The epidemic of obesity and type 2 diabetes is evident in sub-Saharan Africa (SSA). However, their associations have hardly been examined in this region.
The epidemic of obesity and type 2 diabetes is evident in sub-Saharan Africa (SSA). However, their associations have hardly been examined in this region.
A hospital-based case–control study in urban Ghana consisting of 1221 adults (542 cases and 679 controls) investigated the role of anthropometric parameters for diabetes. Logistic regression was used for analysis. The discriminative power and population-specific cut-off points for diabetes were identified by receiver operating characteristic curves.
The strongest association with diabetes was observed for waist-to-hip ratio: age-adjusted odds ratios per 1 standard deviation difference were 1.95 (95% confidence interval [CI]: 1.64–2.31) in women and 1.40 [1.01–1.94] in men. Also, among women, the odds of diabetes increased with higher waist circumference (1.35 [1.17–1.57]) and waist-to-height ratio (1.29 [1.12–1.50]). Among men, this was not discernible. Rather, hip circumference was inversely related (0.69 [0.50–0.95]). Body mass index was neither associated with diabetes in women (1.01 [0.88–1.15]) nor in men (0.74 [0.52–1.04]). Among both genders, waist-to-hip ratio showed the best discriminative ability for diabetes in this population and the optimal cut-off points were ≥0.88 in women and ≥0.90 in men. Recommended cut-off points for body mass index and waist circumference had a poor predictive ability.
Our findings suggest that measures of central rather than general obesity relate to type 2 diabetes in SSA. It remains to be verified from larger population-based epidemiological studies whether anthropometric targets of obesity prevention in SSA differ from those in developed countries.
Examiner l'association entre l'obésité et le diabète de type 2 en Afrique subsaharienne (ASS).
Etude cas-témoins basée sur l'hôpital en milieu urbain au Ghana, portant sur 1221 adultes (542 cas et 679 témoins) investiguant sur l'incidence des paramètres anthropométriques (indice de masse corporelle [IMC], tour de taille, tour des hanches [HC], rapport ceinture-hanche et rapport ceinture-taille dans le diabète. La régression logistique a été utilisée pour l'analyse. Le pouvoir discriminant et des seuils spécifiques aux populations pour le diabète ont été identifiés par des courbes caractéristiques «receiver operating».
La plus forte association avec le diabète a été observée pour le rapport ceinture-hanches: les rapports de cotes (odds ratios) ajustés pour l’âge par une différence de 1 écart-type étaient de 1,95 (intervalle de confiance [IC] à 95%: 1,64 à 2,31) chez les femmes et 1,40 [1,01 à 1,94] chez les hommes. De plus, chez les femmes, les chances de diabète augmentent avec un tour de taille (1,35 [1,17 à 1,57]) et un rapport ceinture-taille (1,29 [1,12 à 1,50]) plus élevés. Chez les hommes, cela n’était pas perceptible. Au contraire, le tour de hanches était inversement associé (0,69 [0,50 à 0,95]). L’IMC n’était ni associé au diabète chez les femmes (1,01 [0,88 à 1,15]), ni chez les hommes (0,74 [0,52 à 1,04]). Chez les deux sexes, le rapport ceinture-hanche a montré la meilleure capacité discriminative pour le diabète dans cette population et les seuils optimaux étaient de 0,88 chez les femmes et 0,90 chez les hommes. Les seuils recommandés pour l’IMC et le tour de taille ont une faible capacité prédictive.
Nos résultats suggèrent que les mesures de l'obésité centrale plutôt que générale se rapportent au diabète de type 2 en ASS. Il reste à vérifier dans de grandes études épidémiologiques de population si les objectifs anthropométriques de la prévention de l'obésité en ASS diffèrent de ceux des pays développés.
Examinar las asociaciones entre la obesidad y la diabetes tipo 2 en África subsahariana (ASS).
Estudio hospitalario de casos y controles en zonas urbanas de Ghana con 1221 adultos (542 casos y 679 controles) para investigar el papel de los parámetros antropométricos (índice de masa corporal (IMC), circunferencia de la cintura, circunferencia de la cadera [CC], el índice cintura-cadera y el índice de cintura-altura en diabetes. El análisis se realizó mediante una regresión logística. El poder discriminatorio y los puntos de corte poblacionales específicos para diabetes se identificaron mediante curvas ROC características.
