Using Neural Machine Translation in Public Health Informatics to Propose Causal Sequence of Death

PROJECT TITLE :

Proposing Causal Sequence of Death by Neural Machine Translation in Public Health Informatics

ABSTRACT:

Over 2.7 million people pass away every year in the United States alone, contributing to the annual global death toll that approaches 57 million. Reporting deaths in a timely manner that are accurate and comprehensive is essential for public health, particularly during the COVID-19 pandemic, as institutions and government agencies rely on death reports to formulate responses to communicable diseases. Regrettably, even for seasoned medical professionals, it can be difficult to ascertain the reasons behind a person's passing. In light of the fact that medical professionals and specialists are still gathering information regarding COVID-related complications, the novel coronavirus and its variants might make the task even more difficult. An advanced Artificial Intelligence (AI) approach is presented with the goal of determining a chronically ordered sequence of conditions that lead to death (named the causal sequence of death), based on the decedent's most recent hospital discharge record. The objective of this approach is to provide medical professionals with assistance in accurately reporting causes of death. The most important aspect of the design is to learn the causal relationship among clinical codes and to recognize conditions that are associated with death. There are three obstacles to overcome: the variety of clinical coding systems, the limitation of medical domain knowledge, and the inability to interoperate data. First, we generate sequences of causes of death by applying neural machine translation models with a variety of attention mechanisms. When assessing the quality of generated sequences, we make use of the BLEU (BiLingual Evaluation Understudy) score in conjunction with three different accuracy metrics. When generating the causal sequences of death, the second thing we do is incorporate medical domain knowledge that has been verified by experts as constraints. In the final step of this project, we create a Fast Healthcare Interoperability Resources (FHIR) interface, which demonstrates the applicability of this work in clinical settings. Our findings are consistent with the most recent reporting and have the potential to assist medical professionals and other specialists in times of public health emergencies such as the COVID-19 pandemic.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here