The association between sedentary behaviour, physical activity and type 2 diabetes markers: A systematic review of mixed analytic approaches

The negative effect of sedentary behaviour on type 2 diabetes markers is established, but the interaction with measures of physical activity is still largely unknown. Previous studies have analysed associations with single-activity models, which ignore the interaction with other behaviours. By including results from various analytical approaches, this review critically summarises the effects of sedentary behaviour on diabetes markers and the benefits of substitutions and compositions of physical activity. Ovid Medline, Embase and Cochrane Library databases were systematically searched. Studies were selected if sedentary behaviour and physical activity were measured by accelerometer in the general population, and if associations were reported with glucose, insulin, HOMA-IR, insulin sensitivity, HbA1c, diabetes incidence, CRP and IL-6. Forty-five studies were included in the review. Conclusive detrimental associations with sedentary behaviour were determined for 2-h insulin (6/12 studies found associations), fasting insulin (15/19 studies), insulin sensitivity (4/6 studies), diabetes (3/4 studies) and IL-6 (2/3 studies). Reallocating sedentary behaviour to light or moderate-to-vigorous activity has a beneficial effect for 2-h glucose (1/1 studies), fasting insulin (3/3 studies), HOMA-IR (1/1 studies) and insulin sensitivity (1/1 studies). Compositional measures of sedentary behaviour were found to affect 2-h glucose (1/1 studies), fasting insulin (2/3 studies), 2-h insulin (1/1 studies), HOMA-IR (2/2 studies) and CRP (1/1 studies). Different analytical methods produced conflicting results for fasting glucose, 2-h glucose, 2-h insulin, insulin sensitivity, HOMA-IR, diabetes, hbA1c, CRP and IL-6. Studies analysing data by quartiles report independent associations between sedentary behaviour and fasting insulin, HOMA-IR and diabetes only for high duration of sedentary time (7–9 hours/day). However, this review could not provide sufficient evidence for a time-specific cut-off of sedentary behaviour for diabetes biomarkers. While substituting sedentary behaviour with moderate-to-vigorous activity brings greater improvements for health, light activity also benefits metabolic health. Future research should elucidate the effects of substituting and combining different activity durations and modalities.


Introduction
Sedentary behaviour (SB), defined as any activity below 1.5 Metabolic Equivalent (MET) in either a lying, reclining or sitting position [1], is unfavourably associated with several T2D biomarkers independent of moderate-to-vigorous activity (MVPA) [2,3]. Despite many countries having introduced guidelines on SB, such as UK [4] and Australia [5], a recent paper [6] summarising the research behind the newest WHO guidelines [7] highlighted the lack of sufficient evidence to set time-based recommendations regarding SB. Additionally, the paper emphasised the need to establish alternative ways to offset the damaging effects of SB, given that the high levels of MVPA (> 300 minutes/week) needed to reduce the mortality risk-as suggested by the current evidence-may be unfeasible for a large part of the population. However, while there is proof of the benefits of MVPA on health outcomes, less clear is the effect of low-intensity physical activity (LIPA). Several works document the benefits of LIPA for markers of T2D (e.g. [8][9][10]), but many studies analysing the cross-sectional association between SB and diabetes markers have not included LIPA in the analysis, given the high collinearity between the two variables, which prevents them from being included simultaneously in regression models. A review by Chastin et al. [11] found that LIPA improves cardiometabolic health, but did not include information on the relationship with SB. Amagasa et al. [12] reviewed the effects of LIPA on cardiometabolic biomarkers and found positive associations independently of MVPA, but could not find evidence of the combined effects of time spent in PA and SB. Studies using the isotemporal substitution model (ISM) [13] or a compositional transformation [14] have been able to include LIPA in the models and reported benefits associated with the reallocation of SB to LIPA. Previous reviews [15][16][17][18] summarised the evidence around the reallocation of time between SB and PA in the adult population, but did not address whether there is a minimum and maximum allocation time of LIPA and MVPA for which such benefits are observed.
This review has three aims. Firstly, to summarise the current evidence on the effects of SB on T2D biomarkers, accounting for MVPA, in the general healthy adult population. Previous systematic reviews [2,19] have analysed the association between device-measured SB and cardiometabolic biomarkers, but did not exclude studies that did not account for MVPA in the analysis. Given that the relationship between SB and health outcomes changes depending on the MVPA level [20], accounting for MVPA is crucial to obtain reliable results. Secondly, this review aims to assess how different time allocations and compositions impact the magnitude of the reduction in T2D markers. Thirdly, we compare the findings obtained with different analytical approaches, namely linear regression, linear regression by quartiles, ISM and compositional transformation. To our knowledge, this is the first review to compare the association between SB and biomarkers of T2D obtained with different analytical methods.