La mayor asociación con diabetes se observaba con el índice cintura-cadera, la razón de probabilidades ajustada por edad con una desviación estándar de diferencia era de 1.95 (IC 95%: 1.64–2.31) en mujeres y 1.40 [1.01–1.94] en hombres. También, entre mujeres, la probabilidad de diabetes aumentaba con una mayor circunferencia de la cintura (1.35 [1.17–1.57]) y el índice cintura-altura (1.29 [1.12–1.50]). Entre hombres esta relación no era discernible. Más bien la circunferencia de la cadera estaba inversamente relacionada (0.69 [0.50–0.95]). El IMC no estaba asociado ni con la diabetes en mujeres (1.01 [0.88–1.15]) ni en hombres (0.74 [0.52–1.04]). En ambos géneros, el índice cintura-cadera mostró la mayor habilidad discriminatoria para diabetes en esta población y los puntos de corte óptimos eran ≥0.88 en mujeres y ≥0.90 en hombres. Los puntos de corte recomendados para IMC y la circunferencia de la cintura tenían una capacidad vaticinadora pobre.
Nuestros hallazgos sugieren que las medidas de obesidad central, más que las de obesidad general están relacionadas con la diabetes tipo 2 en ASS. Aún ha de verificarse, mediante estudios epidemiológicos en poblaciones más amplias, si los objetivos antropométricos de prevención de la obesidad en ASS son diferentes a aquellos en países desarrollados.
The prevalence of type 2 diabetes is rising rapidly in sub-Saharan Africa (SSA), which is mainly attributed to growing rates of obesity, urbanisation and insufficient physical activity (Mbanya et al. 2010). Obesity is the major risk factor for type 2 diabetes, and particularly in urban areas of developing countries, its prevalence is increasing dramatically (Ziraba et al. 2009). It is still controversial which measure of overweight or obesity best reflects an increased risk of diabetes among non-Caucasian populations (Qiao & Nyamdorj 2010a). Standardised measures of obesity such as body mass index (BMI) and waist circumference (WC) have been derived mainly from studies among Caucasians and are therefore considered inappropriate for other populations (Misra et al. 2005; Lear et al. 2007, 2010). In fact, Asian (Nishida et al. 2010; Katzmarzyk et al. 2011) and black populations (Chiu et al. 2011) might bear higher risks of diabetes at lower BMI levels than Caucasians. In Asians and African Americans, central obesity measures including WC, waist-to-hip ratio (WHR) and waist-to-height ratio (WTHR) are better discriminators of diabetes than BMI (Huxley et al. 2008; Nyamdorj et al. 2008; Mackay et al. 2009). This could also be the case with respect to population-specific cut-off points (Lear et al. 2010; Qiao & Nyamdorj 2010b). It has been suggested that this may be due to ethnic differences in the association between BMI, body fat distribution and cardiometabolic risk factors. Also, several aspects of body composition are known to differ between Caucasians and Africans (Wagner & Heyward 2000; Katzmarzyk et al. 2010). Therefore, the ability of the known obesity measures to predict type 2 diabetes in people from SSA may differ from that in other ethnic groups.
Given the limited number of studies on this topic (Fisch et al. 1987; Aspray et al. 2000; Balde et al. 2007; Giday 2010), further investigation is needed to examine the association of obesity indices with diabetes in people from SSA. This may facilitate enhanced screening for diabetes risk factors and provide the basis for improved preventive strategies. We aimed to compare the association between different obesity measures and diabetes and the discriminative power of these measures in an urban Ghanaian population. We also determined the optimal BMI, WC and WHR cut-off points for the identification of diabetes in our study population.
The Kumasi Diabetes and Hypertension Study is an unmatched case–control study conducted between August 2007 and June 2008 at Komfo Anokye Teaching Hospital (KATH), Kumasi, Ghana (Danquah et al. 2012). The primary aim of the study was to identify factors associated with diabetes and/or hypertension in urban Ghana. A detailed description of the examination procedures and the participants' characteristics has been provided elsewhere (Danquah et al. 2012). Briefly, patients attending the diabetes centre (n = 495) and the hypertension clinic (n = 451) were recruited. These participants were encouraged to promote participation in the study among their community members resulting in subsequent recruitment of control participants (n = 222). Further control participants came from the outpatient department (n = 150) and hospital staff (n = 148). In all participants, concentrations of fasting plasma glucose (FPG) (fluoride plasma, tubes cooled at +4 °C) were measured photometrically (Glucose 201+ Analyzer; HemoCue, Ångelholm, Sweden). Cases were defined as patients with documented antidiabetic medication or FPG ≥7 mm (WHO 1999). Controls were defined as participants without diabetes.
