Poor pregnancy outcomes in women with type 2 diabetes

To identify factors associated with poor pregnancy outcomes in type 2 diabetes, and to predict poor pregnancy outcomes.

This project was completed by Laura Markendale, Senior Information Analyst, in the UEC Data Team Team, as part of the Data Science MRes at the University of Leeds

Increasing numbers of pregnancies in the UK are affected by type 2 diabetes, which is linked to rising obesity levels. Type 2 diabetes has recently overtaken type 1 diabetes to become the dominant form of pre-existing diabetes in pregnancy. Diabetes is known to increase the risk of adverse pregnancy outcomes such as congenital anomalies and stillbirth. The study aimed to identify factors associated with poor pregnancy outcomes in type 2 diabetes, and to predict poor pregnancy outcomes, using data from the National Pregnancy in Diabetes (NPID) audit .

Features identified as being important for predicting these outcomes included measures of glycaemic control in early pregnancy for predicting congenital anomalies, and in late pregnancy for predicting perinatal loss, along with diabetes duration, BMI, and deprivation. Deprivation was associated with perinatal loss, with over 50% of pregnancies with this outcome being to women living in the most deprived areas. Models generally performed poorly at predicting outcomes, with the models that were best at identifying the poor outcomes predicting around 45% of the good pregnancy outcomes incorrectly.

Conclusions

The best performing models all used measures of glycaemic control as a predictor, highlighting the importance of glycaemic control in achieving a good pregnancy outcome. Undersampling of the majority class resulted in the best performing models, however none of the models performed well at predicting both good and poor outcomes.