Methods
The PRISMA guidelines [21] were followed to conduct and report the results of this review (see Fig 1).
We searched Ovid Medline, Embase and Cochrane Library for relevant publications, with no specified beginning or end date of publication, on the 15th June 2021. The search term used can be found in the S1 File.
For inclusion, each study had to meet the following criteria: Exclusion criteria included: study on non-general population (diabetic, pre-diabetic, BMI > 30, with metabolic syndrome, pregnant, with pre-existing conditions), sedentary behaviour defined as not meeting the activity guidelines (� 150 mins/week), physical activity and sedentary time assessed by subjective methods or devices other than accelerometer (such as pedometer).
FRC and CG screened all titles and abstracts for the studies obtained through the electronic search, and the studies meeting the inclusion criteria underwent full text review. Any discrepancies were discussed between FRC and CG both at the title/abstract screening stage and the full text review stage. Disagreements were resolved with consensus at both screening stages. The Covidence systematic review software was used for the study selection.

Quality assessment
Each of the selected papers was assessed for quality on a scale from 0 to 7, where scores � 3 represent poor quality, 4-5 medium quality and � 6 good quality. The quality assessment tool was developed based on the Newcastle-Ottawa Scale [23] and adapted following the scale developed by Brocklebank et al. [2]. The Newcastle-Ottawa Scale was developed for prospective studies and thus, many criteria were not applicable to the studies included in the review (for e.g., "Demonstration that outcome of interest was not present at start of study" and "Adequacy of follow up of cohorts"). Therefore, additional criteria from Brocklebank et al. [2] were integrated to reflect the key characteristics of the selected studies. Two points were available for selection of the cohort (representativeness and description), one for measurement of exposure (at least 4 valid days of accelerometer data), three for comparability (adjusted for BMI and/or waist circumference; for sex, age and ethnicity; and for accelerometer wear time) and one for data analysis (if points estimates and measures of variability were reported). The quality assessment was done by FRC and verified by CG.

Data extraction
We extracted author, population, study type, observation period (if applicable), sample size, mean age, outcomes, covariates (gender, age, BMI, diet, etc.), accelerometer device type, unit of exposures (PA and SB), statistical method, and effect measure for each study. Cut-offs and units (hours/day, total time, etc.) were recorded for the exposures. The primary outcome was the association between sedentary behaviour and T2D markers, adjusted or substituted with different PA intensities. Due to the heterogeneity between the studies-especially regarding devices and activity cut-offs-no meta-analysis could be conducted, and the data is presented qualitatively.
The significance of the associations was determined based on the criteria by Sallis et al. [24] for which there is an association between exposures and outcome only if more than 60% of the studies report a significant association (the association is indeterminate if 34-59% report it, and no association if 0-33% report it).
The data extraction was done in Microsoft Excel by FRC and verified by CG.

Results
The initial search identified 2593 studies, 46 of which were included in the review. Three studies had a prospective design [25][26][27]. There was high heterogeneity between the studies (see Table 1). A wide array of monitoring devices was used, with Actigraph being the most common; 11 studies applied activities cut-offs as Metabolic Equivalent (MET), while 34 studies as counts-per-minute (cpm); among the latter, SB was generally classified as < 100cpm, while for MVPA a wide array of values was used (760cpm, 1040cpm, 1486cpm, 1535cpm, 1952cpm, 2020cpm, 2690cpm). Most studies reported sedentary time as hours or minutes per day, while activity was mostly reported as a percentage of the total recorded time. Fasting glucose, fasting insulin and HOMA-IR were the most frequently reported outcomes. 14 studies used a population that was free of existing metabolic conditions (four of which needed to do so as they reported on diabetes incidence [25, 28-30]) and 20 studies adjusted for BMI and/or waist circumference, gender, age, ethnicity and MVPA simultaneously (see S3 Table in S1 File). Eight studies used the ISM, four used a compositional transformation, 14 reported mean differences, and the rest evaluated associations with either linear or logistic regression, reporting Odds Ratios (OR), Risk Ratios (RR) or regression coefficients. Of the selected studies, seven had a high-quality rating (score of 7 or 6) and 14 had a low-quality rating (score � 3). A complete description of the studies included can be found in Table 2, while a detailed quality assessment can be found in S3 Table in S1 File.

