The Association between Body Composition using Dual energy X-ray Absorptiometry and Type-2 Diabetes: A Systematic Review and Meta-Analysis of Observational studies

The association between objective measures of body composition (BC) with type 2 diabetes (T2DM) is inconclusive. We conducted a systematic review and meta-analysis to examine the association between several body composition (BC) indices assessed using dual energy X-ray absorptiometry (DXA), and T2DM. Using PRISMA guidelines, we searched for observational studies investigating BC measures, including total body fat mass (BFM), visceral fat mass (VFM), subcutaneous fat mass (SFM), and fat free mass (FFM); and T2DM. Of 670 titles initially identified, 20 were included. High VFM was consistently associated with T2DM. For every kg increase in VFM, the odds of having T2DM increased by two-fold for males (OR 2.28 [95% CI 1.42 to 3.65], p = 0.001) and more than 4-fold for females (OR 4.24 [1.64 to 11.02], p = 0.003). The presence of T2DM was associated with 2-fold higher odds of low FFM (OR 2.38 [1.44 to 3.95]). We found evidence that greater VFM is a risk factor for prevalent and incident T2DM. While the presence of T2DM is associated with reduced FFM; the relationship between FFM and BFM with T2DM remains unclear. Reducing VFM and increasing FFM through lifestyle changes may reduce the risk of T2DM and mitigate its deleterious effect on BC, respectively.

The prevalence of diabetes mellitus (DM) is rapidly increasing worldwide, with 592 million people expected to have the condition by 2035 1 . As such, the prevention and management of DM has become a crucial public health concern with an emphasis on addressing modifiable risk factors, such as poor diet, sedentary lifestyle, and obesity 2 .
While the relationship between obesity and DM, using surrogate anthropometric measures such as body mass index [BMI] 3,4 , waist circumference [WC] 5,6 , and waist-hip ratio [WHR] 7,8 , have been widely studied, results have been equivocal 9,10 , as these proxy measures are unable to distinguish body composition (BC) indices such as fat tissue deposition from muscle mass and bone density. Furthermore, these surrogate measures do not provide information on the location of fat mass deposition, emphasizing the need to evaluate the contribution of objective components of BC on DM risk.
Specific BC measurements can be estimated indirectly using various objective techniques such as air displacement plethysmography (BodPod), bioelectrical impedance analysis (BIA) 11 , and dual energy X-ray absorptiometry (DXA). However, BC profiles measured by different objective techniques are not interchangeable, as studies have reported poor concordance between these different methods [12][13][14] . In this analysis we only included studies that assessed BC measures using DXA, which is commonly used 15 , and is an acceptable technique for measuring BC in clinical studies 16,17 with a substantially lower cost, no complex post processing requirements, and minimal radiation exposure (~1 µSV) compared to computed tomography (CT) or magnetic resonance imaging (MRI) scan 18 . There is increasing evidence that DXA-assessed BC measures are associated with the onset, progression, treatment response, and health outcomes of cardiometabolic diseases, including type 2 DM (T2DM) [19][20][21][22] .
To date, there is no comprehensive review on the relationship between DXA-assessed BC measures, including total body fat mass (BFM), visceral fat mass (VFM), subcutaneous fat mass (SFM), muscle mass (MM), and the presence and incidence of DM. A better understanding of this association is critical as it can inform clinical guidelines and interventions for the management of DM. We therefore performed a systematic review and meta-analysis of observational studies evaluating the bidirectional associations between objectively assessed BC measures (i.e. BFM including VFM and SFM; and MM) and the prevalence or incidence of T2DM. We also identified key knowledge gaps and suggest future research directions.

Methods
Literature search. We performed a systemic review and comprehensive literature search using three sources (PubMed, Web of Science and the Cochrane Central Register of Controlled Trials) for English language research articles published between January 1998 and August 2018. The databases were systematically searched using a combination of the following keywords and Boolean operators: body composition OR body fat mass OR visceral fat mass OR subcutaneous fat mass OR adiposity OR visceral adipose tissue OR fat free mass OR muscle mass OR lean body mass OR skeletal muscle mass AND type 2 diabetes. The studies included observational epidemiological studies in T2DM populations such as cohort, case control and cross-sectional studies, in which DXA was utilized to assess BC in humans. This process continued until a search saturation point was found; i.e. the point at which additional terms showed no improvement in our search result. Relevant references identified from the bibliographies of pertinent articles were also retrieved. Study selection. Using our search strategy, 670 titles were initially identified. Two authors (PG and CL) assessed the titles independently according to predefined inclusion criteria. Studies were then systematically excluded after detailed examination, if the title and abstract were not relevant. Any potential disagreements were resolved through consulting the senior author (EL). If necessary, full-text articles of studies were also obtained, particularly if there was insufficient information within the abstract to determine exclusion.
Inclusion criteria. Eligibility criteria were based on the PICOS (participants, intervention, comparability, outcomes, study design) framework recommended by the PRISMA guidelines 23 .

