Body composition displays expected and unexpected contributions to blood pressure and glycemia in young South African women Authors: Ravi Retnakaran, Junwei Shen, Rayjean J Hung, Stephen J Lye, Shane A Norris Published: 2026 ABSTRACT AbstractIntroductionSub-Saharan Africa is undergoing rapid demographic, sociocultural, and economic transitions that are contributing to rising rates of overweight/obesity, which in turn may drive cardiometabolic sequelae such as diabetes and hypertension. However, the extent to which excess adiposity may contribute to the prevalence of such sequelae in sub-Saharan Africa remains uncertain. We thus sought to systematically evaluate body composition by dual-energy X-ray absorptiometry (DEXA) and its associations with glycemia and blood pressure (BP) in young women in Soweto, South Africa.Research design and methodsThis study was performed in 7735 women (median age of 22.4 years) undergoing assessment for a clinical trial. Whole-body DEXA scan yielded 12 measures of body composition (reflecting whole-body fat, central adiposity, peripheral adiposity, and lean mass/soft tissue) that were assessed as predictors of A1c, systolic BP, and diastolic BP, respectively.ResultsThe study population had high rates of overweight/obesity (49.6%), pre-diabetes/diabetes (10.9%), and hypertension (11.9%). While specific adiposity measures (fat mass (FM), visceral adipose tissue, subcutaneous adipose tissue (SAT), android fat) were significantly associated with A1c, these associations were exceedingly modest, with these measures additionally explaining only 0.14% to 0.37% of the variance in A1c. In contrast, all 12 body composition measures showed significant associations with systolic BP and diastolic BP. Indices of lean mass/soft tissue (lean mass index, lean mass, fat-free soft tissue mass) explained 5.8% to 6.3% of the variance in systolic BP, while measures of adiposity (android fat, SAT, FM) reconciled 4.6% to 4.8% of the variance in diastolic BP, reflecting much stronger associations with BP than with A1c.ConclusionIn this young adult population with high rates of overweight/obesity and pre-diabetes/diabetes, DEXA-derived adiposity exhibits modest associations with glycemia, underscoring the atypical nature of diabetes in sub-Saharan Africa.Trial registration numberPACTR201903750173871. WHAT IS ALREADY KNOWN ON THIS TOPIC Sub-Saharan Africa is experiencing rising rates of overweight/obesity, which may portend a growing future burden of cardiometabolic sequelae (eg, diabetes, hypertension). Sub-Saharan Africa is experiencing rising rates of overweight/obesity, which may portend a growing future burden of cardiometabolic sequelae (eg, diabetes, hypertension). WHAT THIS STUDY ADDS In this population of young South African women, there were high prevalence rates of overweight/obesity (49.6%), pre-diabetes/diabetes (10.9%), and hypertension (11.9%). In this population of young South African women, there were high prevalence rates of overweight/obesity (49.6%), pre-diabetes/diabetes (10.9%), and hypertension (11.9%). Direct measurement by whole-body dual energy X-ray absorptiometry (DEXA) revealed a negligible contribution of adiposity to A1c. Direct measurement by whole-body dual energy X-ray absorptiometry (DEXA) revealed a negligible contribution of adiposity to A1c. In contrast, DEXA-derived adiposity measures showed the expected associations with blood pressure. In contrast, DEXA-derived adiposity measures showed the expected associations with blood pressure. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY Adiposity exhibits modest associations with glycemia in young South African women but may be associated with the high burden of hypertension. Adiposity exhibits modest associations with glycemia in young South African women but may be associated with the high burden of hypertension. Introduction There is growing recognition that the rapidly evolving societal changes occurring across Africa may hold implications for the future burden of non-communicable diseases such as diabetes and cardiovascular disease.1 2 In particular, sub-Saharan Africa is undergoing demographic (expanding and aging population), sociocultural (changes in lifestyle and dietary habits), and economic (higher income, urbanization, greater food availability) transitions that are contributing to the emergence of overweight and obesity as risk factors that are expected to drive rising prevalence rates of diabetes and other cardiovascular risk factors, such as hypertension.1–3 Notably, from 1980 to 2015 the number of adults with overweight/obesity in sub-Saharan Africa quadrupled from 28 million to 127 million.1 Accordingly, the International Diabetes Federation has projected that, in the coming decades, sub-Saharan Africa is the region of the world that will experience the greatest increase in the prevalence of type 2 diabetes mellitus (T2DM).4 Although well established in other populations, the contribution of adiposity to the prevalence of diabetes in sub-Saharan Africa remains uncertain, as noted in a 2017 Lancet commission addressing this topic.1 Indeed, while some studies have suggested that diabetes may develop at lower body mass index (BMI) in individuals from Africa as compared with those of European descent,5 6 others have indicated a comparatively lower risk of diabetes in Africans for a given level of obesity.7 8 In the context of this uncertainty, we reasoned that detailed characterization of body composition and its associations with diabetes and other adiposity-associated cardiometabolic sequelae (such as hypertension) may be informative. Specifically, we hypothesized that the usual robust associations between adiposity and glycemia (as assessed by A1c) may not be present. Thus, our objective in this study was to systematically characterize body composition by whole-body dual energy X-ray absorptiometry (DEXA) and its associations with glycemia and blood pressure (BP) in young women residing in Soweto, South Africa. Methods The Healthy Life Trajectories Initiative (HeLTI) is a multidisciplinary, international consortium that is conducting four harmonized randomized controlled trials in South Africa,9 India,10 China,11 and Canada,12 respectively. The four HeLTI trials are each testing whether an integrated, culturally tailored, complex continuum of care intervention, starting prior to conception and (in those who become pregnant) continuing across gestation, infancy, and early childhood, can reduce adiposity in the child at 5 years of age (primary outcome of each trial). The current cross-sectional study was performed in the 7735 participants who underwent baseline assessment for the HeLTI South Africa trial, which is called BUKHALI (BUilding Knowledge and a foundation for HeALthy lIfe trajectories), as per the Zulu word meaning ‘smart/powerful.’9 The trial was registered with the Pan African Clinical Trial Registry (PACTR201903750173871) and received ethics approval from the Human Ethics Research Committee of the University of the Witwatersrand (Johannesburg, South Africa; M171137, M1811111), the Mount Sinai Hospital Research Ethics Board (Toronto, Canada; 19-0066 E), and the WHO Ethics Committee (ERC.0003328). All participants provided written informed consent, and the protocol has been previously described in detail.9 Study population and baseline assessment The study population for the BUKHALI trial has been recruited from Soweto, which is a historically disadvantaged urban area of Johannesburg wherein changes in the living environment in recent decades exemplify the rapid transitions affecting sub-Saharan Africa. As recently described in detail,13 the recruitment of the study population was specifically designed to achieve representativeness of young women aged 18–28 years living in Soweto, by recruiting from 30 random community clusters that were defined within Soweto by k-means clustering. Women were ineligible if recently diagnosed with any cancer, type 1 diabetes, and/or epilepsy given their care needs, or if they presented with cognitive impairment that could compromise their ability to provide informed consent. The recruitment process and derivation of the study population of 7735 women has been previously described (with Consolidated Standards of Reporting Trials (CONSORT) diagram).13 The current analysis was performed from data collected when these women attended their baseline assessment visit at the research center at Chris Hani Baragwanath Academic Hospital. These visits occurred between May 2019 and June 2023, with temporary suspension in 2020 owing to the COVID-19 pandemic. This baseline assessment was comprised of interviewer-administered questionnaires, physical examination (anthropometry, BP), phlebotomy, and whole-body DEXA scan. Height was measured to the nearest 0.1 cm with a wall-mounted stadiometer, and weight was measured to the nearest 100 g. Waist circumference was assessed halfway between the iliac crest in the mid-axillary plane and the lowest rib margin. Anthropometric measurements were performed in triplicate. Brachial BP was measured on the left arm with a validated automated oscillometric monitor (Omron; Kyoto, Japan) and appropriately sized cuff, following the International Society of Hypertension guidelines.14 Three measurements were taken at 2-minute intervals with the participant seated and resting throughout. The first reading was discarded, and the mean of the second and third readings was used for analysis. A1c was measured from whole blood by ion-exchange high-performance liquid chromatography on the Bio-Rad D-10 system. The analyzer was calibrated with manufacturer-supplied two-level calibrators following installation and priming of each new analytical cartridge. The assay traces to the reference