EpiFilter

EpiFilter estimates the time-varying effective reproduction number (R) of infectious diseases from real-time and retrospective incidence data using recursive Bayesian smoothing inspired by Kalman filtering to infer transmission dynamics.

Key Features:

• Recursive Bayesian Smoothing: Employs a recursive Bayesian smoother to estimate R using all available incidence data for both real-time and retrospective inference.
• Forward Filtering and Backward Smoothing: Implements forward filtering (epiFilter or epiFilterSm) for causal R estimates and backward smoothing (epiSmooth) to incorporate future data for retrodictive estimates.
• Hidden Markov State Process Modeling: Models R as a flexible hidden Markov state process and solves the forward-backward algorithms exactly to integrate past and future incidence information, unifying elements of EpiEstim and Wallinga-Teunis methods.
• Statistical Robustness: Minimizes bias and variance in R estimates by maximizing information use while minimizing change-point or window-size assumptions, improving stability during low-incidence periods.
• Real-Time Inference and Predictive Capabilities: Supports real-time inference of time-varying transmission and provides one-step-ahead predictions via the recursPredict function.

Scientific Applications:

• Public Health Surveillance: Provides improved estimation of R for public health research and policy-making across various spatial scales.
• Low-Incidence Monitoring: Enables reliable R estimation and monitoring during low-incidence periods where other methods may be destabilized.
• Intervention Evaluation: Assists evaluation of control measures by estimating temporal changes in transmission.
• Real-Time Outbreak Response: Supports timely decision-making through real-time analysis and short-term prediction of transmission trends.

Methodology:

Uses recursive Bayesian smoothing inspired by Kalman filtering; forward filtering (epiFilter or epiFilterSm) and backward smoothing (epiSmooth); models R as a hidden Markov state process solved by exact forward-backward algorithms; and generates one-step-ahead predictions via recursPredict.