Application of Bayesian Latent Variable Model for Early Detection of Gestational Diabetes Mellitus Without A Perfect Reference Standard Test by β‐human Chorionic Gonadotropin

Abstract

Background and Objectives: Gestational diabetes mellitus (GDM) is a medical problem in pregnancy, and its late diagnosis can cause adverse effects in the mother and fetus. The purpose of this research was to estimate the accuracy parameters of a biomarker for early prediction of gestational diabetes in the absence of a perfect reference standard test. Methods: This study was conducted in 523 pregnant women who presented to Mahdieh Hospital and Taleghani Hospital affiliated with Shahid Beheshti University of Medical Sciences, Tehran, Iran 2017-2018. As a predictor for detecting GDM, beta- human chorionic gonadotropin (β-hCG) measurements were recorded during 14-17th weeks’ gestation in a checklist. The Bayesian latent variable model was used to estimate the sensitivity, specificity, and area under receiver operating characteristic curve (AUC). Bayesian parameter estimation was calculated using the R2OpenBUGS package in R version 3.5.3. Results: The median gestational age was 33 years. In the absence of a perfect reference test, the applied model had a sensitivity, specificity, and AUC of 78% (95% credible interval (CrI): 0.66-0.83), 83% (95% CrI: 0.74-0.89), and 0.72 (95% CrI: 0.64-0.88) for β-hCG, respectively. Conclusion: According to the results of this study, β-hCG may be an acceptable biomarker for early diagnosis of diabetes in pregnant women in the absence of a perfect reference test.

Keywords

• latent variable modeling
• roc curve
• bayesian approach
• gestational diabetes
• human chorionic gonadotropin