methods of both the National Glycohemoglobin Standardization Program/Diabetes Control and Complications Trial and the International Federation of Clinical Chemistry. Within-run coefficients of variation (CV) were 0.88% and 0.56% at mean A1c 5.7% and 9.4%, respectively. Corresponding between-day CVs were 0.68% and 0.89%, and technician CV was <1.5%. Whole-body DEXA scans were performed by trained radiographers on a Hologic machine (Hologic Discovery W, software V.4.0.2), with whole-body calibration performed weekly. Instrument quality control (QC) checks were performed each day, with QC variation maintained below 1%. Complete data from DEXA scans were obtained from 6595 women. These scans enabled determination of sub-total (whole body minus head) fat mass (FM), visceral adipose tissue (VAT), subcutaneous adipose tissue (SAT), lean mass, and fat-free soft tissue mass. Trunk, arm, and leg fat were measured, with their respective percentages of FM providing regional measures of peripheral adiposity. Fat mass index and lean mass index were calculated by dividing the respective mass values (in kg) by height squared (in m2). Statistical analyses All analyses were carried out in R V.4.4.2, with p values adjusted for multiplicity via Holm’s method and multiplicity-adjusted p value <0.05 considered as statistically significant. Glycemic status was defined based on pre-existing diagnosis of diabetes, current A1c or random glucose, yielding the following three groups: (1) normal (A1c <6.0% and random glucose <6.1 mmol/L), (2) pre-diabetes (A1c 6.0% to 6.4% inclusive), and (3) diabetes (A1c ≥6.5% and/or random glucose ≥11.1 mmol/L or previous diagnosis of diabetes). BP status was defined based on either pre-existing diagnosis of hypertension or current BP, yielding the following three groups: (1) normal (systolic BP <120 mm Hg and diastolic BP <80 mm Hg), (2) elevated BP (systolic BP ≥120 and <140 mm Hg and/or diastolic BP ≥80 and <90 mm Hg), and (3) hypertension (systolic BP ≥140 mm Hg and/or diastolic BP ≥90 mm Hg or previous diagnosis of hypertension). The demographic and clinical characteristics (table 1) of the study population and DEXA body composition measures (table 2) of those with complete scans were summarized across both (1) glycemic groups and (2) BP groups. Normally distributed variables were summarized by means±SD; skewed variables were summarized by medians and IQRs; and categorical variables were summarized by frequencies and percentages. Spearman correlations of anthropometric and DEXA body composition measures with A1c, systolic BP, and diastolic BP were first assessed in the full study population (table 3) and then reassessed on sensitivity analyses in the participants who did not have diabetes or hypertension (online supplemental table 1). Table 1Demographic and clinical characteristics of the study population, stratified in two ways: (1) by glycemic status (normal, pre-diabetes, diabetes) and (2) by BP status (normal, elevated BP, hypertension)Glycemic statusBP statusNormal(n=6761)Pre-diabetes(n=417)Diabetes(n=413)Normal(n=4633)Elevated BP(n=2169)Hypertension(n=921)Age (years)22.4 (20.2, 24.7)22.0 (19.9, 24.5)22.6 (20.4, 24.9)22.0 (20.0, 24.3)22.6 (20.4, 24.9)23.8 (21.5, 25.8)Parity 0, n (%)3211 (47.8%)203 (48.9%)177 (43.1%)2276 (49.5%)1057 (49.1%)306 (33.3%) 1, n (%)2583 (38.5%)161 (38.8%)160 (38.9%)1746 (38.0%)789 (36.6%)430 (46.8%) ≥2, n (%)922 (13.7%)51 (12.3%)74 (18.0%)578 (12.6%)307 (14.3%)183 (19.9%)Employed Never, n (%)3824 (58.9%)267 (64.3%)242 (59.2%)2701 (60.2%)1252 (60.5%)458 (52.8%) Not currently, n (%)2135 (32.9%)120 (28.9%)133 (32.5%)1395 (31.1%)670 (32.4%)347 (40.0%) Currently, n (%)535 (8.2%)28 (6.7%)34 (8.3%)394 (8.8%)149 (7.2%)62 (7.2%)Government grant recipient, n (%)3175 (48.9%)194 (46.7%)212 (51.8%)2081 (46.3%)1050 (50.7%)513 (59.2%)Hazardous alcohol, n (%)1367 (20.3%)78 (18.8%)73 (17.8%)910 (19.7%)407 (18.8%)217 (23.6%)Current smoking, n (%)1469 (21.8%)93 (22.4%)88 (21.5%)1012 (21.9%)440 (20.4%)228 (24.9%)Recreational drugs, n (%)531 (7.9%)36 (8.7%)41 (10.0%)406 (8.8%)136 (6.3%)68 (7.4%)Anemia, n (%)1851 (27.5%)137 (32.9%)147 (35.6%)1414 (30.7%)520 (24.1%)216 (23.7%)HIV, n (%)497 (7.4%)30 (7.3%)35 (8.5%)364 (7.9%)137 (6.4%)67 (7.3%)Height (cm)158.6±6.2159.3±6.2159.0±6.4158.6±6.2158.7±6.3159.0±5.8BMI (kg/m2)24.9 (21.4, 29.8)25.1 (21.6, 29.9)25.2 (21.5, 30.6)23.8 (20.8, 27.9)26.6 (22.5, 31.9)28.5 (23.7, 34.9)BMI categories Underweight, n (%)458 (6.8%)31 (7.4%)31 (7.6%)398 (8.6%)100 (4.6%)27 (2.9%) Optimal, n (%)2955 (43.8%)175 (42.0%)167 (40.9%)2329 (50.4%)768 (35.6%)259 (28.2%) Overweight, n (%)1695 (25.1%)108 (25.9%)99 (24.3%)1124 (24.3%)579 (26.8%)227 (24.8%) Class I obesity, n (%)928 (13.8%)51 (12.2%)55 (13.5%)490 (10.6%)394 (18.3%)176 (19.2%) Class II obesity, n (%)462 (6.9%)31 (7.4%)37 (9.1%)189 (4.1%)214 (9.9%)131 (14.3%) Class III obesity, n (%)243 (3.6%)21 (5.0%)19 (4.7%)90 (1.9%)103 (4.8%)97 (10.6%)Waist circumference (cm)78 (70, 88)78 (70, 90)78 (70, 89)75 (68, 84)81 (72, 91)85 (75, 98)Central obesity, n (%)1564 (24.1%)115 (28.0%)103 (25.9%)776 (17.6%)668 (31.5%)381 (42.7%)Systolic BP (mm Hg)110.9±12.6108.6±11.5110.7±13.1104.4±8.1118.0±8.9127.1±15.4Diastolic BP (mm Hg)77.1±9.075.7±8.276.2±9.571.8±5.282.5±4.290.8±10.6A1c (%)5.1 (4.7, 5.4)6.1 (6.0, 6.3)7.0 (6.6, 7.6)5.2 (4.8, 5.6)5.2 (4.7, 5.6)5.1 (4.8, 5.5)Continuous variables are presented as mean±SD (if normal distribution) or median followed by IQR in parentheses (if skewed). Categorical variables are presented as n followed by percentage in parentheses.BMI, body mass index; BP, blood pressure. Demographic and clinical characteristics of the study population, stratified in two ways: (1) by glycemic status (normal, pre-diabetes, diabetes) and (2) by BP status (normal, elevated BP, hypertension) Continuous variables are presented as mean±SD (if normal distribution) or median followed by IQR in parentheses (if skewed). Categorical variables are presented as n followed by percentage in parentheses. BMI, body mass index; BP, blood pressure. Table 2DEXA body composition measures of participants with complete DEXA scans stratified by (1) glycemic status (normal, pre-diabetes, diabetes) and (2) BP status (normal, elevated BP, hypertension), respectivelyGlycemic statusBP statusNormal(n=5757)Pre-diabetes(n=390)Diabetes(n=383)Normal(n=3989)Elevated BP(n=1840)Hypertension(n=762)Android fat (g)1482 (930, 2326)1667 (977, 2510)1541 (949, 2458)1324 (855, 2017)1785 (1083, 2685)2052 (1258, 3065)Gynoid fat (g)4810 (3696, 6228)4854 (3782, 6332)4931 (3872, 6562)4478 (3532, 5751)5272 (4012, 6810)5551 (4371, 7334)VAT (cm2)59 (38, 92)64 (40, 100)64 (39, 101)53 (35, 81)69 (43, 105)82 (49, 122)SAT (cm2)296 (202, 433)328 (208, 449)312 (206, 450)269 (187, 386)349 (228, 481)387 (259, 531)FM (kg)23.9 (17.7, 32.5)24.6 (18.1, 33.6)24.3 (17.7, 35.0)22.0 (16.6, 29.5)27.0 (19.6, 36.5)29.2 (21.6, 40.1)Lean mass (kg)34.0±5.734.1±6.034.0±6.533.0±5.335.0±6.036.4±6.2FM index (kg/m2)9.5 (7.0, 12.9)9.8 (7.0, 13.1)9.8 (7.0, 13.7)8.7 (6.6, 11.8)10.8 (7.8, 14.5)11.7 (8.5, 15.9)Lean mass index (kg/m2)13.5±2.013.4±2.113.4±2.313.1±1.913.9±2.114.4±2.3Fat-free STM (kg)32.5±5.532.6±5.832.5±6.331.6±5.233.5±5.834.9±6.0Trunk fat % of FM (%)41.1±4.841.2±5.041.5±5.240.5±4.742.0±4.942.7±4.7Arm fat % of FM (%)11.2±1.711.3±1.611.3±2.111.1±1.711.4±1.611.6±1.7Leg fat % of FM (%)47.7±5.447.5±5.747.2±6.148.5±5.346.6±5.545.7±5.4Continuous variables are presented as mean±SD (if normal distribution) or median followed by IQR in parentheses (if skewed).BP, blood pressure; DEXA, dual-energy X-ray absorptiometry; FM, fat mass; SAT, subcutaneous adipose tissue; STM, soft tissue mass; VAT, visceral adipose tissue. DEXA body composition measures of participants with complete DEXA scans stratified by (1) glycemic status (normal, pre-diabetes, diabetes) and (2) BP status (normal, elevated BP, hypertension), respectively Continuous variables are presented as mean±SD (if normal distribution) or median followed by IQR in parentheses (if skewed). BP, blood pressure; DEXA, dual-energy X-ray absorptiometry; FM, fat mass; SAT, subcutaneous adipose tissue; STM, soft tissue mass; VAT, visceral adipose tissue. Table 3Spearman univariate correlations of anthropometric and DEXA body composition measures with A1c, systolic BP, and diastolic BP, respectivelyA1cSystolic BPDiastolic BPrP valuerP valuerP valueBMI (kg/m2)0.030.060.29 <0.001 0.27 <0.001 Waist circumference (cm)0.030.060.28 <0.001 0.27 <0.001 Android fat (g)0.06 <0.001 0.24 <0.001 0.27 <0.001 Gynoid fat (g)0.040.060.25 <0.001 0.26 <0.001 VAT (cm2)0.06 <0.001 0.22 <0.001 0.25 <0.001 SAT (cm2)0.05 0.003 0.24 <0.001 0.27 <0.001 Fat mass (kg)0.04 0.02 0.26 <0.001 0.27 <0.001 Lean mass (kg)0.030.180.28 <0.001 0.21 <0.001 Fat mass index (kg/m2)0.040.060.26 <0.001 0.27 <0.001 Lean mass index (kg/m2)0.010.400.29 <0.001 0.23 <0.001 Fat-free soft tissue mass (kg)0.030.180.28 <0.001 0.21 <0.001 Trunk fat % of fat mass (%)0.04 0.04 0.18 <0.001 0.22 <0.001 Arm fat % of fat mass (%)0.020.400.11 <0.001 0.12 <0.001 Leg fat % of fat mass (%)−0.04 0.04 −0.19 <0.001 −0.23 <0.001 P values are adjusted for multiplicity via Holm’s method.Bold indicates p<0.05.BMI, body mass index; BP, blood pressure; DEXA, dual-energy X-ray absorptiometry; SAT, subcutaneous adipose tissue; VAT, visceral adipose tissue. Spearman univariate correlations of anthropometric and DEXA body composition measures with A1c, systolic BP, and diastolic BP, respectively P values are adjusted for multiplicity via Holm’s method. Bold indicates p<0.05. BMI, body mass index; BP, blood pressure; DEXA, dual-energy X-ray absorptiometry; SAT, subcutaneous adipose tissue; VAT, visceral adipose tissue. Associations of each DEXA body composition measure with A1c, systolic BP, and diastolic BP and their explanatory capabilities were