Dynamic Bayesian network in infectious diseases surveillance: a simulation study.

Area of Science:

• Epidemiology and Public Health
• Computational Biology and Bioinformatics
• Statistical Modeling

Background:

• Infectious disease surveillance requires identifying dynamic relationships between diseases and influencing factors.
• Challenges include small sample sizes and delayed effects, hindering accurate identification.
• Existing methods may struggle with complex, real-world epidemiological data.

Purpose of the Study:

• To evaluate the performance of dynamic Bayesian networks (DBN) for infectious disease surveillance.
• To compare DBNs against Granger causality and LASSO methods in identifying dynamic relations.
• To assess DBNs' ability to enhance infectious disease forecasting.

Main Methods:

• Conducted two simulations adapted from real-world scenarios, incorporating nonlinearity and nuisance variables.
• Compared DBN performance with Granger causality test and LASSO method using varying sample sizes.
• Assessed DBN's impact on forecasting accuracy for infectious diseases.

Main Results:

• With large sample sizes, DBN showed higher true positive rates and significantly lower false positive rates than Granger causality and LASSO.
• Small sample sizes, especially with nonlinearity and nuisance variables, reduced DBN's true positive rate, a challenge also observed in comparative methods.
• At least three years of weekly data are crucial for reliable dynamic relation identification in surveillance.
• DBN improved infectious disease forecasting by reducing errors by 7%.

Conclusions:

• Dynamic Bayesian networks are recommended for enhancing infectious disease surveillance quality.
• DBNs demonstrate superior performance in identifying dynamic disease relations, particularly with sufficient data.
• The study highlights the importance of data quantity and quality for accurate epidemiological modeling.