A CLINICAL MODEL FOR THE PREDICTION OF DIET CONTROLLED GDM

Objectives Universal screening for Gestational diabetes mellitus(GDM) identifies women with varying degrees of glucose intolerance, many of which will achieve glycemic control by lifestyle and nutritional modification alone. The objective of the study was to develop a prediction model based on biochemical, clinical and sonographic parameters to accurately identify these “lower-risk” women that could then be directed to a less resource intensive care pathway. Methods This was a retrospective cohort study. Relevant clinical, laboratory and sonographic data was extracted from the medical records of women diagnosed with GDM between 24-32 weeks’ gestation at two academic hospitals in Ontario. The following clinical risk factors were considered as candidate predictors for the multivariable predictive model: maternal age, GCT value, GTT fasting plasma glucose value, family history of diabetes, history of GDM, fetal abdominal circumference (AC), gestational age and body mass index at GDM diagnosis visit. Discrimination of the model based on the number of clinical risk factors in combination with fetal AC percentile and alone was assessed using the c-statistic, which is an equivalent concept to the area under the receiver operating characteristic curve (AUC ROC). Results A total of 961 women with GDM met inclusion criteria. 601 (62.5%) women did not require pharmacological management of GDM and were designated as “low-risk”. On univariable and multivariable analysis, fetal AC was not associated with GDM treatment and removed from the final model. The final predictive model had high sensitivity [0.90 (0.87-0.92)] but low specificity [0.29 (0.23-0.34)] and hence a high false positive rate (71%). These results did not change when stratified by method of GDM diagnosis. Conclusions While correctly identifying 90% of women with GDM who will require only dietary and lifestyle modification, this prediction model falsely labeled 71% of women requiring medical therapy as being “low-risk”. A model with increased specificity is needed before it can be safely applied to clinical practice.