recent prospective studies of total sitting time and incident type 2 diabetes, in contrast to cross-sectional studies of sedentary time and breaks measured by self-report [117], found little evidence for an association [118], or associations, between sitting behaviour or time and incident type 2 diabetes, but were limited to inactive [119] or obese [120] participants only. To resolve the discordant findings of prospective versus cross-sectional epidemiological studies, which do suggest an association between sitting time and type 2 diabetes [117], future prospective studies utilising accelerometer assessed total sitting time need to be conducted. Few prospective epidemiologic studies to date have assessed the links between breaks and metabolic outcomes, and even fewer support any associations. Baseline breaks, independent of total sitting time, did not predict any metabolic outcomes at 6-month follow-up [121]. Breaks were not associated with all-cause mortality over 5 years of follow up in older men [122]. To our knowledge, only one prospective epidemiological study to date has found an association between longer sedentary behaviour bouts, synonymous with infrequent sedentary breaks, and mortality risk [123]. Sedentary breaks here refer to any break in sedentary behaviour, measured in observational studies typically with Actigraph accelerometry. Cross-sectional studies of breaks that report device-measured sedentary time and breaks also present an unclear picture, with Actigraph measured breaks being inversely associated with some metabolic markers [124, 125]. However, there was little evidence for an association between sedentary breaks, quantified by a thigh worn ActivPAL inclinometer, and diabetes or metabolic syndrome [126]. Conversely, using the same device, number of long sitting bouts was deleteriously associated with several glucose and lipid biomarkers, although somewhat ameliorated by MVPA [127]. Thus, there is conflicting cross-sectional observational evidence, perhaps or perhaps not supporting our findings for a small advantage of PA breaks over continuous exercise. It should be noted that in the current meta-analysis breaks were PA breaks, with standing breaks excluded; thus PA breaks in the included experimental studies are not the same as sedentary breaks in observational studies.

As sedentary behaviour and physical activity guidelines development require some relative consistency between different types of evidence (experimental, epidemiological, etc.), it is surprising how sedentary breaks became part of several national guidelines [5, 128‐130] given that only one prospective observational study [123] objectively measured sedentary patterns, and none have used inclinometers, in relation to health risk. Additionally, only 1 experimental study [69] has investigated the effects of the patterning of PA breaks, reporting no differences between PA breaks performed every 20 or 60 min. Recently, the United States of America Physical Activity Guidelines Advisory Committee in its Scientific Report to the Secretary of Health and Human Services [24] concluded that there was insufficient evidence that bouts or breaks in SB are important factors in the relationship between SB and all-cause mortality, and incidence of or mortality from CVD, cancer, or incident type 2 diabetes or weight status. Accumulating brief bouts of PA between bouts of sitting throughout the day in a “whole day” approach [131] might be a feasible alternative for a considerable part of the population who do not exercise, a hypothesis that is supported by the results of the current meta-analysis, which found that there was a small advantage for PA breaks compared to one continuous exercise bout, on glucose, and no difference on insulin and TAG measures, especially as those with higher BMI appeared to benefit more. Therefore, given the results of the current meta-analysis of cross-sectional experimental studies, shedding light on the prospective associations between sedentary breaks and metabolic outcomes is an area of absolute priority for future epidemiological research.

In summary, the results of our systematic review and meta-analysis, viewed in the context of the wider literature, suggest that PA breaks, performed for example, throughout a normal working day, might be an alternative, or at worst, complementary for those who are unable to perform one bout of structured exercise training, particularly in those with higher BMIs, for the prevention of type 2 diabetes and atherosclerosis.

There was moderate to substantial heterogeneity (Figs. 4, 7) in the results that might be explained by the PA/health status of participants, sex, and also whether a study utilised single or multi-day designs.

It is unclear if the number, duration, intensity, amount and modality of PA breaks within a period of prolonged sitting, and the total duration of the sitting bout, are mediators in the metabolic responses to sitting, with only one study investigating and reporting that such variables did not affect glucose outcomes [69]. Most currently researched modalities involve light to moderate walking or running [7, 8, 13, 20, 21, 46, 68]. To date, only Dempsey et al. [47] and Hawari et al. [78] have examined metabolic responses to simple resistance activities (SRAs) as a means to interrupt sitting. Interestingly, engaging in own body weight resistance type exercises was associated with similar decreased risk of mortality compared to engaging in aerobic type exercise [132]. Pertinently, the modality of the exercise interrupting sitting, walking or cycling at very low intensity, even when energy matched, might play a role in modulating post-prandial glycaemic responses [14]. Future research should attempt to explore the effects of very light intensity breaks, such as fidgeting [133‐135], that can be performed at a low enough intensity, or very short duration HIIT [89, 90, 136, 137] breaks in sitting which constitute, “exercise snacks” [138], so as to address concerns about productivity, impracticality, the habitual nature of sitting [139] and management support [140].

