DEVELOPING A MULTIVARIABLE PROGNOSTIC PREDICTION MODEL FOR THE RISK OF DELIVERING LARGE-FOR-GESTATIONAL-AGE INFANTS AMONG WOMEN WITH TYPE 2 DIABETES MELLITUS

Abstract

Background: The incidence of large-for-gestational-age (LGA) infants, associated with adverse pregnancy outcomes, is notably higher among women with diabetes mellitus compared to healthy pregnant women. Traditional diagnostic methods for LGA have limitations, necessitating the development of effective predictive models [1-3]. Aim: This study aims to develop and internally validate prediction model and nomogram to identify pregnant women with T2DM at risk of delivering LGA infants. Methods: This retrospective cohort study analyzed data from 1,224 women with type 2 diabetes mellitus (T2DM) who delivered at a major hospital in Qatar between 2016 and 2022. The primary outcome was large-for-gestational-age (LGA), defined as birth weight above the 90th percentile. A Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression approach was then applied to select the most relevant predictors and construct the multivariable logistic regression model. Missing data was minimal (<5%). The model was developed following TRIPOD guidelines, using STATA 18.5 for analysis, and a nomogram was created to support clinical application. Results: Among 1,224 pregnant women with T2DM, 18.2% delivered LGA infants. Using LASSO-based selection, a multivariable logistic regression model was developed incorporating eight predictors: diabetes duration, use of metformin, use of bolus insulin, glycemic control in the first and last trimester, pre-pregnancy BMI, gestational weight gain ,and ethnicity. In the adjusted analysis, poor glycemic control in both the first and last trimesters, obesity (BMI ≥30 kg/m²), and excessive GWG (≥0.26 kg/week) were significantly associated with increased LGA risk. Other predictors, including ethnicity, diabetes duration, and use of metformin or insulin, were not statistically significant. The model demonstrated good performance, with 82% correct classification, an AUC of 0.68 (95% CI: 0.64–0.82), and good calibration (Hosmer–Lemeshow P = 0.85). A nomogram was constructed based on the final model to support individualized LGA risk estimation in clinical practice. Conclusion: The developed nomogram, incorporating easily accessible risk factors, provides an individualized prediction tool for identifying pregnant women withT2DM at risk of delivering LGA infants, supporting early prevention and intervention strategies. The use of TRIPOD guidelines ensures the model's transparency and reliability, while the nomogram enhances its clinical applicability.