Oral Presentation Australasian Diabetes in Pregnancy Society Annual Scientific Meeting 2019

Novel Web-based Risk Calculators and Nomograms for Predicting Adverse Pregnancy Outcomes in GDM Women (#15)

Tang Wong 1 2 3 , Glynis Ross 1 2 , Robyn Barnes 1 4 , N Wah Cheung 2 5 , Jeff Flack 1 3 6
  1. Department of Diabetes and Endocrinology, Bankstown-Lidcombe Hospital, Sydney, NSW, Australia
  2. University of Sydney, Sydney, Australia
  3. University of NSW, Sydney, Australia
  4. University of Newcastle, Newcastle, NSW, Australia
  5. Department of Diabetes & Endocrinology, Westmead Hospital, Westmead, NSW, Australia
  6. Western Sydney Univerisity, Sydney, NSW, Australia

 

Background:
Risk prediction calculators, such as the Framingham fracture risk calculator (FRAX) are commonly used in medical practice but are not yet available for GDM management.

Aim:
1)To create, compare and validate predictive models of adverse pregnancy outcomes in GDM across ADIPS1998 and IADPSG criteria
2)To use these models to create nomograms and online risk calculators available for web access

Methods:
The training dataset for model creation included women diagnosed with GDM (ADIPS1998 criteria2) at Bankstown-Lidcombe hospital from 1992-2013. Datasets from Jan2014-Feb2016(also ADIPS1998 criteria) and March2016-March2019 (IADPSG criteria) were used for validation purposes. Antenatal variables were analysed for correlation and significance on univariate analyses and included in multivariate models(if p<0.05). The final models were fitted using logistic regression. End-points assessed included need for insulin therapy, large for gestational-age infant (LGA) and a composite neonatal outcome (≥1 of the following; needing insulin therapy, pre-term labour, caesarean section, LGA, neonatal hypoglycaemia/jaundice). Two models were created per outcome evaluated; one for continuous variables(model 1) and a second model for their categorical counterparts(model 2).

Results:
There were a total of 3095 singleton births to GDM women in the training dataset. There were 759 and 1280 women in the validation datasets according to ADIPS1998 and IADPSG criteria respectively. Risk calculators and nomograms were generated and are accessible on www.gdmriskcalculator.com. LGA and the composite outcome could only be modestly predicted by the models. However models for predicting Insulin therapy, in particular model 1, performed well; AUC-ROC 0.757(95% CI.737-0.776). Across all outcomes, model 1 performed better than model 2 in both training and validation datasets. Performance of each model in the training and validation datasets is displayed in the table below.

5cf203c429322-NEW+Comparison+validation.jpg

Conclusion:
The risk of adverse pregnancy outcomes in GDM, particularly insulin therapy can be reliably predicted using GDM risk calculators. Risk calculators incorporating continuous variables (model 1) performed better than those exclusively using categorical variables (model 2) and performed more consistently across ADIPS1998 and IADPSG criteria.

  1. Unnanuntana, A., et al., The assessment of fracture risk. The Journal of bone and joint surgery. American volume, 2010. 92(3): p. 743-753.
  2. Hoffman, L., et al., Gestational diabetes mellitus--management guidelines. The Australasian Diabetes in Pregnancy Society. Med J Aust, 1998. 169(2): p. 93-7.