Optimal cut-offs of four anthropometric measures and their predictive ability of diabetes in a nationally representative Kenyan study


 Background Type 2 diabetes (T2D) is one of the top non-communicable diseases in Kenya and prevention strategies are needed. Intervening to reduce obesity is the most common prevention strategy. However, black ethnicities develop T2D at lower obesity levels and it is unclear what anthropometric cut-offs could provide the best predictive ability for T2D risk. This study, therefore, aimed to determine the optimal anthropometric cut-offs and their predictive ability of T2D. Methods The study included 2159 participants (59% women) aged 35-69 years from the Kenya STEPwise survey conducted in 2014. Four anthropometric measures – body mass index (BMI), waist circumference (WC), waist to hip ratio (WHR), and waist to height ratio (WHtR) – were used. Diabetes was defined as a fasting blood glucose of ≥7.0 mmol/l or a previous diagnosis for diabetes. Optimal anthropometric cut-offs and their receiver operating characteristics, such as the area under the curve (AUC), were computed. Results Overall, the optimal cut-off for BMI, WC, WHR and WHtR were 24. 8kg.m-2, 90.0cm, 0.88 and 0.54. On disaggregation by sex, the optimal cut-off for BMI, WC, WHR and WHtR was 27.1 kg.m-2, 87cm, 0.85 and 0.55 in women and 24.8 kg.m-2, 91cm, 0.88, and 0.54 in men. Overall, WC (AUC 0.71 (95% confidence interval 0.65, 0.76)) had a better predictive ability for T2D than BMI (AUC 0.68 (0.62, 0.73), p=0.0314)). Also, WHtR (AUC 0.71 (0.66, 0.76)) had better a predictive ability for T2D than BMI (AUC 0.68 (0.62, 0.73), p=0.0182)). Conclusions WC and WHtR were better predictors of T2D than BMI and should be used for risk stratification in Kenya. A WC cut-off of 87cm in women and 91cm in men, or a WHtR cut-off of 0.55 in both men and women should be used to identify individuals at high risk of T2D.


Introduction
The prevalence of diabetes is on the rise in sub-Saharan Africa (SSA) exerting a disease and economic burden (1). Type 2 diabetes (T2D), the main type of diabetes, is preventable through lifestyle modification such as weight loss (2). Excess body fat is an important risk factor for T2D and anthropometric measures that estimate body fat such as body mass index (BMI), waist circumference (WC), waist to hip ratio (WHR), and waist to height ratio (WHtR) have been used to predict risk (3).
However, debate remains on which is the best measure to characterise risk. Despite BMI being the most used measure it has been challenged on its inability to: 1) differentiate between body lean mass and fat mass; and 2) detect the location of body fat (4). Central obesity measures (WC, WHR, WHtR) that are proxies of visceral fat may be better measures of risk as visceral fat is more important for T2D development than peripheral fat (5). Consequently, a meta-analysis of 20 studies reported that WHtR and WC provide a more robust measure for detecting T2D than BMI (6). In SSA, several studies have reported central obesity measures as better predictors of T2D than BMI (7)(8)(9)(10)(11)(12)(13). However, it remains unclear which measure has a better predictive ability. For instance in Ghana and Ethiopia, WHR was a better predictor of diabetes in three studies (8,11,12) while WC was a better predictor of diabetes in five studies from Ethiopia, Cameroon, Kenya, and Guinea (7,(9)(10)(11)(12).
Further, there is increasing evidence that black populations develop T2D at lower obesity levels compared to whites, and therefore current anthropometric cut-offs may be less appropriate for them.
For example, the age-adjusted prevalence of T2D in whites at 30 kg.m -2 , WC of 102cm in men, and 88cm in women equated to a BMI of 26 kg.m -2 , WC of 88cm and 79cm in men and women respectively in the black ethnicity (14). There is emerging but limited evidence of the optimal cut-offs to predict T2D in SSA. In Ghana, the optimal cut-offs for BMI, WC and WHR were 26.2 kg.m -2 , 91.7cm and 0.88 for women, and 26.7 kg.m -2 , 83.4cm and 0.90 for men (8). Recently, an Ethiopian study reported that the optimal cut-offs for BMI, WC, WHR and WHtR were 20.5 kg.m -2 , 82.9cm, 0.86 and 0.51 in women, and 23.0 kg.m -2 , 88.6cm, 0.97 and 0.52 in men (12). This study, therefore, aimed to determine the optimal anthropometric cut-offs of four measures (BMI, WC, WHR, WHtR) for predicting T2D in Kenya and compare their predictive ability.

Methodology
The study used data from the Kenya STEPwise survey 2014 which was a nationally representative study that investigated non-communicable diseases risk factors. A multi-stage cluster sampling method was used to recruit 4479 participants aged 18-69 years. Data were collected by trained personnel and involved the administration of a questionnaire, physical and biological measurements (15). More information on the survey can be found elsewhere (15).

Inclusion and exclusion criteria
To limit analyses to type 2 diabetes cases we included participants between 35 and 70 years of age as type 2 diabetes prevalence increases with age (16) and type 1 diabetes is relatively low from ≥35 years of age (17). A total of 2159 participants with complete data on all anthropometric measures and diabetes diagnosis were included in the analysis.