PLOS ONE
The association between sedentary behaviour, physical activity and type 2 diabetes markers

PLOS ONE
The association between sedentary behaviour, physical activity and type 2 diabetes markers Generalised linear models and logistic regression.

PLOS ONE
The association between sedentary behaviour, physical activity and type 2 diabetes markers

PLOS ONE
The association between sedentary behaviour, physical activity and type 2 diabetes markers

PLOS ONE
The association between sedentary behaviour, physical activity and type 2 diabetes markers

PLOS ONE
The association between sedentary behaviour, physical activity and type 2 diabetes markers Gennuso et al. [45] found associations with glucose for every hour/day increase in SB, but only for the top quartile of MVPA, which included participants accumulating � 300 minutes of MVPA per week: each hour of SB increased the odds of high glucose (defined as � 5.55 mmol/L) by 13%.
Isotemporal substitution. Three studies using substitution analyses found no associations with SB [27, 38,67]. On the other hand, three other studies found significant results: Ekblom-Bak et al. [39] found that substituting 30 minutes of SB with 30 minutes of MVPA was associated with a 0.9% improvement in fasting glucose; substituting with LIPA did not produce significant results. Buman et al.
[33] reported a 1.3% reduction in fasting glucose when substituting 30 minutes of SB with MVPA; no reduction was observed when substituting with LIPA. Healy et al. [49] reported a 2% reduction when reallocating 2 hours/day of sitting time with standing time; however, reallocations with stepping were not significant.
Linear regression. Three studies using linear regression [25, 26, 49] did not find significant associations with SB while two studies found significant associations between sedentary behaviour and 2-h glucose after adjusting for MVPA.
Healy et al. [47] found an increase in 2-h glucose for every extra hour of SB (β = 0.23, p = 0.019); Maher et al. [55] also found that higher SB was associated with higher 2-h glucose, but with a smaller effect size (β = 0.02, p<0.05).
Linear regression by quartiles. Qi et al. [61] found a significant difference between quartiles of SB, with a difference between bottom and top quartiles (9.9 and 13.7 hours/day respectively) of 8mg/dL (p-value < 0.0001). Diaz et al. [37] reported significant differences between quartiles, with a difference of 11.2 mg/dL between the top and bottom quartiles. Farrahi et al. [43] did not find significant differences between quartiles of sedentary bouts accumulations after adjusting for MVPA. Mossavar-Rahmani et al. [57] did not find significant differences between quartile of SB (cut-offs of 10.8, 12, 13 hours/day) over 6 years of follow-up. Healy et al. [48] also did not find significant differences between quartiles of SB.

Isotemporal substitution.
Healy et al. [49] found that reallocating 2 hours of sitting to standing did not produce significant benefits, while the reallocation to stepping lead to a 12% reduction in 2-h glucose (p-value < 0.001) and the reallocation from standing to stepping lead to an 11% reduction (p-value = 0.005).
Compositional transformation. With a compositional paradigm, Farrahi et al. [42] found significant associations between SB and 2-h glucose however only for participants accumulating less than 7.5 hours/day of sleep; MVPA was beneficially associated to glucose regardless of sleep time (β <7.5h/d = -0.06, p-value < 0.001, β >7.5h/d = -0.04, p-value = 0.04), while no significant associations were reported for LIPA. Lynch et al. [54] found significant associations only for SB greater than 9.84 hours/day, but the difference between quartiles was still significant (p-value = 0.01), with a difference of 12 pmol/ L between top (9.84 hours/day) and bottom quartiles (7.74 hours/day); however, after adjusting for waist circumference, only the difference between the first and third quartile remained significant. Healy et al. [48] reported a significant difference of 11.6 pmol/L between top and bottom quartiles of SB, which differed by 2.3 hours of sedentary time. Peterson et al. [59] found a significant difference between high, low and moderate SB groups, but only for the high MVPA subgroup. Qi et al. [61] reported significant differences between quartiles of SB, with the most sedentary quartile having fasting insulin increased by 1.3 mU/L compared to the least sedentary quartile. Farrahi et al. [43] found significant differences between couch potatoes (highest number of sedentary bouts interrupted less frequently) and short sitters (accumulating SB in short bouts) with a % difference of -8.8 (p-value < 0.001); no significant difference were found between couch potatoes, prolonged sitters (accumulating SB in bouts > 15-30 minutes) and breakers (accumulating less SB, which was frequently interrupted by longer non-SB time).

