Integration and Harmonization of Multi-Source Obstetric Data Using Rule-Based NLP for Fetomaternal Risk Modelling

Area of Science:

• Medical Informatics
• Obstetrics and Gynecology
• Data Science

Background:

• Over 80% of German pregnancies are high-risk, causing inefficient resource allocation.
• Limited data standardization across clinical databases hinders accurate fetomaternal risk prediction models.

Purpose of the Study:

• To develop a Natural Language Processing (NLP) pipeline for integrating and standardizing obstetric data.
• To create a harmonized real-world fetomaternal dataset (FEMAR) for improved risk prediction.

Main Methods:

• A rule-based NLP pipeline was developed to process structured and unstructured obstetric data.
• Data was integrated from multiple hospital IT systems, standardizing 449 fetomaternal factors.
• The FEMAR dataset includes 123,183 unique birth deliveries.

Main Results:

• Successfully integrated and standardized diverse obstetric data from multiple sources.
• Created the comprehensive FEMAR dataset, a foundation for advanced predictive modeling.
• Established a standardized approach to fetomaternal data for future research.

Conclusions:

• The developed NLP pipeline effectively harmonizes obstetric data, overcoming standardization barriers.
• The FEMAR dataset provides a robust foundation for developing accurate risk prediction models for pregnancy complications.
• This work facilitates more efficient resource allocation in high-risk pregnancies.