Improving Risk Assessment of Miscarriage during Pregnancy with Knowledge Graph Embeddings

Abstract

AbstractMiscarriages are the most common type of pregnancy loss, mostly occurring in the first 12 weeks of pregnancy due to known factors of different natures. Pregnancy risk assessment aims to quantify evidence in order to reduce such maternal morbidities during pregnancy, and personalized decision support systems are the cornerstone of high-quality, patient-centered care in order to improve diagnosis, treatment selection, and risk assessment. However, the increasing number of patient-level observations and data sparsity requires more effective forms of representing clinical knowledge in order to encode known information that enables performing inference and reasoning. Whereas knowledge embedding representation has been widely explored in the open domain data, there are few efforts for its application in the clinical domain. In this study, we discuss differences among multiple embedding strategies, and we demonstrate how these methods can assist on clinical risk assessment of miscarriage both before and specially in the earlier pregnancy stages. Our experiments show that simple knowledge embedding approaches that utilize domain-specific metadata perform better than complex embedding strategies, although both are able to improve results comparatively to a population probabilistic baseline in both AUPRC, F1-score, a proposed normalized version of these evaluation metrics that better reflects accuracy for unbalanced datasets.