Integer-valued functional data analysis for measles forecasting.

Area of Science:

• Epidemiology
• Statistical Modeling
• Public Health

Background:

• Measles is highly contagious, posing an imminent threat due to declining vaccination rates.
• Rising measles cases necessitate improved forecasting methods for public health preparedness.

Purpose of the Study:

• To develop a novel methodology for forecasting measles counts with quantified uncertainty.
• To enhance preparedness against potential measles outbreaks by improving predictive models.

Main Methods:

• Modeling disease counts as an integer-valued functional time series using a negative-binomial distribution.
• Employing a real-valued latent process with basis expansion and dynamic coefficients to capture complex seasonality.
• Utilizing a fully Bayesian approach for robust uncertainty quantification.

Main Results:

• The proposed framework accurately models dynamic, year-to-year seasonality in measles cases.
• Achieved improved multi-month-ahead point forecasts and tighter forecast intervals compared to existing models.
• Provided well-calibrated and precise uncertainty quantification for epidemiological features, including peak timing.

Conclusions:

• The developed statistical model offers enhanced capabilities for forecasting infectious diseases like measles.
• Improved forecasting and uncertainty quantification are crucial for effective public health response to outbreaks.
• An R package is available to facilitate the application of this novel forecasting methodology.