Maternal BMI and diabetes in pregnancy: Investigating variations between ethnic groups using routine maternity data from London, UK

Objective To investigate the ethnicity-specific association between body mass index (BMI) and diabetes in pregnancy, with a focus on the appropriateness of using BMI cut-offs to identify pregnant women at risk of diabetes. Study design Analysis of routinely-collected data from a maternity unit in London, UK. Data were available on 53 264 women delivering between 2004 and 2012. Logistic regression was used to explore the association between diabetes in pregnancy and BMI among women of different ethnicities, and adjusted probability estimates were used to derive risk equivalent cut-offs. ROC curve analysis was used to assess the performance of BMI as a predictor of diabetes in pregnancy. Results The prevalence of diabetes in pregnancy was 2.3% overall; highest in South and East Asian women (4.6% and 3.7%). In adjusted analysis, BMI category was strongly associated with diabetes in all ethnic groups. Modelled as a continuous variable with a quadratic term, BMI was an acceptable predictor of diabetes according to ROC curve analysis. Applying a BMI cut-off of 30 kg/m(2) would identify just over half of Black women with diabetes in pregnancy, a third of White (32%) and South Asian (35%) women, but only 13% of East Asian women. The 'risk equivalent' (comparable to 30 kg/m(2) in White women) threshold for South Asian and East Asian women was approximately 21 kg/m(2), and 27.5 kg/m(2) for Black women. Conclusions This study suggests that current BMI thresholds are likely to be ineffective for diabetes screening in South and East Asian women, as many cases of diabetes will occur at low BMI levels. Our results suggest that East Asian women appear to face a similarly high risk of diabetes to South Asian women. Current UK guidelines recommend diabetes screening should be offered to all pregnant South Asian women; extending this recommendation to include women of East Asian ethnicity may be appropriate.