The development repository for thedisbayes R packagefor chronic disease epidemiology estimation with incomplete data.
disbayes can estimate age-specific case fatality fora disease, given:
published information on age-specific mortality and at least oneof incidence or prevalence
some indication of the uncertainty associated with the publishedestimates, either as a credible interval, or by expressing the estimateas a number of cases with associated denominator.
The underlying model is a three-state multi-state model withstates given by no disease, disease and death. Remission from thedisease is optional.
Case fatality, incidence or remission rates can be modelled assmooth functions of age, through a spline model, or estimatedindependently for each age. Case fatality or remission can also bemodelled as age-constant.
Two alternative estimation methods can be used, both based on theStan software.
exact point estimation using optimisation to obtain the posteriormode, with credible intervals based on an approximation to the Bayesianposterior. This is generally instant to compute, but the uncertaintyquantification is approximate.
full Bayesian estimation using Markov Chain Monte Carlo. Thisgives more accurate uncertainty quantification but is computationallyintensive.
The following more advanced models are provided, which are allmore computationally intensive:
hierarchical models for data by age and area, which shareinformation between areas to strengthen estimates from areas with lessdata
hierarchical models for data by age, area and gender, where theeffect of gender is assumed to be the same for every area
models with assumed trends in disease incidence or case fatalitythrough calendar time, where trends can be age-specific(non-hierarchical models only)
It is inspired by theDisMod IIandDisMod-MR packagesused for the Global Burden of Disease studies. It modifies and extendsthe formal, fully Bayesian framework described in thebookby Flaxman et al..
The method is fully described inJackson etal. (2023).
Source code is at theGitHubrepository
install.packages("disbayes")install.packages("devtools")# if devtools not already installedlibrary(devtools)install_github("chjackson/disbayes")If this fails, make sure that therstan package is setup properly, asexplainedhere. If you are on Windows, then follow these instructions forinstallingrstan from source on Windows.
Bayesianestimation of chronic disease epidemiology from incomplete data: thedisbayes package