evaluated individually in a series of linear regression models. These models were adjusted for the following variables that could potentially affect body composition, glycemia, or BP: age, parity, employment status, hazardous alcohol consumption, current smoking, recreational drugs, anemia, and HIV positivity. Regression assumptions were evaluated by residual versus fitted plots, scale-location plots, normal quantile-quantile plots of residuals, and residuals versus leverage plots. To improve model assumptions, a natural log transformation was applied to A1c, systolic BP, and diastolic BP. Continuous predictors in the models were standardized to account for their varying scales. Missing data were addressed by multiple imputations: 30 imputed datasets were generated; predictive mean matching, logistic regression, and proportional odds models were used as the imputation models for missing continuous, binary, and ordinal variables, respectively, with the inclusion of all variables considered in the regression models along with those which were included in table 1 but not in the regression models. For anthropometric variables, only height, weight, and waist circumference were included in the imputation models, as other variables such as BMI were derived from these three variables. Pooled estimates of associations, 95% CIs, p values, model R2 statistics, and incremental R2 relative to the covariate-only models were reported in table 4. The linear regression analyses were also repeated in participants who did not have diabetes or hypertension (online supplemental table 2). Table 4Associations of body composition measures with (A) A1c, (B) systolic BP, and (C) diastolic BP, respectively, after adjustment for age, parity, employment, alcohol, smoking, recreational drugs, anemia, and HIV(A) A1cBeta95% CIP valueModel R2 (%)Incremental R2 (%)Total fat Fat mass (kg)0.0061(0.002 to 0.010) 0.03 0.510.14 Fat mass index (kg/m2)0.0051(0.001 to 0.009)0.110.470.10Central adiposity VAT (cm2)0.0101(0.006 to 0.014) <0.001 0.740.37 SAT (cm2)0.0067(0.003 to 0.011) 0.01 0.540.17 Android fat (g)0.0088(0.005 to 0.013) <0.001 0.660.29 Gynoid fat (g)0.0054(0.001 to 0.010)0.100.480.11Peripheral adiposity Trunk fat % of fat mass (%)0.0036(−0.001 to 0.008)0.850.430.06 Arm fat % of fat mass (%)−0.0010(−0.006 to 0.004)10.390.02 Leg fat % of fat mass (%)−0.0031(−0.008 to 0.002)0.970.410.04Lean mass/soft tissue Lean mass (kg)0.0010(−0.003 to 0.005)10.380.007 Lean mass index (kg/m2)−0.0014(−0.006 to 0.003)10.380.01 Fat-free soft tissue mass (kg)0.0010(−0.003 to 0.005)10.380.007(B) Systolic BPBeta95% CIP valueModel R2 (%)Incremental R2 (%)Total fat Fat mass (kg)0.023(0.021 to 0.026) <0.001 7.14.2 Fat mass index (kg/m2)0.022(0.019 to 0.024) <0.001 6.53.7Central adiposity VAT (cm2)0.019(0.017 to 0.022) <0.001 5.72.8 SAT (cm2)0.021(0.019 to 0.024) <0.001 6.33.5 Android fat (g)0.022(0.019 to 0.024) <0.001 6.43.6 Gynoid fat (g)0.021(0.019 to 0.024) <0.001 6.43.5Peripheral adiposity Trunk fat % of fat mass (%)0.010(0.006 to 0.013) <0.001 3.60.7 Arm fat % of fat mass (%)0.007(0.003 to 0.010) 0.004 3.20.3 Leg fat % of fat mass (%)−0.012(−0.015 to −0.009) <0.001 4.01.1Lean mass/soft tissue Lean mass (kg)0.029(0.026 to 0.031) <0.001 9.26.4 Lean mass index (kg/m2)0.028(0.025 to 0.030) <0.001 8.75.8 Fat-free soft tissue mass (kg)0.029(0.026 to 0.031) <0.001 9.26.3(C) Diastolic BPBeta95% CIP valueModel R2 (%)Incremental R2 (%)Total fat Fat mass (kg)0.025(0.022 to 0.028) <0.001 9.14.6 Fat mass index (kg/m2)0.024(0.022 to 0.027) <0.001 8.84.3Central adiposity VAT (cm2)0.022(0.019 to 0.025) <0.001 8.03.5 SAT (cm2)0.025(0.023 to 0.028) <0.001 9.14.6 Android fat (g)0.026(0.023 to 0.028) <0.001 9.34.8 Gynoid fat (g)0.023(0.020 to 0.026) <0.001 8.43.8Peripheral adiposity Trunk fat % of fat mass (%)0.012(0.009 to 0.015) <0.001 5.61.0 Arm fat % of fat mass (%)0.006(0.003 to 0.010) 0.004 4.80.3 Leg fat % of fat mass (%)−0.014(−0.017 to −0.011) <0.001 6.01.4Lean mass/soft tissue Lean mass (kg)0.021(0.019 to 0.024) <0.001 7.83.3 Lean mass index (kg/m2)0.022(0.019 to 0.025) <0.001 7.93.4 Fat-free soft tissue mass (kg)0.021(0.018 to 0.024) <0.001 7.83.3Beta estimate, 95% CI, multiplicity-adjusted p values, model R2 and incremental R2 relative to covariate-only models are reported.Bold indicates p<0.05.BP, blood pressure; SAT, subcutaneous adipose tissue; VAT, visceral adipose tissue. Associations of body composition measures with (A) A1c, (B) systolic BP, and (C) diastolic BP, respectively, after adjustment for age, parity, employment, alcohol, smoking, recreational drugs, anemia, and HIV Beta estimate, 95% CI, multiplicity-adjusted p values, model R2 and incremental R2 relative to covariate-only models are reported. Bold indicates p<0.05. BP, blood pressure; SAT, subcutaneous adipose tissue; VAT, visceral adipose tissue. Results The study population consisted of 7735 women, with a median