Authors: Jingxue Feng ([email protected]), Liangliang Wang ([email protected])
We propose a Beta-Dirichlet switching state-space transmission model to track underlying dynamics of disease and evaluate the effectiveness of interventions simultaneously. Bayesian inference are performed using a particle MCMC algorithm.
-
$y_t$ is the observed infectious proportion at time$t$ ; -
$x_t$ is a discrete latent variable describing the regime of model at time$t$ ; -
$p$ is the identification rate representing the fraction of infectious population identified by diagnosis; -
$\boldsymbol{\theta}_t = [S_t, E_t, I_t, R_t]^\top$ is the vector of the underlying proportion of the susceptible, exposed, infectious, and recovered population, where$S_t+E_t+I_t+R_t=1 \ \forall t$ ; -
$\alpha$ ,$\beta$ and$\gamma$ are the latency rate, transmission rate and recovery rate in the SEIR system respectively; -
$\lambda$ and$\kappa$ are the precision parameters controlling the variances (or randomness) for the observation and transition process; -
$f_{x_t}$ is the transmission rate modifier that changes with respect to the state variable$x_t$ . -
$\psi$ represents the set of model parameters to be estimated. -
$r(.)$ is the solution of the modified SEIR system starting at$\boldsymbol{\theta}_{t-1}$ .
This repository contains R code for simulation studies and real data analysis of BDSSS-SEIR model, described as follows:
- "Code" contains source R code for particle Gibbs samplers in simulation study and real data analysis.
- In "Simulation Study", the two-regime and three-regime cases are simulated and estimated.
- In "Real Data", the particle Gibbs sampler is implemented for K=1,2,3 and 4
- "Data" contains simulated data in two-regime and three-regime settings, as well as the real data retreived from here.