Bayesian nowcasting during the STEC O104:H4 outbreak in Germany, 2011.

Area of Science:

• Epidemiology
• Biostatistics
• Public Health Surveillance

Background:

• Real-time public health surveillance requires accurate prediction of occurred-but-not-yet-reported events.
• The 2011 German Shiga toxin-producing Escherichia coli (STEC) O104:H4 outbreak highlighted the need for effective outbreak prediction models.

Purpose of the Study:

• To develop a Bayesian approach for predicting daily hospitalizations during an outbreak.
• To evaluate and compare predictive models using proper scoring rules for count data.

Main Methods:

• A novel Bayesian approach using negative binomial sampling for count data.
• Incorporation of right-truncated reporting delay distributions using generalized Dirichlet distributions.
• Development of a hierarchical model combining survival regression and penalized splines for epidemic dynamics.

Main Results:

• The Bayesian approach effectively predicted hospitalizations, accounting for the count nature of time series data.
• Significant changes in the reporting delay distribution were observed, linked to intervention measures.
• Hierarchical modeling improved the analysis by adapting to evolving outbreak dynamics.

Conclusions:

• Bayesian nowcasting is a valuable tool for real-time trend analysis in time-critical outbreaks.
• Accurate prediction requires accounting for data characteristics and dynamic changes in reporting delays.
• Intervention measures can alter reporting patterns, necessitating adaptive modeling approaches.