Nowcasting influenza epidemics using non-homogeneous hidden Markov models.

Area of Science:

• Epidemiology
• Public Health Surveillance
• Biostatistics

Background:

• Timeliness is crucial for public health surveillance systems, particularly for infectious diseases like influenza.
• Current influenza surveillance in Europe relies on sentinel networks, with weekly bulletins often delayed.
• Nowcasting, predicting the present situation with incomplete data, is of high public health interest.

Purpose of the Study:

• To develop and evaluate a non-homogeneous hidden Markov model (HMM) for nowcasting weekly influenza-like illness (ILI) rates.
• To assess the model's ability to predict current ILI rates and epidemic probabilities.
• To improve the timeliness of public health surveillance for influenza activity.

Main Methods:

• Development of a non-homogeneous hidden Markov model (HMM).
• Utilized weekly early observations of ILI incidence and positive test counts as covariates.
• Employed Bayesian inference for parameter estimation and nowcasting.

Main Results:

• The non-homogeneous HMM demonstrated added value compared to a homogeneous HMM.
• The model successfully nowcasted the current week's ILI rate and epidemic state probability.
• Implementation with the Portuguese influenza surveillance system showed improved timeliness by two weeks.

Conclusions:

• Non-homogeneous hidden Markov models offer a significant advancement for influenza nowcasting.
• This approach enhances the timeliness of public health surveillance systems.
• The developed model provides valuable insights for managing influenza epidemics.