All participants (n = 1466) provided written informed consent; the study protocol was approved by the Ethics Committee, University of Science and Technology, Kumasi. For the current analysis, we excluded 245 participants with incomplete information on anthropometry, nutrition behaviour, genetic variants or socio-economic status, resulting in a final sample size of 1221 participants (Figure 1). Excluded participants did not differ in key characteristics (age, gender, anthropometric measures) from those included in the study.
All participants were instructed to have an overnight fast. Anthropometric measurements were taken in the standing position by a trained nurse. The participants wore light clothes without shoes. Weight was measured on an electronic person scale to the nearest 0.1 kg and height with a stadiometer to the nearest 0.1 cm. WC was determined two fingers' breadth below the lowest rib using a measuring tape and hip circumference (HC) at the level of the widest diameter around the gluteal protuberance (all devices, Seca, Germany). Body composition was assessed by bioelectric impedance (BIA) (50 kHz; Nutrigard-S, NutriPlus 1.0; Data Input Germany).
BMI was calculated as weight/(height)2 [kg/m²]. Overweight was defined as BMI 25.0–29.9 kg/m² and obesity as BMI ≥30.0 kg/m². WC cut-offs corresponding to BMI-defined overweight were added in a sensitivity analysis, that is, WC ≥80 cm in women and ≥94 cm in men. WHR was calculated as WC/HC and WTHR as WC/height. Central obesity was defined as WC ≥88 cm in women and ≥102 cm in men. Alternatively and applied for the identification of the optimal cut-off point, central obesity was defined as WHR ≥0.85 for women and ≥0.90 for men (each WHO 2011).
The participants underwent a routine clinical examination. Blood pressure and heart rate were measured in triplicates on a comfortable chair after 10 min resting time in an air-conditioned room (M8 Comfort, Omron, Japan). Data on general demographics, medical history and socio-economic status were documented. Nutrient intake (g/1000 kcal) was assessed by 24-h dietary recall and was calculated applying local nutrient databases (FAO 1975). An interview on physical activity included way to work, work activities and leisure activities (min/week). Daily energy expenditure (kcal) was calculated as the sum of metabolic equivalents (ml/kg/min) × body weight (kg) × duration (min) (Ainsworth et al. 1993). Smoking status (never, quit, current) and educational attainment (none, primary, secondary, tertiary, other) were self-reported. Also, Plasmodium falciparum infection was included as a potential confounder (Danquah et al. 2010). Detection of P. falciparum was performed by primer-specific PCR (Snounou et al. 1993).
All analyses were performed separately for women and men. Between-group differences in socio-demographic and anthropometric characteristics were assessed by Mann–Whitney U-test for continuous variables and by chi-square test for categorical variables. The relationships between anthropometric measures were investigated using age-adjusted Spearman correlations.
To evaluate associations between anthropometric measures and diabetes risk, the measures of interest (BMI, WC, HC, WHR and WTHR) were categorised into quintiles based on their distribution among controls. We then estimated the odds ratios (ORs) and 95% confidence intervals (CIs) for the comparisons of quintiles using logistic regression. The lowest quintile served as the reference category among women and, for sample size reasons, the two lowest quintiles among men. The trend test across categories used the median value among controls for each category as a continuous exposure variable. We also estimated ORs and 95% CIs per 1 standard deviation (SD) difference in anthropometric measures.
Initially, we adjusted for age (model 1); then, we added smoking status (never/current or ex-smoker), family history of diabetes (yes/no), educational attainment (any/none), fat and fibre intake (g/1000 kcal) and energy expenditure (kcal) (model 2). In the final model, we further included BMI (model 3). In a sensitivity analysis, we additionally entered systolic and diastolic blood pressure and P. falciparum infection into model 2.
As a secondary analysis, the discriminative abilities of anthropometric measures for identifying diabetes cases were compared by means of ROC curve analysis. The ROC area under the curve (AUC) estimated the discriminative capabilities of those anthropometric measures associated with diabetes.
Finally, sensitivity and specificity of sex-specific cut-off points for BMI, WC and WHR recommended by the WHO were estimated using ROC analysis. The Youden Index was computed to identify population-specific cut-off points of these measures for the optimal differentiation between cases and controls. The Youden Index is derived from (sensitivity + specificity) −1 and ranges from 0 to 1 (Youden 1950).
A P-value of < 0.05 was used to ascertain statistical significance. All analyses were performed using SAS statistical software (version 9.2; SAS Institute, Cary, NC, USA).
