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

More than 80% of pregnancies in Germany are classified as high risk, leading to inefficient resource allocation. Although data-driven approaches could improve fetomaternal risk prediction, limited standardization across clinical databases remains a major barrier. We developed a rule-based Natural Language Processing (NLP) pipeline to integrate and standardize structured and unstructured obstetric data from multiple hospital IT systems. The resulting harmonized real-world fetomaternal dataset (FEMAR) comprises 123,183 unique birth deliveries and 449 fetomaternal factors, providing a foundation for developing more accurate risk prediction models for pregnancy complications.