| Autor: |
Lin, Zhiming, Hong, Linghong, Cai, Xiaohai, Chen, Siyao, Shao, Zhiyu, Huang, Yan, Yang, Chenhui, Chen, Longbiao |
| Zdroj: |
CCF Transactions on Pervasive Computing and Interaction; 20220101, Issue: Preprints p1-16, 16p |
| Abstrakt: |
There are many risks that may lead to serious consequences in the prescriptions. The traditional prescription risk detection methods rely on rational medication monitoring systems based on the database, which cannot cover the medication patterns, and often conflicts with clinical practice, resulting in false-reporting and under-reporting. To effectively detect the risks in prescriptions, we propose a prescription risk detection framework based on knowledge graph leveraging medical big data. Firstly, we construct a medical knowledge graph using the knowledge extracted from medical text data and historical prescriptions data. Then, based on the constructed knowledge graph, we detect the medication risk and complete the risk edges of the knowledge graph in three aspects: (1) we extract the medication usages that are consistent with both the description of the drug instructions and the historical prescriptions, and then label the risk degree in the graph. (2) we collect the medication patterns that did not conform to the instructions and extract the features leveraging knowledge graph to detect the risk of off-label drug use. (3) we represent the new drugs that lack risk information using graph model, then detect the risk of new drugs using knowledge graph completion task. Finally, we get a complete medical risk knowledge graph to detect the risk of clinical prescriptions. We use real-world prescriptions data from a tertiary hospital in Fujian Province for verification. The results show that our framework performs best among the baseline and can effectively detect the risks in the prescription. |
| Databáze: |
Supplemental Index |
| Externí odkaz: |
|
Full text from SpringerLink