Additionally, different sitting periods were used, along with different patterns of PA breaks. Some used 2 day laboratory designs [7, 8, 11, 45], whereas others used 1 day [20, 60, 62, 77]. A free-living protocol over 1 day [70] or 4 days [74, 75] was also used. Participants sat for bouts between 2.5 [16], 4 [77], and 7–9 h [7, 8, 20, 45‐47, 60, 62]. Breaking up sitting with exercise might have different effects depending on the duration of sitting, given that observational findings suggest that extended sitting time negatively affects metabolic health [123‐125]. However, this is as yet untested experimentally. Additionally, the duration of individual discrete sitting bouts varied, for example 1.5 min of brisk walking every 15 min [20] or every 30 min [62]. Interestingly, when participants had their sleep restricted, PA breaks did not attenuate post-prandial glucose measures compared to sitting. Therefore, future experimental research could systematically explore the effects of the number, duration and intensity of PA breaks, and also the total duration of the sedentary bout in which PA breaks occurred.

People who were overweight [71] or had lower CRF [81] experienced greater attenuation in post-prandial glucose. Subgroup analysis showed that those who were physically inactive, or had IFG or type 2 diabetes, experienced statistically significant greater glycaemic benefits from PA breaks, and attenuation of insulin also approached significance (Figs. 5, 8). In support of this, meta-regression revealed that PA breaks had a greater effect on glucose in participants with higher BMI.

No or small differences in glycaemia or lipaemia between EX and INT were reported in studies involving highly fit young men with maximal aerobic capacity (\(\dot{V}\) O₂max) above 50 ml.kg.min⁻¹ [7, 8, 45], whereas studies involving sedentary or metabolically unhealthy people appeared to find in favour of INT for glycaemia [20, 60, 62, 73] or continuous for lipaemia [62]. Glucose transport, uptake and metabolism might be higher in magnitude in the insulin sensitive compared to the insulin resistant [44]. Moreover, trained [42, 43] or insulin sensitive [41] participants demonstrate a greater response to a glucose or lipid challenge. Differences in glycaemia or lipaemia between sitting and PA breaks possibly would be greater in participants who are not exercise trained. Endurance training might alter lipid metabolite levels, composition and localisation, and thus muscle lipid metabolism and insulin sensitivity [41, 141].

We found a small body of evidence suggesting that sex might mediate glucose responses [13, 15, 47, 69] (Figs. 6, 11). Conversely, 2 studies [46, 72] reported no sex interactions for any outcomes when sitting was interrupted by light or moderate walking. Both sex specific PA break protocols or the underlying mechanisms, such as oestrogen levels [48‐51], responsible for any possible sex divergent metabolic responses to PA breaks could be avenues for further research. It would be desirable if future studies recruited more than one sex and were powerful enough to analyse and report sex-specific results, even if this was done only for completeness and subsequent findings were put in the appendices.

Meal timings, type of meals, whether high fat [7, 8, 20, 45], high carbohydrate [60], or mixed meals [62, 77, 83], liquid [60, 62] or solid [7, 8, 20, 77, 83], varied. Liquid meals might lower the magnitude of post-prandial excursions [142], thus the results might be affected by whether liquid or solid meals were used. The macronutrient and amino acid composition [143‐146], and the glycaemic index (GI) of meals may also modulate post-prandial metabolic responses.

Recently, Bailey et al. [63] reported that the GI of the breakfast meal and PA breaks both independently affected post-prandial glucose excursions, with little evidence that there were additive effects from combining PA breaks with a low GI meal.

Additionally, participants were fed one meal [46, 77] two meals [20], three meals [62] or six small meals [77]. Moreover, participants consumed their own breakfast prior to arrival in the laboratory for exercise trials [7, 8], whereas breakfast was provided for them as part of the test meal in another trial [20]. Therefore, feeding protocols might explain some of the heterogeneity in post-prandial responses, and should be investigated more comprehensively.