The characteristics of 1221 participants included in the analysis are presented in Table 1. Of the 542 cases, 97% have already been known to the diabetes centre or the hypertension clinic (mean time since diagnosis, 6.5 ± 5.8 years). The majority were on metformin-based medication or on combination therapy with sulphonylureas (Danquah et al. 2012). Compared to controls (n = 679), cases were on average older and were more likely to have a family history of diabetes. There was no difference in residence between women with diabetes and those without, whereas men with diabetes were less likely than unaffected men to live in the Kumasi metropolitan area. Men and women with diabetes were more frequently unemployed, illiterate and without formal education than controls. WC, WHR and WTHR were all higher in women with diabetes than in controls. This was observed only for WHR in men (Table 1).
|Characteristics||Women (922)||Men (299)|
|Controls (523)||Diabetes cases (399)||Controls (156)||Diabetes cases (143)|
|Age (years)||47.0 ± 15.7||54.6 ± 13.1a||46.2 ± 16.0||55.1 ± 14.5a|
|Ethnic group (Akan)||451 (86.2)||347 (87.0)||133 (85.3)||127 (88.8)|
|Residence (Kumasi metropolitan)||397 (75.9)||285 (71.4)||120 (76.9)||94 (65.7)a|
|Smoking status (ever)||n/ab||n/ab||27 (17.3)||37 (25.9)|
|Family history of diabetes||141 (27.0)||245 (61.4)a||30 (19.2)||72 (50.4)a|
|Formal education (none)||102 (19.5)||168 (42.1)a||11 (7.1)||23 (16.1)a|
|Literacy (illiterate)||159 (30.4)||218 (54.6)a||18 (11.5)||31 (21.7)a|
|Unemployed||89 (17.0)||152 (38.1)a||21 (13.5)||40 (28.0)a|
|Body mass index (kg/m²)||26.5 ± 5.6||26.8 ± 5.1||23.5 ± 3.9||23.1 ± 3.7|
|Prevalent overweightc||164 (31.4)||155 (38.9)||40 (25.6)||34 (23.8)|
|Prevalent obesityd||131 (25.1)||93 (23.3)||7 (4.5)||7 (4.9)|
|Waist circumference (cm)||87.0 ± 13.4||92.3 ± 12.0a||84.6 ± 11.9||86.6 ± 11.0|
|Prevalent central obesitye||258 (49.3)||262 (65.7)a||13 (8.3)||11 (7.7)|
|Hip circumference (cm)||101.6 ± 11.0||101.8 ± 10.7||95.5 ± 8.6||93.9 ± 8.2|
|Waist-to-hip ratio||0.85 ± 0.07||0.91 ± 0.07a||0.88 ± 0.08||0.92 ± 0.07a|
|Waist-to-height ratio||0.55 ± 0.08||0.58 ± 0.07a||0.50 ± 0.07||0.51 ± 0.06|
Gender-specific partial correlation coefficients for different anthropometric measures, controlling for age, are given in Table 2. BMI, WC, HC and WTHR were generally strongly correlated in both genders. WHR showed a comparatively weaker correlation with the other anthropometric measures.
We evaluated associations with diabetes comparing quintiles for anthropometric measures stratified by gender. Among women (Table 3), BMI and HC were not associated with diabetes. Comparing the highest with the lowest quintile, the multivariate-adjusted ORs for diabetes were 2.18 (1.19–4.00) for WC, 5.06 (2.70–9.46) for WHR and 2.77 (1.46–5.25) for WTHR. The strength of association generally increased after adjustment for BMI. Similarly to the quintile-based analysis, WHR showed the strongest association with diabetes per 1 SD difference (1.98 [1.63–2.42]) followed by WC (1.33 [1.11–1.60]) and WTHR (1.27 [1.06–1.51]). Further adjustment for BMI strengthened these associations.
