Revised gestational diabetes prediction instrument reveals improved efficiency

The study dealt with in this summary was published as a preprint on medRxiv.org and has not yet been reviewed.

The central theses

• An external validation study provided evidence of the need for revisions to an earlier model used to predict a pregnant woman’s risk of developing gestational diabetes (GDM) in many situations and revised ethnic categories early in pregnancy.

• The revision applied best practices using large validation and update datasets obtained from an ethnically diverse population diagnosed with GDM based on current criteria and a universal screening strategy with a GDM prevalence of 18.0%.

• The report demonstrates the potential value of working with an existing validated model and updating it to maintain predictive performance over time rather than starting from scratch.

Why is that important?

• The increasing evidence supports the integration of the revised GDM prediction model into routine practice in order to accelerate and improve the strategic risk management of women at risk of GDM.

• Both the original and newly revised models use variables that are routinely collected and recorded in clinical practice, avoiding the barriers and costs of collecting additional information.

Study design

• Researchers used routinely collected health data from 26,474 singleton pregnancies that resulted in childbirth between January 2016 and December 2018 at Monash Health, Australia’s largest health service, which includes three maternity hospitals and serves an ethnically diverse population.

• They defined the diagnosis of GDM based on the criteria of the International Association of Diabetes and Pregnancy Study Groups.

• Investigators updated the ethnic classification system to reflect international ethnic categories and self-reported ethnic names. In contrast, the original model was based on extrapolation of ethnicity from the country of birth.

• They selected the best model using predictive performance measures and a closed-loop testing process. C-statistics (the area under the receiver’s operating curve) were used to evaluate and compare the models developed.

Main results

• The original model produced a C statistic of 0.698, indicating “fair” discrimination.

• Model C2 turned out to be the preferred model because of the comparable calibration plot in the high prevalence region, a superior C statistic of 0.732, the use of generalizable ethnicity categories, and because it showed significantly better fit in closed tests.

• The new model uses the variables of age, body mass index, family history of diabetes, history of GDM, history of poor birth outcome, and ethnicity.

limitations

• The revised model treats the continuous variables body mass index and age as categorical variables, an approach that can reduce predictive power and can be replaced by electronic risk calculators. The reassessment of the relationship between body mass index and age as continuous variables and the diagnosis of GDM would produce a completely new predictive model that goes beyond the scope of validation and updating.

Disclosure

• The preprint currently does not contain any information on funding or author information.

• A previously published description of the study design indicated that it did not receive commercial funding and none of the authors had commercial information.

This is a summary of a preprint research study, External Validation and Update of a Predictive Model for Diagnosing Gestational Diabetes Mellitus, written by researchers primarily based at Monash University, Clayton, Australia, via medRxiv, made available to you by Medscape. This study has not yet been reviewed. The full text of the study can be found on medRxiv.org.

Mitchel L. Zoler is a reporter for Medscape and MDedge based in Philadelphia. @mitchelzoller

For more news on diabetes and endocrinology, follow us on Twitter and Facebook.

Follow Medscape on Facebook, Twitter, Instagram, YouTube and LinkedIn.