Effects of neonicotinoid pesticide exposure in the first trimester on gestational diabetes mellitus based on interpretable machine learning

Abstract

Background

Gestational diabetes mellitus (GDM) is one of the most common pregnancy complications and seriously threatens the health of mothers and offspring. Neonicotinoids (NEOs) is a new class of pesticide and widely used worldwide. Prenatal NEOs exposure had negative effects on fetal growth, but the potential effect of NEOs exposure on pregnancy complications remain unclear.

Objectives

To examine the individual and jointed effects of serum neonicotinoids (NEOs) pesticide exposure on gestational diabetes mellitus (GDM), and explore the application of NEOs exposure levels as predictor of GDM.

Methods

We conducted a prospective cohort study based on Guangxi Zhuang Birth Cohort, China. A total of 1450 mather-infant pairs were included from 2015 to 2019. Ten NEOs were measured by UPLC-MS. Maternal serum samples were collected during gestational age 0–12 weeks. Individual and jointed effects of NEOs on GDM were assessed through binomial regressions, Bayesian Kernel Machine Regression and quantile g-computation. Prediction of GDM using XGboost machine learning and SHapley Additive exPlanations (SHAP).

Results

A total of 122 (8.4%) mothers were diagnosed with GDM. In the individual exposure models, sulfoxaflor and thiamethoxam exposure in the first trimester significantly increased the risk of GDM (OR = 1.48, 95%CI: 1.21, 1.82; OR = 1.42, 95%CI: 1.14, 1.78). Moreover, GDM risk increased significantly with NEOs mixture concentration was above 75th percentile, compared with the 50th percentile. Sulfoxaflor and thiamethoxam as the main positive contributing factors in NEOs mixture to increase the GDM with a weight of 29.3% and 27.6%, respectively. Furthermore, sulfoxaflor and thiamethoxam were the most important contributing factors for predicting GDM after combining traditional risk factors in machine learning model, with predicted contribution values of 0.79 and 0.46, respectively.

Conclusion

Our findings suggested that elevated maternal serum sulfoxaflor, thiamethoxam and NEOs mixture were positively associated with GDM, and sulfoxaflor, thiamethoxam were the important contributing factors for predicting GDM.