DemoGraphDB- A Novel Knowledge Graph Approach to Predicting Inpatient Mortality
Open Access DepositedThis praxis explores the integration of Social Determinants of Health (SDOH) with healthcare data using Knowledge Graph (KG) embeddings to predict inpatient mortality at the time of admission. The research is moti vated by the significant impact of SDOH on health outcomes and the poten tial for KGs to provide a structured representation of complex relationships between entities. The methodology involves creating a novel framework, DemoGraphDB, which combines healthcare data from the 2021 New York State Inpatient Database with SDOH data from the American Census Survey geospatially. The SDOH data is structured into a KG and transformed into embeddings that captures all information present for a specific FIPS code using the StarE graph neural network algorithm. This embedding is then appended to eachpatient’s FIPS code andusedasinputintoatransformermodel. The full framework improves mortality prediction accuracy by 86 basis points compared to a baseline logistic regression model. The research demonstrates the feasibility of using KGs to integrate di verse datasets and enhance predictive power in healthcare applications. The study also highlights the potential of using KGs to inject domain knowledge, improve data integration, and enable causal reasoning in ma chine learning models. The findings suggest that incorporating SDOH into predictive models leads to increased prediction accuracy that may enable proactive healthcare interventions, ultimately benefiting patient outcomes and reducing healthcare costs.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.