LoginSign Up
← Back to all sparks
D

dcurves

ANALYTICS
Velocity0.0

Decision Curve Analysis for Model Evaluation

github.com/ddsjoberg/dcurves ↗

Decision curve analysis, settled since 2022 and now moving only when its neighbours do

clinical-predictiondecision-analysisnet-benefitbiostatisticsr-package
◆Current state
dcurves implements decision curve analysis — evaluating a prediction model or diagnostic test by net benefit across the range of thresholds a clinician might plausibly use, rather than by a single discrimination statistic. Its API stabilised in 2022 around dca() and test_consequences(). The two releases since exist because gtsummary and CRAN documentation rules changed, not because the method did.
◆Where it's heading
The package reached its intended scope quickly and then stopped. Its 2022 releases did the substantive work: adding threshold-level diagnostic accuracy, tightening argument validation, and taking one breaking change to make net-interventions-avoided plots show the treat-all and treat-none reference lines by default. Since then it has moved only as a dependent of the wider tidy-modelling documentation ecosystem it plugs into.
◆Prediction
Nothing in these entries points to method or API work; expect the next release to be another compatibility or CRAN documentation patch.

◆Recent moves

  1. 9mo ago

    HTML5 documentation rebuild for CRAN

    Regenerates documentation as HTML5 to satisfy a CRAN requirement. Archive-avoidance maintenance, and the package's only release in over a year.

    View source ↗
  2. 2y ago

    broom.helpers added to Suggests for gtsummary 2.0

    Adds broom.helpers to Suggests to keep working with gtsummary v2.0. A dependency accommodation carrying a minor version number despite changing nothing in the package itself.

    View source ↗
  3. 3y ago

    Net interventions plots now show treat-all and treat-none by default

    A breaking default change: net-interventions-avoided figures include the treat-all and treat-none reference lines, and nper drops to one. Both make the plot harder to misread on its own, which matters for a method whose output is usually shown to a clinical audience.

    View source ↗
  4. 4y ago

    test_consequences() reports accuracy across thresholds

    Adds test_consequences() for diagnostic accuracy at multiple thresholds, allows zero and one as threshold values, and renames the prevalence column to pos_rate. The release that completed the package's analytical surface.

    View source ↗