age of 22.4 years (IQR 20.2–24.7) and featured high prevalence rates of overweight/obesity (49.6%), pre-diabetes/diabetes (10.9%), and hypertension (11.9%). There were 144 women with missing glycemic data and 12 with missing BP status, resulting in 7591 and 7723 women, respectively, who could be classified by these features. Table 1 shows the demographic and clinical characteristics of the study population when stratified in two ways: (1) by glycemic status (normal (n=6761), pre-diabetes (n=417), diabetes (n=413)) and (2) by BP status (normal (n=4633), elevated BP (n=2169), hypertension (n=921)). Of the 413 women with diabetes, only 13 had a pre-existing diagnosis thereof (and only 3 of whom were receiving diabetes medication). Age, parity, socioeconomic indicators (employment, receipt of government grant for social assistance), and substances (alcohol, smoking, recreational drugs) were similar across the glycemic groups. Anemia was more common in those with diabetes, and the prevalence of HIV positivity was comparable between the groups. Of note, waist circumference, central obesity, and BMI categories were similar across the groups. In contrast, many of these features appeared to differ across the BP strata (table 1). Of the 921 women with hypertension, 378 had a pre-existing diagnosis thereof (only 34 of whom were receiving anti-hypertensive medication). Women with higher BP were slightly older, more likely to be parous and recipients of government grants, and less likely to have anemia. Both BMI and waist circumference progressively increased from the normotensive group to elevated BP to those with hypertension, accompanied by the analogous patterns for weight categories and central obesity. Thus, it appeared that the clinical measures of adiposity tracked with BP groups but not glycemic status. We next assessed indices of body composition from whole-body DEXA scan (table 2). Intriguingly, all 12 DEXA measures assessing whole body fat, central adiposity, peripheral adiposity, and lean mass/soft tissue appeared comparable across glycemic groups. In contrast, all but one of the measures progressively increased from the normotensive group to elevated BP to hypertension, with the sole exception of leg fat percentage of FM, which showed the inverse progression. Correlations with continuous measures of A1c and BP We then evaluated Spearman univariate correlations of anthropometric and DEXA body composition measures with (1) A1c, (2) systolic BP, and (3) diastolic BP, respectively (table 3). While android fat, VAT, SAT, FM, trunk fat percentage, and leg fat percentage were associated with A1c by statistical significance (all multiplicity-adjusted p<0.05), it is notable that the correlations were exceedingly modest, with correlation coefficients all near 0 (the largest being 0.06). In contrast, all body composition measures showed significant associations with systolic BP and diastolic BP, respectively, with correlation coefficients in the range of ~0.2 to 0.3 (the sole inverse correlation with leg fat percentage of FM showing similar absolute magnitude). Moreover, these respective patterns of the Spearman correlations with A1c and systolic/diastolic BP persisted on sensitivity analyses that were restricted to the 6428 women who did not have diabetes or hypertension (online supplemental table 1). Adjusted associations with A1c and BP We then constructed a series of multiple linear regression models to individually evaluate the associations of each of the body composition measures with (dependent variables) A1c, systolic BP, and diastolic BP, respectively, after adjustment for age, parity, employment, hazardous alcohol consumption, current smoking, recreational drugs, anemia, and HIV positivity. As shown in table 4A, the A1c models revealed that specific adiposity measures (FM, VAT, SAT, and android fat) were statistically significantly associated with A1c. However, it should be recognized that these associations were exceedingly modest, as indicated by the small magnitudes of the effect estimates and the narrow CIs. Their incremental R2 values showed that these measures additionally explained only 0.14% to 0.37% of the variance in A1c beyond that provided by the adjustment variables. In contrast, as shown in table 4B, all 12 body composition measures showed statistically significant associations with systolic BP. Their incremental R2 showed that measures of lean mass/soft tissue (lean mass index, lean mass, and fat-free soft tissue mass) explained 5.8% to 6.3% of the variance in systolic BP (ie, explaining >10 times more than did the body composition predictors of A1c). Similarly, all 12 body composition measures showed statistically significant associations with diastolic BP (table 4C). As indicated by their incremental R2 values, measures of adiposity (android fat, SAT, FM) accounted for an additional 4.6% to 4.8% of the