|Women (n = 922)||Odds ratios (95% CI) for quintiles||P for Trenda||Odds ratio for 1 SD|
|Quintile 1||Quintile 2||Quintile 3||Quintile 4||Quintile 5|
|No. of cases/controls||65/105||71/104||95/104||94/105||74/105|
|Model 1: age adjusted||1.00||1.07 (0.68–1.68)||1.29 (0.84–2.00)||1.27 (0.82–1.96)||1.02 (0.66–1.60)||0.85||1.01 (0.88–1.15)|
|Model 2: multivariate adjusted||1.00||1.14 (0.67–1.93)||1.34 (0.81–2.22)||1.21 (0.73–2.00)||0.80 (0.46–1.38)||0.38||0.92 (0.77–1.09)|
|No. of cases/controls||32/109||64/102||71/103||122/106||110/103|
|Model 1: age adjusted||1.00||1.68 (0.99–2.83)||1.63 (0.97–2.74)||2.65 (1.61–4.36)||2.41 (1.46–3.98)||< 0.001||1.35 (1.17–1.57)|
|Model 2: multivariate adjusted||1.00||1.69 (0.93–3.05)||1.60 (0.88–2.91)||2.56 (1.44–4.53)||2.18 (1.19–4.00)||0.006||1.33 (1.11–1.60)|
|Model 3: multivariate + BMI adjusted||1.00||2.41 (1.30–4.46)||3.29 (1.67–6.48)||6.81 (3.32–13.94)||10.51 (4.21–26.23)||< 0.001||3.63 (2.50–5.27)|
|No. of cases/controls||72/105||76/105||94/106||71/103||86/104|
|Model 1: age adjusted||1.00||0.87 (0.56–1.36)||1.12 (0.73–1.72)||0.84 (0.54–1.31)||1.02 (0.66–1.57)||0.97||0.98 (0.85–1.12)|
|Model 2: multivariate adjusted||1.00||0.79 (0.47–1.32)||1.06 (0.65–1.73)||0.74 (0.44–1.26)||0.79 (0.47–1.34)||0.39||0.88 (0.74–1.04)|
|Model 3: multivariate + BMI adjusted||1.00||0.82 (0.48–1.41)||1.14 (0.65–2.00)||0.83 (0.43–1.61)||0.94 (0.41–2.13)||0.90||0.80 (0.57–1.13)|
|No. of cases/controls||23/104||42/106||67/104||87/104||180/105|
|Model 1: age adjusted||1.00||1.40 (0.78–2.54)||2.15 (1.22–3.80)||2.60 (1.48–4.57)||5.01 (2.89–8.70)||< 0.001||1.95 (1.64–2.31)|
|Model 2: multivariate adjusted||1.00||1.33 (0.69–2.56)||2.37 (1.25–4.46)||3.03 (1.59–5.76)||5.06 (2.70–9.46)||< 0.001||1.98 (1.63–2.42)|
|Model 3: multivariate + BMI adjusted||1.00||1.42 (0.73–2.74)||2.70 (1.42–5.12)||3.71 (1.92–7.19)||6.36 (3.32–12.17)||< 0.001||2.20 (1.78–2.72)|
|No. of cases/controls||25/104||81/104||72/105||112/105||109/105|
|Model 1: age adjusted||1.00||2.44 (1.41–4.20)||1.88 (1.08–3.27)||2.87 (1.68–4.92)||2.64 (1.54–4.55)||0.002||1.29 (1.12–1.50)|
|Model 2: multivariate adjusted||1.00||2.97 (1.61–5.49)||2.16 (1.15–4.07)||3.19 (1.73–5.88)||2.77 (1.46–5.25)||0.015||1.27 (1.06–1.51)|
|Model 3: multivariate + BMI adjusted||1.00||4.35 (2.29–8.28)||4.40 (2.15–9.04)||8.56 (3.96–18.52)||12.28 (4.76–31.72)||< 0.001||3.20 (2.23–4.59)|
Among men (Table 4), BMI and WC were not linked to diabetes. Comparing the highest with the lowest quintile, the multivariate-adjusted odds of diabetes were 0.41 (0.17–0.98) for HC and 3.12 (1.31–7.38) for WHR. BMI adjustment strengthened these associations. Similarly to the quintile-based analysis, WHR showed a positive association with diabetes per 1 SD difference (1.71 [1.19–2.46]) and HC was inversely linked (0.65 [0.44–0.97]). Further adjusting for BMI strengthened these associations. WTHR was associated with diabetes in the multivariate-adjusted model in quintile 3 and 4. After adjustment for BMI, the association became significant across all quintiles (5th vs. 1st quintile: 6.43 [1.24–33.36]) and per 1 SD difference (3.05 [1.34–6.98]).