Fasting insulin
Only one study [57] did not find significant differences between quartiles of SB. Isotemporal substitution. Buman et al. [33] found that substituting 30 minutes of SB to LIPA reduced the risk ratio to 0.99 and caused a reduction in insulin of 2.4%; substituting with MVPA reduced the RR to 0.87 and caused a reduction in insulin of 14.5%.; moreover, reallocating 3 hours/day of sedentary time to LIPA produced the same insulin reduction as reallocating 30 minutes/day of sedentary time to MVPA. Ekblom-Bak et al. [39] reported a -0.001% change about the mean when 10 minutes of SB were replaced with 10 minutes of LIPA.
Whitaker et al.
Compositional transformation. Using a compositional framework, Chastin et al.
McGregor et al. [56] reported no significant associations for SB, but found a significant association with MVPA (β = -0.116, p-value < 0.001) using a compositional paradigm, which implies reallocating other behaviours including SB for MVPA.
Linear regression. One study analysed associations with 2-h insulin with linear regression, reporting no associations [26].
Linear regression by quartiles. Farrahi et al. [43] found a significant % difference of -6.5 (p-value = 0.048) between couch potatoes (highest number of sedentary bouts interrupted less frequently) and short sitters (accumulating SB in short bouts), but differences with the other sedentary groups disappeared after adjustment for MVPA.
Compositional transformation. Farrahi et al. [42] reported significant associations with SB (β = 0.22, p-value = 0.001), LIPA (β = -0.3, p-value < 0.001) and MVPA (β = -0.28, pvalue < 0.001) within a compositional framework, and additionally they reported significant reallocations between SB and both LIPA and MVPA, with the latter yielding a higher effect size. One study [25] found that each hour of SB was associated with an increase in HOMA-IR of 5.8%, which persisted after adjustments for covariates but not in the longitudinal analysis.

HOMA-IR
Linear regression by quartiles. Three studies found no differences between groups or quartiles of SB [37,57,59] [54] reported significantly lower HOMA-IR between the first and third quartiles of SB (difference of 2.19) in the fully adjusted model. Qi et al. [61] found significant differences between quartiles of SB (9.9, 11.6, 12.6, 13.7 hours/day), with a difference of 0.33 between the top and bottom quartiles.
Isotemporal substitution. Ekblom-Bak et al. [39] reported that substituting 30 minutes of SB with LIPA was associated with 3.1% lower HOMA-IR, and 12.4% lower if replaced with MVPA; they also found that the RR decreased linearly as increasing bout lengths were substituted.
Substitution analyses with a compositional transformation reported significant results: Chastin et al.
Linear regression. Barone Gibbs et al. [25] reported that participants spending more than 10 hours/day in SB had 3.8 times greater odds ratio of diabetes compared to participants who were sedentary for less than 6 hours/day over 7 days; each hour of SB was associated with an odds ratio increasing of 22%. Van der Velde et al. [66] found that for each 1.63 hours/day of SB, the odds ratio increased by 1.35 (CI 1.18, 1.55).
Linear regression by quartiles. Honda et al. [29] found that after adjusting for MVPA, only SB of more than 10 hours/day was associated with an odds ratio of diabetes (OR = 1.84, p-value = 0.04).
Linear regression. Three studies [25, 49, 64] did not find significant associations between SB and HbA1c. Barone Gibbs et al.
[25] did not find cross-sectional associations with SB, but the longitudinal association almost reached significance (p-value = 0.06). Mossavar-Rahmani et al. [57] found differences between quartiles of SB, but they disappeared after adjustment for MVPA.
Linear regression by quartiles. Bakrania et al.
[31] compared sedentary and inactive individuals to sedentary active and non-sedentary active individuals, who had a HbA1c reduction of 0.11% (p-value = 0.009) and 0.12% (p-value = 0.003) respectively; no significant benefits in HbA1c were observed for the non-sedentary inactive group.
Maher et al. [55] reported a coefficient β = 0.03 for the association between SB and CRP adjusted for MVPA.
Linear regression by quartiles. Three studies found no differences in CRP between quartiles or groups of SB [54,59,61].
Healy et al. [48] found a difference of 0.04 mg/dL between top and bottom quartiles of SB (the difference in SB was 2.3 hours/day). Between groups of activity (G1: active and non-sedentary, G2: active and sedentary, G3: inactive and non-sedentary, G4: inactive and sedentary), Loprinzi et al. [53] found that G1 and G2 had reduced levels of CRP compared to G4 (of -0.12 and -0.1 respectively) while G3 was not significantly different.
Linear regression. Elhakeem et al. [41] found no significant association with SB after adjusting for MVPA. However, Parson et al. [58] found an increase of 4.7% for every 30 minutes of SB.
Isotemporal substitution. Phillips et al. [60] found that replacing 30 minutes of SB with LIPA increased IL-6 of 0.34 (standardised β), and replacing SB with MVPA produced a decrease of 0.3.