variance in diastolic BP relative to the adjustment variables. Thus, taken together, these multiple linear regression analyses showed that only certain adiposity measures (FM, VAT, SAT, android fat) related to A1c, and they explained <0.4% of its variance. In contrast, all 12 measures of body composition related to both systolic BP and diastolic BP. After adjusting for the same variables, the incremental proportion of the variance explained by these measures could be up to >10-times greater for BP than for A1c. Moreover, the robustness of these findings was confirmed on sensitivity analyses in which the models were repeated in the 6428 participants who did not have diabetes or hypertension (online supplemental table 2). Discussion In this study, we demonstrate three main findings. First, in this population of young South African women with a median age of 22.4 years, there were high prevalence rates of overweight/obesity (49.6%), pre-diabetes/diabetes (10.9%), and hypertension (11.9%). Second, among 12 DEXA-derived measures of body composition, only FM, VAT, SAT, and android fat showed significant associations with A1c. Third, and in contrast, body composition exhibited relationships to BP, with lean mass/soft tissue indices and adiposity measures emerging as predictors of systolic BP and diastolic BP, respectively. Taken together, these findings suggest that adiposity has modest associations with glycemia in young South African women but may be associated with the high burden of hypertension in this population. As highlighted by the 2017 Lancet commission on diabetes in sub-Saharan Africa,1 there is growing recognition that diabetes in African populations exhibits differences from that in other settings. For example, it has been suggested that the pathophysiology of T2DM in African patients may be distinct from that in other populations, with reduced hepatic insulin clearance posited as playing a comparatively greater role.3 Notably, as compared with White Europeans, Africans with normal glucose tolerance or pre-diabetes exhibit hyperinsulinemia that may be independent of differences in adiposity and insulin sensitivity.15 16 Similarly, in contradistinction to most other populations, the relationship between overweight/obesity and diabetes in the African context has yielded conflicting signals,1 5–8 with diabetes often presenting in lean individuals (BMI <25 kg/m2) who do not have islet-cell autoimmunity.3 17 18 In the past year, the International Diabetes Federation recognized type 5 diabetes as a distinct entity that presents with moderate-to-severe hyperglycemia in underweight patients (BMI <18.5 kg/m2) with a history of undernutrition early in life (previously described as malnutrition-associated diabetes).19 Moreover, an insulin-deficient, non-autoimmune form distinct from both classical type 1 diabetes and type 5 diabetes was recently reported in African children and young adults,20 further supporting the complexity of dysglycemic presentations in this setting.21 Against this backdrop, the current study adds an intriguing angle in revealing minimal association between DEXA-derived body composition and glycemia in young South African women. The robustness of these findings is supported by three key features. First, the study population of 7735 women reflects the full spectrum of adiposity and glycemia, with sizeable representation of all weight categories and glycemic groups (table 1). Second, in such a population, we show modest associations between DEXA-derived indices of body composition and glycemia. The biologic validity of the DEXA measures is supported by their clear relationships with BP and the emergence of VAT and android fat as predictors of A1c (although with low magnitude of association). Third, these findings were unchanged on sensitivity analyses restricted to participants without diabetes and hypertension, suggesting that these data may reflect the physiology of the study population (rather than the pathophysiology of the indicated conditions or the effects of treatment thereof). We are not aware of other populations in which DEXA-derived adiposity (whether visceral, subcutaneous, central, or peripheral) shows such modest association with A1c and glycemia. While some previous reports have suggested a comparatively lesser effect of VAT on glycemia in Black populations,3 22–24 its near-negligible contribution in the current study (explaining <0.4% of the variance in A1c) is striking (particularly in the absence of evidence of significant associations with other body composition measures). In a study in which 49.6% of participants had overweight/obesity, these findings raise the question of the biologic mechanisms that are enabling the maintenance of glucose tolerance in response to this presumed metabolic challenge. Though it would be reasonable to invoke beta-cell compensation as an explanation, it is