|Men (n = 299)||Odds ratios (95% CI) for quintiles||P for Trenda||Odds ratio for 1 SD|
|Quintile 1||Quintile 2||Quintile 3||Quintile 4||Quintile 5|
|No. of cases/controls||61/63||30/30||32/32||20/31|
|Model 1: age adjusted||1.00||0.99 (0.52–1.89)||0.77 (0.41–1.47)||0.51 (0.25–1.01)||0.07||0.74 (0.52–1.04)|
|Model 2: multivariate adjusted||1.00||1.23 (0.57–2.66)||0.79 (0.37–1.72)||0.64 (0.28–1.47)||0.37||0.88 (0.58–1.34)|
|No. of cases/controls||40/63||36/31||44/31||23/31|
|Model 1: age adjusted||1.00||1.29 (0.66–2.49)||1.25 (0.64–2.45)||0.66 (0.32–1.37)||0.55||0.92 (0.68–1.23)|
|Model 2: multivariate adjusted||1.00||1.59 (0.73–3.47)||1.86 (0.82–4.21)||0.85 (0.35–2.09)||0.82||1.02 (0.71–1.47)|
|Model 3: multivariate + BMI adjusted||1.00||1.82 (0.77–4.28)||2.45 (0.81–7.47)||1.35 (0.30–6.16)||0.22||1.74 (0.81–3.76)|
|No. of cases/controls||72/64||30/30||19/33||22/29|
|Model 1: age adjusted||1.00||0.78 (0.41–1.47)||0.44 (0.22–0.88)||0.51 (0.26–1.01)||0.01||0.69 (0.50–0.95)|
|Model 2: multivariate adjusted||1.00||0.84 (0.38–1.82)||0.27 (0.12–0.64)||0.41 (0.17–0.98)||0.009||0.65 (0.44–0.97)|
|Model 3: multivariate + BMI adjusted||1.00||0.59 (0.25–1.37)||0.15 (0.05–0.41)||0.14 (0.04–0.52)||< 0.001||0.36 (0.18–0.72)|
|No. of cases/controls||29/63||23/30||37/31||54/32|
|Model 1: age adjusted||1.00||1.22 (0.59–2.55)||1.75 (0.87–3.53)||2.01 (0.97–4.16)||0.015||1.40 (1.01–1.94)|
|Model 2: multivariate adjusted||1.00||1.36 (0.56–3.29)||2.21 (0.96–5.13)||3.12 (1.31–7.38)||0.002||1.71 (1.19–2.46)|
|Model 3: multivariate + BMI adjusted||1.00||1.47 (0.60–3.62)||3.07 (1.26–7.45)||5.75 (2.08–15.89)||< 0.001||2.17 (1.42–3.32)|
|No. of cases/controls||33/63||40/31||43/30||27/32|
|Model 1: age adjusted||1.00||1.81 (0.93–3.51)||1.66 (0.84–3.28)||0.87 (0.42–1.81)||0.97||0.92 (0.67–1.25)|
|Model 2: multivariate adjusted||1.00||3.20 (1.41–7.26)||3.11 (1.35–7.16)||1.50 (0.61–3.69)||0.20||1.14 (0.78–1.65)|
|Model 3: multivariate + BMI adjusted||1.00||4.84 (1.94–12.11)||7.18 (2.23–23.05)||6.43 (1.24–33.36)||0.001||3.05 (1.34–6.98)|
We evaluated whether additional adjustment for blood pressure and P. falciparum infection confounded these associations, but only minor changes in the odds ratios were observed (<4.5% among women and <4.0% among men).
As BMI was not associated with diabetes, we speculated that diabetic atrophy in cases with poor glycaemic control (FPG <7 mm) may be responsible. Among women, mean BMI (± SD) was significantly lower in cases with FPG ≥7 mm (26.1 ± 5.4 kg/m2) as compared to cases with good glycaemic control (27.3 ± 4.8 kg/m2; P = 0.01). In men, these figures were 22.5 ± 3.9 kg/m2 and 23.7 ± 3.4 kg/m2, respectively (P = 0.04). The same pattern was observed for mean body cell mass, assessed by BIA, in women (20.1 ± 3.3 vs. 21.0 ± 3.4 kg; P = 0.02) and in men (25.0 ± 5.4 vs. 27.5 ± 5.2 kg; P = 0.005). However, the lack of association between BMI and diabetes persisted after the exclusion of cases under poor glycaemic control (data not shown). This was similar when using FPG ≥7.8 mm as the cut-off value.
We further assessed the discriminative power of selected anthropometric measures for identifying diabetes by ROC-AUC comparisons (Figure 2). In women, WHR (ROC-AUC: 0.785 [95% CI: 0.756–0.815]) was the best obesity measure for discriminating diabetes, followed by WC (0.762 [0.731–0.793], P = 0.0013) and WTHR (0.759 [0.728–0.790], P = 0.0003). Among men, HC and WHR had similar discriminative power: HC (0.749 [0.694–0.804]) and WHR (0.753 [0.698–0.808]) P = 0.80).