Data collection procedures and definitions
Height was measured in centimetres using a portable stadiometer and weight was measured in kilogram using a portable electronic weighing scale. Waist circumference was measured at the umbilicus level using a constant tension tape. Blood pressure was measured three times using an automated blood pressure measuring instrument (OMRON®). A minimally invasive prick was used to draw a blood sample and CardiocheckPA analyser® was used to measure blood glucose levels (15).

Analysis
All data analysis was done using R statistical package version 3.5.2 (20). Means and standard deviations were used to summarise continuous variables, and frequencies and percentages used to summarise categorical variables. The 'OptimalCutpoints' package was used to compute the optimal anthropometric cut-offs and their receiver operating characteristics (ROC) using the Youden Index.
The characteristics computed were: the Area Under the Curve (AUC), sensitivity, specificity, positive predictive value, and negative predictive value associated with each cut-off (21). The 'pROC' package was used to compare the AUC of anthropometric measures using BMI as the reference (22). The ROC curves of all four anthropometric measures were plotted to show sensitivity (y-axis) and specificity (xaxis). Data are presented for the overall sample and by sex disaggregation. A statistical significance level of p<0.05 was used. Table 1 presents the characteristics of the 2159 study participants included in the analyses. The average age was 48 years, and the diabetes prevalence was 6.6% in women and 4.3% in men. The majority of both women and men were of normal weight status. WHtR-mean ± SD 0.51 ± 0.09 0.52 ± 0.10 0.48 ± 0.08 BMI -body mass index; WHR -waist to hip ratio; WHtR -waist to height ratio; SD -standard deviation Figure 1 and Figure 2 present the ROC analyses for diabetes by anthropometric measures while

Discussion
This study aimed to determine optimal anthropometric cut-offs to predict T2D and compare their predictive ability. The optimal BMI cut-off was 27.1 kg.m -2 and 24.8 kg.m -2 in women and men, respectively. These findings were consistent with a Ghanaian study which found an overweight BMI of 26.2 kg.m -2 as the optimal cut-off in women (8) and an Ethiopian study that reported a normal weight BMI of 23.0 kg.m -2 as the optimal cut-off in men (12). Also, this is in agreement with increasing evidence that Africans develop T2D at a lower BMI (23) than the current obesity cut-off of 30kg.m -2 (24). Further, WC and WHtR had a better discriminatory ability for T2D than BMI. This agrees with other findings from Guinea, Kenya, Cameroon, and Ethiopia which have shown that WC is a better predictor of T2D (7,(9)(10)(11)(12). Globally, these findings are consistent with a meta-analysis that found that WC and WHtR had better discrimination for T2D than BMI (6). WC and WHtR are proxies of visceral fat which confers a higher risk of metabolic complications such as T2D compared to peripheral fat (25).
This implies that these measures, rather than BMI, should be used for risk stratification in Kenya.
We found the optimal WC cut-off as 87cm for women and 91cm for men. These cut-offs differ from those from other SSA settings: in Ghana, the optimal cut-off was 91.7cm for women and 83.4cm for men (8) while in Ethiopia the optimal cut-off was 82.9cm for women and 88.6cm for men (12). Despite these contrasts, our findings show a similar trend to the other two studies, that is, the cut-offs for women are above the WHO cut-off for metabolic complications (≥80cm) while the cut-offs in men are below the WHO thresholds (≥94cm) (24). A similar trend has been reported by a large analysis of data (n=24181) from eight SSA countries which showed that men were at increased cardiometabolic risk at lower WC (≥81.2cm) than current guidelines (≥94cm) (26). This implies that a WC cut-off of ≥91cm should be used for predicting T2D risk in men instead of the recommended ≥94cm. In women, the WHO recommended cut-off of ≥88cm which corresponds to "substantially increased risk" of metabolic complications (24) should be used to predict T2D risk in Kenya.
The optimal WHtR cut-off was relatively similar for women and men -0.55 and 0.54, respectively. This is slightly higher than what was reported in Ethiopia -0.51 in men and 0.52 in women (12). The derived WHtR cut-offs are relatively similar to the recommended and widely used cut-off of ≥0.5 in women and men which translates to a simple screening message of "keep your waist to less than half your height" (27)(28)(29). Apart from a simple public health message, the same WHtR cut-off is used for both men and women, and adults and children (29). Further, WHtR is easier to measure and calculate than BMI and a string can be used for measurement (a string measuring half someone's height should fit around their waist) making it a very cheap screening method (27,29). This implies there is need to consider the use of WHtR in risk stratification and associated simple public health messaging to intervene on central obesity and prediction of T2D risk.
This study adds to the limited evidence on optimal anthropometric cut-offs and their predictive ability of T2D in Kenya and SSA. Its strengths are the use of a large nationally representative sample, and identification of optimal cut-offs, and predictive ability of all four commonly used anthropometric measures. However, just like the recommended cut-offs, the study uses cross-sectional evidence which limits causal inference of derived cut-offs on T2D. More evidence, particularly from prospective studies, is needed to confirm if the derived cut-offs are the most appropriate for T2D prediction.
In conclusion, WC and WHtR were better predictors of T2D than BMI highlighting their utility in risk stratification. The optimal cut-offs of WC (87cm for women and 91cm for men) and WHtR (0.54 for women and men) should be used for risk stratification in Kenya.

Availability of data and materials
The Kenya STEPwise survey 2014 data can be assessed from the Kenya National Bureau of Statistics using this link: http://statistics.knbs.or.ke/nada/index.php/catalog/90 Ethics approval and consent to participate.

Consent for publication
Not applicable