notable that this minimal impact of adiposity is being observed in a population with 10.9% prevalence of pre-diabetes/diabetes (ie, individuals with beta-cell dysfunction by definition), suggesting the possibility of additional factors at play. It can be hypothesized that further study of the physiologic basis for the discordance between adiposity and glycemia in this population potentially could identify novel mechanisms for adaptation in response to the accrual of fat and its typical metabolic sequelae (although this possibility remains conjecture at this time). The relationship between overweight/obesity and higher BP has been well established in previous studies,25 26 and it has been suggested that sub-Saharan Africa may be disproportionately affected in this regard.27–29 Of note, two studies have shown associations of DEXA-derived measures of central adiposity with elevated BP and hypertension in middle-aged South African women and men.28 29 Importantly, the current study extends these observations to a much larger population of young women with a median age of 22.4 years, 11.9% of whom already meet the diagnosis of hypertension. The robust associations between measures of adiposity and higher BP observed at this young age raise the concern of a rising burden of early-onset hypertension in this population in the coming years, the importance of which is further underscored by the recognition that higher systolic BP has been previously identified as the largest contributor to global disability-adjusted life years.30 A limitation of this study is the cross-sectional design, which precludes definitive commentary on causality for the observed associations. In addition, the assessment of glycemic status by A1c can be confounded by both hemoglobin variants and iron deficiency (which can compromise the interpretation of A1c as a measure of glycemic exposure), although it should be noted that all models were adjusted for anemia. That said, determination of glycemic status by A1c (without fasting glucose or oral glucose tolerance test) carries a risk of misclassification. Similarly, the diagnosis of hypertension cannot be confirmed from BP measurement on a single occasion. The findings reported herein also cannot be extrapolated to men given that there are sex-specific differences in body composition,31 including higher percentage body fat in women from sub-Saharan Africa as compared with men.32 Another limitation is the absence of measures of islet autoimmunity, though the persistence of the findings after exclusion of those with diabetes argues against unrecognized autoimmune diabetes as the underlying explanation. Finally, the absence of measures of insulin sensitivity/resistance and beta-cell function limits our capacity to elucidate the pathophysiologic determinants of dysglycemia in this population. Conversely, however, the observed findings in such a large population of young women undergoing whole-body DEXA assessment in the setting of high prevalence rates of overweight/obesity and pre-diabetes/diabetes provide compelling evidence of minimal associations between body composition and glycemia in this population. Our findings hold clinical and research implications. They suggest that better understanding of the pathophysiology of diabetes in this setting may be needed to inform appropriate targeting of therapy. Indeed, research aimed at elucidating this pathophysiology may be revealing, particularly since the presenting features herein (no relationship to underweight or short stature; modest degree of glycemia) are not really suggestive of type 5 diabetes.19 The range of possibilities remains broad and potentially could include distinct genetic factors, early-life undernutrition/deprivation resulting in reduced pancreatic mass (although with more indolent beta-cell dysfunction than that of type 5 diabetes), unrecognized infections, or other environmental insults. Finally, the high rates of both pre-diabetes/diabetes and hypertension in such young women underscore the clinical imperative of early screening for these conditions in this at-risk population. In conclusion, in young South African women with high rates of overweight/obesity, dysglycemia, and hypertension, an array of 12 DEXA-derived measures of body composition showed minimal associations with A1c, suggestive of an atypical pathophysiology in line with growing recognition of the complexity and heterogeneity of diabetes presentations in sub-Saharan Africa. In contrast, there were clear associations between body composition and BP, with lean mass/soft tissue indices and adiposity measures emerging as predictors of systolic BP and diastolic BP, respectively. These findings thus suggest that adiposity has modest associations with glycemia in young South African women but may be associated with the high burden of hypertension in this population. Supplementary material