Furthermore, blood was drawn, for example, just before PA breaks [46, 47] or in rested, sedentary conditions, 1 day after PA breaks [7, 8], every 10 min [16], once every 2 h [10, 20] or assessed via CGMS [12, 14, 15, 22, 71, 82, 83]. Thus, results could have been affected by differing blood draw protocols [99, 147]. Notably, PA breaks reduced post-prandial iAUC up to 2 h after a meal, but not up to 4 h after [82], suggesting that meal timing in relation to blood draw schedule can significantly affect results.

Few studies so far have attempted to assess the underlying mechanisms responsible for metabolic responses to PA breaks, even if merely by assessing c-peptide, which would determine whether decreases in insulin are the result of decreased insulin secretion or increased clearance [148‐151]. Additionally, only a few studies have assessed lipoproteins [77, 79], adipose tissue gene expression [73], molecular signalling involved in glucose metabolism [152], or used new metabolomics methods [153] to assess lipidomics [154], and none have assessed branched-chain amino acids [153].

Since visual inspection of funnel plots is subjective and can lead to incorrect interpretations, even by medical researchers [155], a variety of methods were used to assess publication bias [53, 54, 156]. There might have been publication bias, or selective outcome or analysis reporting [56] in TAG measures from comparing PA breaks with sitting especially (Table 3) (Fig. 15). Additionally, there might have been publication bias for insulin measures from PA breaks compared to sitting and PA breaks compared to continuous exercise (Table 3). The Vevea and Woods [58] method estimated that the SMD for TAG measures comparing PA breaks with sitting would be reduced from − 0.27 to − 0.20, assuming severe 2-tailed selection bias (Table 3). Assuming the existence of severe 2-tailed bias, the effects of PA breaks on glucose and insulin would be still moderate, i.e. SMD of − 0.41 and − 0.43, respectively.

No subgroup analysis or meta-regression was performed to assess possible moderating effects of risk of bias on effect size because as stated, only a small number of studies reported randomisation, blinding and handling of data attrition clearly [15, 46, 47]. Future research should more clearly report randomisation, blinding and data attrition procedures, and should also more clearly fully report all data collected, even if statistically non-significant.

The current work has a number of strengths. Experimental controlled studies that evaluated the metabolic effects of PA breaks and those of continuous or prolonged exercise in the context of prolonged sitting were systematically synthesised. The metabolic effects of PA breaks compared to no exercise sitting were also systematically synthesised. A variety of publication bias analyses were conducted, and effect sizes in the event of severe publication bias were also calculated. A meta-regression identified BMI as a moderator for glucose and insulin responses to PA breaks. When data were not reported in a study, they were obtained from the authors.

Despite this, the limitations of the current work must be mentioned. Selection bias is a possibility, as only published peer-reviewed studies were included. The inclusion criteria for the meta-analysis could be a limitation, as only trials with explicitly controlled PA break protocols were included. The exclusion of studies with “free living” protocols might affect the results. However, when free-living trials that did not use strictly controlled laboratory protocols [70, 74‐76] were included in the meta-analyses the results and interpretation were not qualitatively substantively altered. For example when comparing EE matched PA breaks with continuous exercise, the SMD for glucose measures would have changed from − 0.26 (95% CI − 0.49 − 0.03, p = 0.03) to − 0.23 (95% CI − 0.41, − 0.05, p = 0.01). The SMD for insulin measures would have changed from 0.36 (95% CI − 0.45, 1.18, p = 0.38) to 0.24 (95% CI − 0.37, 0.84, p = 0.44), whereas the SMD for TAG would have changed from 0.06 (95% CI − 0.23, 0.36, p = 0.67) to − 0.01 (95% CI − 0.27, 0.26, p = 0.96). Similarly, when PA breaks were compared with sitting, with free-living trials included, the SMD for glucose measures changed from − 0.54 (95% CI − 0.70, − 0.37, p = 0.00001) to − 0.51 (95% CI − 0.67, − 0.35, p = 0.00001), for insulin measures from − 0.56 (95% CI − 0.74, − 0.38, p = 0.00001) to − 0.54 (95% CI:− 0.71, − 0.38, p = 0.00001), and for TAG measures from − 0.26 (95% CI − 0.44, − 0.09, p = 0.004) to − 0.31 (95% CI − 0.48, − 0.15, p = 0.0002).