Lastly, we calculated sensitivity and specificity of recommended cut-off points for the identification of diabetes and assessed the optimal cut-offs for BMI, WC and WHR with the Youden Index (Table 5). Using BMI ≥25 kg/m2, 62% of the cases and 44% of the controls were correctly classified in women. Sensitivity was considerably lower among men (29%), at higher specificity (70%). Recommended WC cut-off points identified diabetes well in women (high sensitivity), but not in men. The sensitivity and specificity for WHR were >60% in both sexes. Compared to the recommended cut-off points, the optimal BMI cut-offs were slightly higher in both women and men. The optimal cut-off point for WC was higher than the recommended cut-offs for both, overweight and obesity, in women (91.7 vs. 80.0/88.0 cm) and lower in men (83.4 vs. 94.0/102.0 cm). For WHR, the population-specific cut-off point in women (0.88) exceeded the reference value (0.85). These values were identical for men (0.90). In a sensitivity analysis, we excluded cases with poor glycaemic control. Only for BMI, the optimal cut-off point changed markedly, decreasing from 26.2 to 25.6 kg/m2 among women and from 26.7 to 20.7 kg/m2 among men.
|Variable||Women (n = 922)||Men (n = 299)|
|Sensitivity (%)||Specificity (%)||Sensitivity (%)||Specificity (%)|
|Optimal cut-off:a 26.2||54.6||53.9||26.7||86.0||21.8|
|Optimal cut-off:a 91.7||54.1||64.4||83.4||60.8||54.5|
|Optimal cut-off:a 0.88||65.7||62.0||0.90||63.6||61.5|
This study investigated the association between different obesity measures and diabetes in urban Ghana. We found that measures of central, but not of general obesity, were associated with diabetes in women and men. Specifically, BMI was not linked to diabetes, while WHR showed the strongest association in both sexes. Also, WHR showed the best discriminative ability for diabetes; a cut-off point of ≥0.88 in women and ≥0.90 in men were the optimal cut-off points in this SSA population.
So far, only a few studies have determined the association between obesity measures and diabetes in SSA. These studies have been conducted in different regions of SSA, applied varying obesity measurements and diabetes ascertainment and included diverse study types with a wide range of sample sizes. Consequently, their analyses are heterogeneous and results difficult to compare. For instance, BMI was not consistently associated with diabetes and associations differed between women and men. Also, the magnitude of association diverges expansively (Fisch et al. 1987; Aspray et al. 2000; Balde et al. 2007; Giday 2010). Indeed, BMI was neither associated with diabetes in women nor in men in our study. We examined whether this could be due to the high prevalence of poorly controlled diabetes in this population. The interpretation of Hb1Ac in areas with high prevalence of haemoglobinopathies is difficult (Smaldone 2008). Therefore, we used FPG as the indicator of glycaemic control in our population. In fact, the reduced mean BMI and mean body cell mass in patients with FPG ≥7 mm argues for diabetic muscle atrophy as an underlying mechanism (Park et al. 2006). After exclusion of poorly controlled diabetes cases, the odds ratios for diabetes increased slightly, but were still not significant. Thus, a poor glycaemic control is unlikely to explain the different associations observed for BMI and central obesity measures in our study population. As for the strongest anthropometric measure in our study, WHR, data are even scarcer: A Guinean survey (Balde et al. 2007) reports an odds ratio of 1.63 (95% CI: 1.01–2.63) for prevalent type 2 diabetes whereas this figure was 3.96 (1.76–8.92) in a cross-sectional Ethiopian study (Giday 2010). Interpretation is further complicated by high levels of ethnic diversity possibly resulting in different body compositions.
To the best of our knowledge, this is the first study comparing the discriminative power of different obesity measures for diabetes in SSA. In the existing studies on cardiovascular diseases from SSA, central obesity measures were better predictors of disease risk than BMI. One cross-sectional study in Ethiopia investigated the association between measures of adiposity and cardiovascular disease risk (Wai et al. 2011): WC in women and WTHR in men were the measures most strongly associated. Overall, comparisons of ROC curves identified that WC was the best predictor of cardiovascular disease risk among this population. In several other non-Caucasian indigenous populations, such as Taiwanese, South Indians, Australian Aboriginal people and Torres Strait Islanders, WHR has been consistently observed to be a better predictor of diabetes risk than general obesity measured by BMI (Wang et al. 2007; Kaur et al. 2008; Cheng et al. 2010).
Even though data from African Americans are not directly comparable, they also point to a particularly important role of central (rather than general) obesity as a risk factor for diabetes among blacks. In the Insulin Resistance Atherosclerosis Study, measures of central and general adiposity similarly predicted type 2 diabetes, except in African Americans, in whom central obesity measures were more predictive (Mackay et al. 2009). In the Atherosclerosis Risk in Communities Study, including both Caucasians and African Americans, WC was a better predictor for incident diabetes than BMI in the latter (Stevens et al. 2001).