Discussion
From the combined analyses of all the studies included, no association exists between SB and fasting glucose, HbA1c or CRP; the overall quality of the studies included is medium for all three biomarkers. We found evidence of a positive association with fasting insulin, 2-h insulin, incident diabetes and IL-6-with overall medium study quality. A negative association was for insulin sensitivity in healthy adults, but the overall study quality was found to be poor. No clear association could be determined for HOMA-IR, with overall study quality being medium. Table 3 provides a summary of findings for each biomarker, stratified by analytical approach.
The present finding that fasting insulin is associated with SB is consistent with a previous review by Powell et al. [19], though they reported associations between SB and fasting glucose, which is not found in this review. This difference may be due to the use of unadjusted data in their meta-analysis, as in the present review we found that the negative association between SB    and fasting glucose disappeared when models were adjusted for MVPA. Brocklebank et al. [2] reported a negative association with insulin sensitivity-also found in this review-but insufficient evidence for other biomarkers of T2D. This difference may be due to the lower number of studies included in the previous review, to the inclusion only of populations with diabetes

PLOS ONE
The association between sedentary behaviour, physical activity and type 2 diabetes markers and/or at high risk of metabolic conditions and to the lack of studies using compositional or isotemporal substitution analyses, which reported significant associations in the present review. A review on elderly individuals [3] reported similar findings to the present study: SB was negatively associated with fasting insulin and HOMA-IR, while no association was found with glucose levels. Studies using the ISM or a compositional transformation report significant associations between T2D markers and reallocations between SB and PA. Benefits were reported when SB was substituted with either LIPA or MVPA for 2-h glucose, fasting insulin, 2-h insulin, HOMA-IR, insulin sensitivity, CRP and IL-6. However, for 2-h glucose and HOMA-IR, the overall quality for studies using ISM was poor and therefore, we cannot exclude the presence of bias in the composite result. For incident diabetes and hbA1c, none of the included studies employed ISM or a compositional transformation to analyse potential associations. For insulin, substituting SB with LIPA resulted in a lower reduction in RR-albeit still significantthan for MVPA substitutions. Other reviews analysing the benefits of reallocating SB with PA found reductions in cardiometabolic biomarkers [15,16] for substitutions of both LIPA and MVPA. In children, only MVPA reallocations seem to be associated to a reduction in body fat, with greatest benefits for 60-minute reallocations, while substitutions for LIPA do not have evidence of association [17]. A review including only compositional associations with biomarkers of glucose and insulin control [18] did not find conclusive evidence of association with measures of SB and PA. Despite this disagreements in the evidence obtained through ISM and compositional analysis, this review observed evidence of agreement between the two methods, as also speculated by a recent consensus on methodology to analyse associations with accelerometer-measured PA [69].
On the other hand, the results derived from standard linear regression, from regression stratified by SB levels, from ISM and from compositional approaches are in disagreement for fasting glucose, 2-h glucose, 2-h insulin, insulin sensitivity, HOMA-IR, incident diabetes, CRP and IL-6. The absence of associations between SB and T2D markers when standard regression is employed, indicates that the assumption of linearity between exposure and outcomes may be wrong. The studies included in this review that analysed the participants by quartiles report a dose-response relationship between SB and biomarkers and independent associations with SB only for high duration of sedentary time: Lynch et al. [54] reported significant associations between SB, fasting insulin and HOMA-R for more than 8.8 hours/day of SB; Healy et al. [48] found a cut-off of 7.24 hours/day of SB that increase fasting insulin; Honda et al. [29] found that more than 10 hours/day of SB significantly increase the OR of insulin resistance and diabetes. A meta-analysis [70] found evidence of a logarithmic dose response relationship between SB and all-cause mortality and an accelerometer-measured 9-hour cut-off of SBadjusted for MVPA-after which hazard ratios of all-cause mortality increased. For studies measuring SB with subjective methods, the cut-off was around 7 hours; a similar cut-off (6-8 hours) for subjective SB was also found by Patterson et al. [71]. The discrepancy between cutoffs found with device-measured and self-reported measures of SB are likely due to participants underestimating the time spent in SB: a recent