Abstract:
AIM. We aimed to develop a prediction model for the diagnosis of gestational diabetes and to evaluate the performance of published
prediction tools on our population. METHODS. We conducted a cohort study on nondiabetic women < 26 weeks gestation at a level 1
clinic in Johannesburg, South Africa. At recruitment, participants completed a questionnaire and random basal glucose and HbA1c
were evaluated. A 75 g 2-hour OGTT was scheduled between 24–28 weeks gestation, as per FIGO guidelines. A score was derived
using multivariate logistic regression. Published scoring systems were tested by deriving ROC curves. RESULTS. In 554 women, RBG,
BMI, and previous baby ≥ 4000 g were significant risk factors included for GDM, which were used to derive a nomogram-based
score. The logistic regression model for prediction of GDM had R2 0.143, Somer’s Dxy rank correlation 0.407, and Harrell’s
c-score 0.703. HbA1c did not improve predictive value of the nomogram at any threshold (e.g,. at probability > 10%, 25.6% of
cases were detected without the HbA1c, and 25.8% of cases would have been detected with the HbA1c). The 9 published scoring
systems performed poorly. CONCLUSION. We propose a nomogram-based score that can be used at first antenatal visit to identify
women at high risk of GDM.