Many studies throughout the world have suggested the use of ethnic-specific cut-off points for assessing diabetes risk (Lear et al. 2010; Qiao & Nyamdorj 2010b). As for the population-specific anthropometric cut-off points, data from Asian populations have revealed that lower values of general and central obesity measures might be meaningful for the identification of individuals at risk of diabetes (Misra et al. 2005). However, there is insufficient evidence for specific cut-offs for the identification of type 2 diabetes in SSA populations. Only one cross-sectional study reported on WC cut-offs in SSA and recommended 71.5 and 81.5 cm for women and 75.6 and 80.5 cm for men of Nigerian and Cameroon origin, respectively, for the identification of hypertension (Okosun et al. 2000). In our study, the optimal cut-off for WC was higher in women (91.7 cm) than in men (83.4 cm). West African men have generally lower WC values than Western populations (Okosun et al. 2000). WHR was the best obesity measure to identify diabetes cases at the recommended cut-off with high sensitivity and specificity in our study population.
In our study, WHR and BMI were weakly correlated, but WHR was the strongest predictor of diabetes, independent of BMI. Clearly, this indicates that body fat distribution is more important in diabetes development than generalised obesity. Visceral fat is highly metabolically active, and its accumulation, reflected by larger WHR, causes an increased delivery of free fatty acids to the liver, resulting in insulin resistance, hyperinsulinaemia and hypertriglyceridaemia (Despres et al. 2008) and hence increases greatly the risk of diabetes. We found a protective association of larger HC on diabetes among men. Our findings support other studies that have reported an inverse association between HC and diabetes in South Africans (Motala et al. 2008), Asian-Indian men (Asghar et al. 2011) and four non-Caucasian ethnic groups (Snijder et al. 2004). Several mechanisms could explain this inverse association including increased muscle mass or femoral fat mass accumulation reflected by large hips. The subcutaneous femoral fat tissue has been suggested to be protective, because of high lipoprotein lipase activity and due to the lower lipolysis rate in this region compared to the visceral region (Arner 1995). These fat depots are therefore more likely to protect liver and muscle by taking up and storing increased concentrations of free fatty acids (Snijder et al. 2003).
In contrast to studies in other ethnic populations (Vazquez et al. 2007), we found no association between BMI and diabetes. Presumably, general body fat mass may play a substantially different role among populations with African ancestry compared to Caucasians. This is supported by the observation that associations between the fat mass and obesity-related (FTO) gene, a major obesity risk locus, and diabetes are weaker or even contrary in populations with African ancestry compared to European populations (Hennig et al. 2009; Adeyemo et al. 2010; Bressler et al. 2010).
A strength of our study is that standardised screening for diabetes ensured that no undiagnosed individuals remained in the analyses. Only 3.7% of cases were on insulin monotherapy; the majority was on medical record. This argues for our definition of type 2 diabetes. All anthropometric measurements were carried out by trained staff, which reduces the impact of measurement error in our study. One limitation is the case–control design that impedes a straightforward interpretation of the temporal relation between obesity measures and diabetes. Prospective studies would be needed to provide stronger evidence for the association between obesity measures, fat distribution and diabetes risk among people from SSA. At the same time, not all potential confounders may have been detected and therefore accounted for. Also, we cannot rule out that the hospital- and community-based selection of controls has led to a comparison population supposedly heterogeneous and not fully representative of the source population from which the cases were drawn from. While this could have masked an association between BMI and diabetes, it seems implausible that it might also explain the higher risk observed with higher WHR in our study. Thus, selection bias seems an unlikely explanation for the stronger effect of WHR compared to BMI observed in our study.
In conclusion, our study suggests that WHR is one of the obesity measures most strongly linked with diabetes in this SSA population. It remains to be examined whether preventive strategies against type 2 diabetes should take into account WHR in addition to the conventional measure of BMI. In this region, women and men may equally benefit from public health efforts on the prevention of and the reduction in central obesity. Further investigations are needed to evaluate the rationale of country- or region-specific cut-off points for anthropometric indices to identify individuals with diabetes in SSA.
We thank all participants at KATH and acknowledge the study team of the Kumasi Diabetes and Hypertension Study for on-site recruitment, data and sample collection as well as laboratory analysis. This study was supported by Charité Universitätsmedizin Berlin (grant 89539150).