meta-analysis quantified the difference between device-measured and self-reported measures SB as~1.74 hours/day [72]. As such, a threshold between 7-9 hours/day of SB may be significant not only for all-cause and CVD mortality, but also for individual biomarkers of T2D. However, due to the scarce evidence found by this review, it is not possible to recommend a specific cut-off for SB that significantly increases biomarkers of T2D. Despite the evidence of deleterious effects of high SB, a review on the interaction between subjectively measured MVPA and SB [20] found that adults who engaged in MVPA for 60-75 minutes/day did not have an increase in mortality risk even for sitting times greater than 8 hours/day. However, the present finding that biomarkers of T2D are associated to SB for many hours per day suggests that such levels of MVPA are not achieved by the participants in the selected studies and are likely unattainable by a large part of the population. As a result, displacing SB with LIPA, or a mixture of MVPA and LIPA, could be the most feasible option. Supporting the case for displacing SB with longer bouts of LIPA rather than shorter MVPA bouts, a randomised crossover trial [73] found that substituting one hour of daily sitting with MVPA (cycling) does not improve insulin nor glucose, while replacing 6 hours/day of sitting with 4 hours of leisurely walking and 2 hours of standing has significant beneficial effects on both insulin and glucose.

Implications of findings
The present findings support the case for time-based guidelines to reduce prolonged periods of sedentary time in the 18-64 healthy population. Several studies included in the review reported clinically significant changes in biomarkers. For example, Barone Gibbs et al. [25] reported a 4.8% increase in fasting insulin for every additional hour of SB; Celis-Morales et al. [35] found that the highest SB quartile has a 6.2% increase in insulin compared to the lowest quartile of SB. According to Buman et al. [33], substituting 30 minutes of SB with LIPA reduced insulin by 2.4%, and substituting with MVPA reduced it by 14.5%. These effect sizes are comparable to the effect of a LIPA intervention [74], which produced a 18.2% reduction after three months in high-risk patients for T2D. Additionally, a 20% difference in area under the insulin response curve, was found to be associated with a 10% difference in coronary mortality risk [75], thus suggesting that SB reallocations could have a clinically meaningful effect. However, other studies included in the review reported much smaller percent reductions in insulin: Carson et al.
[34] reported a 0.03% change for every additional hour of SB; Healy et al. [48] and Lynch et al. [54] found 0.32% and 0.34% insulin difference between top and bottom quartiles of SB (difference in SB of 2.3 and 2.1 h/day), respectively. This potentially invalidates the claim of clinical relevance, and a meta-analysis to calculate a pooled effect size is needed to formulate a reliable conclusion.
Despite finding some evidence of a cut-off between 7-9 hours/day of SB, more evidence is required to support this value specifically for T2D biomarkers. Moreover, this review highlights the need for more evidence on how to spend the time gained from the reduction of SB most effectively. Studies analysing associations by quartiles found evidence on the deleterious effects of SB-especially for long sedentary periods-but, unlike other analytical approaches, cannot inform on how to combine PA of different kinds and intensity to benefit health. Current PA guidelines focus on moderate and vigorous PA, but lower intensity activities can also benefit T2D markers, as found in this review. More prolonged but less intense exercise may be easier for the elderly or impaired, while short but more intense activities could be as beneficial for those who cannot engage in prolonged sessions. Future research should analyse the effects of different bout lengths on a wide array of metabolic biomarkers to provide the data necessary to develop new and quantified guidelines on SB and LIPA, alongside the already present MVPA recommendations.
Additionally, future studies should consider using analytical approaches that examine the interaction between SB and PA of various intensities. While ISM allow to evaluate the association with reallocations of SB with PA, it does not allow to examine time compositions including different kinds of PA, as the reallocation is done between SB and one behaviour at a time. Moreover, this method presents the same limitations as standard linear regression, i.e. multicollinearity and assumption of linearity, and therefore is not recommended [69]. Other approaches, such as compositional transformation, may be better suited to assess the joint effect of SB and PA on measures of health. For example, Debache et al. [36] analysed different combinations of daily behaviours (lying, sitting, standing, LIPA and MVPA) with compositional models and reported varying differences in cardiometabolic biomarkers from the sample mean.

