Causal insights into gestational diabetes mellitus

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Computational Biology

Background:

• Gestational diabetes mellitus (GDM) is the most common metabolic disorder during pregnancy.
• GDM increases risks for adverse maternal, neonatal, and long-term metabolic health outcomes.
• Targeted interventions are needed to mitigate GDM-associated complications.

Purpose of the Study:

• To identify potential causal relationships within clinical data of GDM patients.
• To support the development of more targeted and effective GDM interventions.
• To enhance clinical decision-making through quantifiable insights.

Main Methods:

• Analysis of a curated GDM patient dataset (2014-2016, 2020).
• Construction of a knowledge graph integrating clinical expertise, literature, and GPT-4 insights.
• Application of 20 structure learning algorithms to infer Causal Bayesian Networks (CBNs).
• Utilized model-averaging for a consensus-based causal structure.

Main Results:

• An integrative model provided stable representations and quantifiable insights for clinical decision-making.
• Clinicians reported increased confidence in personalized, evidence-based GDM care strategies.
• Key findings include the impact of birth weight on NICU admissions and dietary intervention on maternal glucose regulation.
• Sensitivity analysis identified birth weight, gestational age, and mode of delivery as major determinants of outcomes.
• Non-modifiable factors like multiple pregnancies and prior GDM contributed to risk stratification.

Conclusions:

• Structure learning techniques applied to observational GDM data identified clinically relevant relationships.
• The study's insights can generate hypotheses for refining intervention strategies.
• The findings aim to improve patient outcomes in GDM care through data-driven approaches.