Advancing Underlying Cause of Death Inference Through Wide and Deep Model

Area of Science:

• Public Health
• Medical Informatics
• Artificial Intelligence

Background:

• Accurate death certificates are crucial for effective death surveillance.
• Manual determination of underlying cause of death can be imprecise.
• Existing methods require enhancement for improved reliability.

Purpose of the Study:

• To investigate the Wide and Deep framework for improving accuracy in inferring the underlying cause of death.
• To enhance the reliability of cause-of-death determination in surveillance systems.
• To leverage artificial intelligence for more precise public health data.

Main Methods:

• Analysis of 403,547 death reports from Fujian Province (2016-2022).
• Development of a Wide and Deep model embedded with Convolutional Neural Networks (CNN).
• Performance evaluation using weighted accuracy, precision, recall, and AUC, compared against XGBoost, CNN, GRU, and Transformer models.

Main Results:

• The Wide and Deep model achieved high performance: 95.75% precision, 92.08% recall, 93.78% F1 Score, and 95.99% AUC.
• The model demonstrated strong performance across different cause-of-death chain lengths, with F1 Scores ranging from 79.50% to 97.13%.
• Outperformed other benchmark models in accuracy and reliability.

Conclusions:

• The Wide and Deep framework significantly improves the determination of underlying causes of death.
• This AI-driven approach offers a valuable tool for enhancing cause-of-death surveillance quality.
• Integration of AI is expected to streamline death registration and reporting, boosting public health data precision.