In studies with more than 2 experimental conditions, the sample size of the control condition, uninterrupted sitting was divided by the number of times it was used as a control. For example, Pulsford et al. [21] had 3 experimental conditions, walking, cycling, and sitting. Therefore, in the meta-analyses, the sample size for the sitting condition was divided in half, as the sitting condition was used twice as the control comparison. When the results of experimental PA breaks conditions were combined instead, the SMD for glucose measures, comparing PA breaks with sitting, changed from − 0.54 (95% CI − 0.70, − 0.37, p = 0.00001) to − 0.52 (95% CI − 0.70, − 0.35, p = 0.00001). The SMD for insulin measures changed from − 0.56 (95% CI − 0.74, − 0.38, p = 0.00001) to − 0.54 (95% CI − 0.73, − 0.35, p = 0.00001) and the SMD for TAG measures changed from − 0.26 (95% CI − 0.44, − 0.09, p = 0.004) to − 0.26 (95% CI − 0.44, − 0.09, p = 0.005). The SMD for glucose measures, comparing PA breaks with continuous exercise, changed from − 0.26 (95% CI − 0.49, − 0.03, p = 0.03) to − 0.26 (95% CI − 0.50, − 0.02, p = 0.04).

Furthermore, the scales used to measure metabolic responses—CGMS, iAUC, tAUC—were heterogeneous. One study collected both CGMS [15] and venous blood [47] measurements. The current meta-analysis utilised the CGMS data from Dempsey et al. [15]. If the venous blood data from Dempsey et al. [47] were utilised instead, the SMD for blood glucose, for PA breaks versus sitting, would have changed from − 0.54 (95% CI − 0.70, − 0.38) to − 0.50 (95% CI − 0.65, − 0.35).

Altenburg et al. [67] was omitted from the meta-analysis, as data were not normally distributed and reported in medians, but inclusion would likely not have altered the main results. The meta-analysis included one study that was not randomised [77]. However, removing it would not have affected the results. TAG SMD for INT compared to SIT changed from − 0.27 (95% CI − 0.45, − 0.08, p = 0.005) to − 0.28 (95% CI − 0.48, − 0.08, p = 0.005). Similarly, TAG for INT compared to EX changed from 0.04 (95% CI − 0.23, 0.31, p = 0.77) to 0.00 (95% CI − 0.30, 0.29, p = 0.98).

Studies that included only interrupted sitting with standing were not included, as standing might not exceed 1.5 METs [29, 30] and heterogeneity in EE during standing might be affected by leg or body displacement [30]. Additionally, normal weight men and women, BMI: 22.5 ± 1.5 kg/m², had higher leg muscle activity during sitting compared to the overweight, BMI: 28.4 ± 2.9 kg/m². Conversely, leg muscle activity was higher in overweight adults during standing [31]. However, standing might confer positive [157] or negative [158, 159] physiological effects beyond simply EE, and thus a future meta-analysis should evaluate the effects of using standing to break up sitting.

Only BMI was assessed as a moderator variable in the meta-regression. \(\dot{V}{\text{O}}_{2\text{max} }\) could not be evaluated, as generally only studies with aerobically fit or physically active participants reported \(\dot{V}{\text{O}}_{2\text{max} }\) or \(\dot{V}{\text{O}}_{{2{\text{peak}}}}\) values [7, 8, 11, 16, 45, 62]. Future experimental work should attempt to assess CRF, as it has been suggested that CRF might modulate responses to PA breaks in sitting [160]. Similarly, exercise intensity could not be assessed in the meta-regression due to intensity being reported either as absolute or relative intensity. Nor could the prior PA levels of participants be assessed as a continuous moderator because some studies merely reported highly aerobically fit participants as “recreationally active” [7, 8] whereas others did not report PA status [63, 68, 77]. Similarly, this meant that in the subgroup analyses (Figs. 3, 5, Electronic Supplementary Material Appendix S2—Figs S3, S5 and S6), studies were grouped such that participants in one subgroup were physically active, or sedentary, or were overweight/obese or had type 2 diabetes or IFG, compared to another subgroup with active people. It should be noted that differences observed in subgroup analyses (Fig. 15) based on summary data are considered observational, and need to be specifically tested in within subjects experimental designs, for example in participants with lower compared to higher BMI. Similarly, even though BMI was identified as a moderator in the meta-regression, this was based on summary data, is observational [86] and needs to be specifically tested in future experimental studies. Similarly, effect sizes were calculated using summary data from individual studies, and not individual participant level data.