Background: Due to the increased global burden of glucose intolerance and type 2 diabetes (T2D) among women who have had gestational diabetes mellitus (GDM), prediction models are prioritised for early risk stratification and timely intervention. This systematic review aimed to examine methodological characteristics, risk of bias and reporting quality of existing prognostic models predicting postpartum glucose intolerance and T2D following GDM.
Method: A systematic review was conducted searching seven databases (MEDLINE, Embase, Scopus, Web of science, CINHAL, Maternity & Infant Care Database (MIDIRS), and Global Health 1910 to 2022 Week 18) with no date restriction. The Checklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies (CHARMS) was applied. To assess the risk of bias and applicability, Prediction models Risk Of Bias Assessment Tool (PROBAST) was used. The protocol was registered at PROSPERO CRD42022327239.
Results: The systematic review retrieved 15 eligible publications. Models were developed in the USA (n=4), Europe n=6), Australia (n=2), Asia (n=1), Canada (n=1), and Ethiopia (n=1) between 1995 and 2022. Only two prognostic models were identified to have low overall risk of bias based on the PROBAST tool. Traditional statistical models were used most, with only few applying machine learning (13.3%). Common predictors included in the final models were body mass index (73%), fasting glucose concentration during pregnancy (53%), maternal age (40%), family history of T2D (33%), biochemical variables (lipid metabolites, triacylglycerols, cholesterol) (27%), oral glucose tolerance test result during pregnancy (27%), use of insulin during pregnancy (20%), postnatal fasting glucose level (20%), genetic risk factors (13%), Haemoglobin A1c (13%), and weight (13%). Of the 15 prognostic models, only 4 were internally validated and none externally validated. Model discrimination and calibration were reported in 13 and 4 studies, respectively. Only seven studies (47%) included model presentations, mostly using a risk score.
Conclusions: This systematic review identified that existing prognostic models for glucose intolerance following GDM were not externally validated and only a few were internally validated. In addition, there were high risk of bias, unreported model calibration, and low use of model presentation methods. Future research should focus on the development of robust, high-quality risk prediction models through incorporating easily accessible prognostic determinates to enhance the practical application and accuracy of risk prediction models for glucose intolerance and T2D following GDM. External validation and safety, clinical and cost effectiveness assessments are also required before implementing these prediction models in clinical practice.