Study Type:
We included observational studies (cross-sectional, case-control and prospective). 2. Participants: Studies that included human participants with T2DM. 3. Exposures: Four objective BC measurements (total BFM, VFM, SFM, and MM) obtained using DXA were selected as the exposures of interest. For MM, we included studies which reported total lean mass (TLM) or appendicular skeletal muscle mass (ASM) -two of the most commonly used measures of MM. Furthermore, we accepted studies reporting on BFM or MM and DM as either exposure/outcome. 4. Outcomes: Outcomes were the prevalence or incidence of T2DM. We accepted studies using different T2DM assessment methods, including but not limited to: random glucose ≥11.1 mmol/L, HbA1c ≥6.5% (≥48 mmol/moL), fasting plasma glucose >125 mg/dl or ≥7.0 mmol/l, self-reported use of oral hypoglycaemic medications or insulin, and/or history of physician-diagnosed diabetes.
Exclusion criteria. The following types of papers were excluded: 1. Conference abstracts. 2. Papers not written in the English language. 3. Studies on animals, and in-vitro/in-vivo studies. 4. Studies involving individuals with type 1 DM. 5. Studies in people without diabetes, or those with participants with impaired glucose tolerance or pre-diabetes. 6. Studies measuring biomarkers of BC in serum, blood, or urine without any link to BC assessment using DXA. 7. Exposures which were assessed using objective methods other than DXA, such as bioelectrical impedance analysis (BIA), MRI or CT. 8. Studies only measuring outcomes of "cardiometabolic diseases" or "metabolic risk factors" without specific reference to T2DM. 9. Articles with insufficient data to draw conclusions. This included any form of data insufficiency which did not enable us to draw conclusions from/evaluate the study, (e.g. lack of exposure/outcome definitions, or lack of statistical analysis). 10. For our meta-analysis, we also excluded studies that did not report the required statistical parameters and which we could not obtain despite repeated attempts to contact the study authors.

Data extraction.
A standardized data extraction form based on the "Strengthening the Reporting of Observational Studies in Epidemiology" (STROBE) statement 24 was used to extract the following relevant data from each included article: authors, year, study design, sample size, population characteristics, age of participants, objective BC measures, DM outcome type, method of T2DM diagnosis, adjustment for confounders used in analysis, statistical methods used, and summary of key findings. Data extraction was done by one author (PG) and confirmed by another (CL).
Statistical analysis. All statistical analyses were performed using Intercooled Stata version 15 for Windows (StataCorp, Lake Station, TX). Due to the paucity of studies reporting on the total BFM-T2DM relationship, and no studies on SFM, we could not perform meta-analysis of these BC measures and T2DM. We conducted random effects meta-analysis to pool the crude odds ratios (ORs) relating VFM and MM with T2DM, including studies that provided this information. ORs were calculated from two-way contingency tables where they were not reported but cell counts were available. We chose not to pool adjusted ORs as studies varied substantially in their adjustment for potential confounders which might have affected the direction and significance of the associations reported. Given that BC varies by gender, the meta-analysis of VFM-T2DM was stratified by gender. In contrast, no stratification was done for the meta-analysis of MM-T2DM as too few studies provided gender-stratified data. Heterogeneity was quantified using the I-squared statistic. A higher I-squared value meant greater heterogeneity in study effects. We reported all pooled estimates with 95% confidence intervals and judged a p-value < 0.05 as statistically significant.

Results
Description of studies. Of 670 titles screened, 120 were extracted for detailed evaluation, of which 20 adhered to our inclusion criteria ( Fig. 1). They comprised five prospective, eight cross-sectional, and seven case-control studies. The majority had high NOS scores, with 19 classified as "high quality" (≥7 stars) and one as "moderate quality" (5 stars). No study was excluded because of "low quality" (Tables 1 and 2).