Strengths and limitations
While this review found evidence of the benefits of different PA intensities (LIPA and MVPA), it was not possible to systematically assess how bout duration affects T2D biomarkers, as all studies but two only reallocated for 30 minutes of activity. Chastin et al. [14] found a minimal reduction in glucose, insulin and HOMA-IR for a 10-minute reallocation; but given that the results were reported as % change about the mean, it was not possible to compare their results to the studies substituting for 30-minute bouts which used risk ratios. Healy et al. [49] reallocated 120 minutes of SB to LIPA and MVPA and found significant but small, reductions in fasting and 2h-glucose, suggesting that longer substitutions may influence glucose, which is not seen for shorter time substitutions. Therefore, more studies are needed to assess the impact of substituting different bout lengths. Moreover, a greater variety of activities should also be included in the analyses, as different types of activities have different effects on T2D markers [76]. Additionally, this review raises the issue around the discrepancy between results obtained with linear regression and other analytical methods. Finally, a major limitation of the present review is the lack of a pre-registered protocol. However, despite the absence of a review protocol, we adhered to the best practices for conducting a systematic review in order to minimise the potential for bias. This review has several strengths: it solely includes device-measured methods for monitoring activity (specifically by accelerometer) and studies that adjusted for MVPA in the analyses. Additionally, only studies examining associations in healthy adults (18-64) were included, to consider evidence relevant to the current age-specific guidelines. To the authors' knowledge, this is the first review including both results from standard and stratified linear regression and compositional approaches (including ISM) whilst also comparing the methods. Despite other analytical approaches exist for analysing the association between accelerometer data and health outcomes, such as functional analysis [77] and multivariate pattern analysis [78], we were unable to include them in the present review, as studies employing such methods did not fulfil the selection criteria. This review presents other limitations such as the lack of prospective studies and randomised controlled studies, which are lacking in the literature; consequently, no considerations around causality can be made at this time. We could not answer the question about how different types of activity affect marker levels for several of the included outcomes, as few studies reported on reallocations with LIPA. Equally, we could not find evidence for how substitutions with different bout durations impact the risk reduction magnitude, as most studies reallocated only 30 minutes between activities, and reallocated only one kind of activity at a time. No meta-analysis could be done given the high heterogeneity of studies in terms of devices and activity cut-offs used, effect sizes and statistical models.

Conclusion
This review found evidence of the negative association between SB and fasting 2-h glucose, fasting insulin, 2-h insulin, incident diabetes and IL-6. In addition, we found some evidence of a threshold of 9 hours/day, after which the effect of SB on T2D biomarkers is independent of MPVA, likely due to failing to accumulate enough MVPA to counteract the increased risk. However, the evidence is too scarce to provide definite recommendations regarding a timespecific cut-off for SB. While this review provides evidence of the health benefits associated with LIPA and MVPA, it was not possible to determine what durations and compositions are required to compensate for the reduction in intensity. Further work to confirm a time-based threshold and enable quantitative recommendations for SB, as well as flexible and achievable replacement of SB time is needed to provide actionable evidence for policy makers and clinicians.