LoginSign Up
← Back to all sparks
C

cfrnow

ANALYTICS
Velocity5.0

Real-time case fatality ratio from a line list via a Bayesian mixture-cure survival model

github.com/epiforecasts/cfrnow ↗

A Bayesian real-time CFR estimator that now ships stratified fits and posterior-predictive checks

epidemiologybayesian-modellingcfr-estimationr-packagemodel-checking
◆Current state
cfrnow estimates case fatality ratios from line-list data while an outbreak is still running, using a Bayesian mixture-cure survival model registered as an `epidist` model type. Three releases in a month took it from first public release to stratified, partially-pooled fits with posterior-predictive checking. It leans on `distspec` for delay parameterisation, which reached CRAN alongside the 0.2.1 patch.
◆Where it's heading
The arc runs from producing a single corrected CFR number toward supporting a full model-checking workflow. 0.2.0 added the pieces a modeller needs to defend an estimate: replicate line lists replayed through the real-time truncation, an ascertainment-ratio correction for when fatal and non-fatal cases enter the line list at different rates, and per-group CFRs from `brms` formulas. Delay coverage widened from LogNormal and Gamma to Weibull in the same release.
◆Prediction
Expect the next release to keep widening covariate and pooling support rather than adding new outcome types, since every 0.2.0 addition extended the existing formula interface instead of replacing it.

◆Recent moves

  1. 5d ago

    distspec dependency moves to CRAN

    A no-behaviour-change patch that drops `distspec` from `Remotes` now that it is on CRAN. It closes out the dependency work 0.2.0 started rather than adding anything users can call.

    View source ↗
  2. 5d ago

    Stratified CFR fits, Weibull delays, posterior-predictive checks

    The release that turns a CFR estimate into a checkable one: posterior-predictive replication via `pp_check_cfr()`, an ascertainment-ratio correction in `summary()`, and per-group CFRs from intercept-free formulas such as `cfr ~ 0 + group`. Weibull joins LogNormal and Gamma as an onset-to-death delay, and a new vignette walks the no-, complete- and partial-pooling fits.

    View source ↗
  3. 1mo ago

    First release: real-time CFR from a Bayesian mixture-cure model

    ⚡ SPARK

    The first release, and the one that sets the package's shape: a Bayesian mixture-cure survival model registered as an `epidist` model type, so both the CFR and the onset-to-death delay accept `brms` formulas. Everything in 0.2.0 extends this interface rather than replacing it.

    View source ↗