Association of BFM and its components with T2DM.
Total BFM as the exposure. The association between total BFM and T2DM was equivocal [27][28][29][30] . In a cross-sectional study of Korean adults aged 50 years or older, Choi and colleagues reported increased total body fat % in both men (23.44 ± 4.91 vs 21.98 ± 5.23, p < 0.001) and postmenopausal women (34.85 ± 5.05 vs 33.85 ± 5.59, p < 0.001) with T2DM compared to those without 29 . In contrast, in a longitudinal study, Neeland and associates found no association between total BFM or body fat % and incident T2DM 30 . However, as both studies used simple univariate methods, the lack of multivariate adjustment for potential confounders, may bias their results. www.nature.com/scientificreports www.nature.com/scientificreports/ Total BFM as the outcome. We found only one cross-sectional study of 1315 Caucasians (mean age 58.5 ± 6.6), published study by Heska and colleagues, who reported that subjects with T2DM had reduced total BFM, compared to those without (34.5 vs 35.8 kg, p < 0.001) 20 . www.nature.com/scientificreports www.nature.com/scientificreports/ VFM as the exposure. We found evidence of a strong association between accumulation of VFM and prevalent T2DM, after adjusting for relevant confounders [31][32][33][34][35] . In a cross-sectional sample of 939 Italian men and women, VFM was independently associated with a two-fold risk of having T2DM, even after adjustment for BMI and WC 35 . Similarly, Nordstrom and colleagues demonstrated almost a four-fold (OR = 3.64, 95% CI: 2.53-5.25) and 1.5-fold (OR = 1.41, 95% CI: 0.93-2.13) higher odds of having T2DM in men and women for every SD increase in VFM/body weight, respectively 33 .
Subcutaneous Fat Mass (SFM) as the exposure or outcome. We found no published data on the relationship between SFM (assessed using DXA) and T2DM.
MM as the exposure. The association between MM and T2DM was equivocal 21,45,46 . For instance, in the Korean National Health and Nutrition Examination Survey (KNHANES), low muscle mass was found to be an early marker for T2DM (individuals with low ASM/wt had higher odds of T2DM (OR = 2.44, 95% CI: 1.69-3.53; p < 0.001) than those without) 21 . Conversely, in a fully-adjusted prospective, population-based, randomly selected cohort of men (mean age 54.1 ± 11.4 years), Li and associates found that reduced TLM and ASM/ht 2 were not significant risks factors for incident T2DM (per 5 kg increase in TLM and per 1 kg/m 2 increase in ASM/ ht 2 on T2DM incidence, OR = 1.03, 95% CI: 0.87-1.24; OR = 1.08: 95% CI: 0.83-1.39, respectively) 46 .

Discussion
In this systematic review and meta-analysis of the relationship between DXA-measured BC indices and T2DM, we found a consistent evidence of an association between high levels of visceral fat and the risk of T2DM, as well as a significant association between T2DM presence and low MM, after accounting for traditional diabetes risk factors. These findings suggest that reducing VFM and improving MM through lifestyle changes such as diet and physical activity, may reduce the risk of T2DM, and mitigate the deleterious effect of the condition on MM. In contrast, the relationship between total BFM, MM and T2DM, remains unclear, while no studies to date have explored the relationship between DXA-assessed SFM and T2DM. Longitudinal studies are however needed to better understand the temporality and mechanisms underlying the relationships between total BFM, SFM and MM; and T2DM.  www.nature.com/scientificreports www.nature.com/scientificreports/ Our finding that VFM may increase the risk of T2DM 31,33,35 could be explained by several potential underlying mechanisms. First, increased VFM is related to low adiponectin levels 47 . Adiponectin plays a pivotal role in energy metabolism, and has antiglycemic (increases insulin sensitivity), anti-inflammatory, antiangiogenic, and cardio protective properties 48,49 . Therefore, reduced secretion of adiponectin in individuals with high fat mass, particularly VFM, may lead to a cascade of biochemical reactions including an increase in insulin resistance and impaired glucose homeostasis (by augmenting hepatic gluconeogenesis and inhibiting glucose uptake in skeletal muscles) 50 resulting in hyperglycaemia. Second, lower adiponectin levels increases the secretion of pro-inflammatory cytokines (such as c-reactive proteins [CRP], interlukins and tumor necrosis factor [TNF] alpha) 51 , which have been linked to the pathogenesis of T2DM. Third, the accumulation of visceral fat has been suggested to have lipolytic potential. This may, in turn, result in an increased delivery of free fatty acid to the liver's portal circulation, which may induce hepatic insulin resistance by stimulating gluconeogenesis and interfering with hepatic insulin removal 50 .
Given that the relationship between total BFM and T2DM remain inconsistent, future studies are needed. While there is some evidence in studies using CT or MRI to measure subcutaneous fat 52 , data on the relationship between DXA-assessed SFM and T2DM are lacking. This is likely because previous versions of DXA were unable to compartmentalise SFM. Further studies using newer DXA models which can provide SFM in a large, population-based sample across the spectrum of BMI are warranted to assess the impact of SFM on the prevalence and incidence of T2DM. Studies unequivocally showed that people with T2DM were more likely to have lower MM [36][37][38][39][40][41][42][43] through several mechanisms, such as decreased glucose utilization by muscle 53 , increased levels of systemic inflammatory cytokines such as interlukin-6, TNF-alpha, and C-reactive protein 54,55 oxidative stress 56 , and mitochondrial dysfunction 57 . However, the relationship between reduced MM and risk of developing T2DM is still unclear 21,45,46 , and further studies are needed to determine this relationship and associated underpinning mechanisms.
There is substantial evidence indicating percent body fat differs between Asian and Western populations 58,59 , suggesting that the body composition profiles of patient with T2DM might also differ in these populations. As such, we also analysed various BC parameters separately in Asians and Caucasians. For BFM, most of the included studies (6 out of 8) were from Caucasians, limiting our ability to conclude whether objective fat measures differed between Asians and Caucasians. Of the 9 studies (4 Asians, 5 Caucasians) included in the systematic review of the DM-MM relationship, we found that T2DM was associated with low MM measures in both populations. However, there were only 3 studies (1 Asian, 2 Caucasians) for MM-DM relationship, with equivocal findings, suggesting the need for more studies to untangle this relationship in general, and between different populations.
There are several strengths of our systematic review. First, we only evaluated the relationship between BC measures and T2DM in human subjects, which allows for a more direct translation of results into clinical recommendations for patients with T2DM. Second, the studies in our review had wide geographic diversity, which can aid its generalizability. Third, we did not limit the timeframe, allowing a broad range of literature from 1998 to 2018. Finally, we excluded studies involving people with T1DM. This is important as there are pathophysiological, etiological, epidemiological and disease management differences between diabetes types which means that the influence of BC may differ between T1DM and T2DM.
While the majority of studies included in our review had high NOS scores, several limitations must be highlighted. First, eight studies were cross-sectional, which limit their ability to establish a causal relationship between BC and T2DM. The relatively low number of longitudinal studies (n = 5) included in the analysis, out of the 20 available in the literature, demonstrates the need for more well-designed cohort studies. Second, we excluded non-English publications which may have resulted in some publication bias. Third, as data from only eight articles were included in the meta-analysis, the generalizability of these findings might be limited. Fourth, the studies in our systematic review varied in their adjustment for potential confounders, thus we chose not to pool the adjusted odds ratio. Our meta-analysis summarised crude odds ratios, therefore our observed associations may be confounded by characteristics strongly related to T2DM risk that vary depending on body composition, such as diabetes duration and physical activity. Last, most studies only assessed a single BC parameter, and did not consider the interrelation between the key BFM measures (including both VFM and SFM) and MM. Future studies should assess the different BC phenotypes like low fat-high muscle mass, high fat-high muscle mass, low fat-low muscle mass, and high fat-low muscle mass to better reflect real world BC outcomes, which can then be translated into clearer clinical BC guidelines [60][61][62][63][64] .
In conclusion, our systematic review and meta-analysis demonstrated that higher BFM, particularly VFM, is associated with greater T2DM risk, and conversely the presence of T2DM was associated with a likelihood of lower MM. However, the relationship between MM (exposure) and T2DM remains unclear. Our findings suggest that optimal diabetes management and reducing VFM and increasing MM through lifestyle changes in the form of a more balanced diet and increased physical activity, may reduce the risk of T2DM and mitigate its deleterious impact on BC, respectively. However, further prospective studies to untangle the relationship between total BFM, SFM, and MM, on T2DM are needed in order to better inform clinical guidelines for disease prevention and management.