meddecide

The clinical-decision sub-module — fourteen analyses covering test performance, agreement, and three capabilities nobody had written down: agreement sample-size planning, sequential testing, and analysis with no gold standard.

Purpose

Answer the three questions that arise whenever a new test, biomarker, or AI model is evaluated: how good is the test, does using it help the patient, and do observers agree.

Data used

Any dataset with a test result and a reference standard, or with multiple rater columns.

Methods

Package meddecide, version 1.0.1, 14 analyses, 55 vignettes.

Analysis What it is
decision Medical Decision
decisioncalculator Medical Decision Calculator
decisioncompare Compare Medical Decision Tests
decisioncombine Combine Medical Decision Tests
enhancedROC Clinical ROC Analysis
psychopdaROC Advanced ROC Analysis
agreement Interrater Reliability
kappasizeci Sample size — CI approach
kappasizepower Sample size — power approach
kappasizefixedn Lowest expected value for fixed n
cotest Co-Testing Analysis
sequentialtests Sequential Testing Analysis
nogoldstandard Analysis Without Gold Standard
lassologistic LASSO Logistic Regression

The agreement claims all hold. Checked in agreement.a.yaml: weighted kappa with unweighted / linear / quadratic options, ICC, CCC, and Bland-Altman with limits of agreement, a proportional-bias test and a plot. This module genuinely is the agreement engine Interobserver Agreement describes.

Decision-curve analysis is here, but not where you would look for it. It has no standalone analysis: psychopdaROC performs “clinical utility analysis including net benefit curves, decision curve analysis, and clinical impact assessment” with a treatment-threshold range, and enhancedROC adds clinical-impact metrics (NNT, NND) and net-benefit regression. The umbrella adds three standalone variants — decisioncurve, bayesdca (Bayesian) and timedependentdca. So Decision Curve Analysis’s attribution to this module is correct; that page now also records where in the module it lives and what the umbrella adds.

Three capabilities that were not on this page

Each answers a question already open elsewhere in this wiki:

  • kappasizeci / kappasizepower / kappasizefixedn — sample-size planning for agreement studies, three approaches. Interobserver Agreement describes several planned and running agreement studies and nothing anywhere records how any of them were powered. The tool for doing it properly is already installed.
  • nogoldstandard — Analysis Without Gold Standard. This is the latent-class situation that arises constantly in pathology, where the reference standard is itself a fallible reader. Diagnostic Test Accuracy and Interobserver Agreement both circle the problem; this is a direct implementation.
  • sequentialtests and cotest — sequential and parallel multi-test strategies, which is what a real diagnostic pathway looks like rather than a single test in isolation.

Current state / open questions

Active; last commit 2026-07-16 at version 1.0.1, in step with the other five. Used as the agreement engine across several projects (HER2 Intratumoral Heterogeneity, Omentum, Konsultasyon).

Documented pitfalls: PPV and NPV depend on prevalence, so report sensitivity/specificity and predictive values together with the cohort prevalence; quadratic versus linear kappa weights give materially different values on ordinal scales — pick one and document it; decision curves are not ROC curves and should not be read as “higher line is always better”. The chapter explicitly recommends this module rather than raw accuracy for framing AI model versus pathologist comparisons.

  • Were any of the group’s agreement studies powered using the kappasize* analyses? Nothing records it, and the pilot-then-full-study pattern in HER2 Intratumoral Heterogeneity is exactly where it would apply. [unverified]
  • Has nogoldstandard been used? For Konsultasyon, where “the referring lab’s diagnosis” is a fallible reference rather than truth, it is arguably the correct analysis and no page mentions it. [unverified]

Related: ClinicoPath jamovi Module — the umbrella, which adds the three standalone decision-curve analyses. Interobserver Agreement and Decision Curve Analysis — the two concept pages this module underpins, both updated from this reading.

Derived from: repository working tree read 2026-07-26 — DESCRIPTION, the 14 jamovi/*.a.yaml files, jamovi/agreement.a.yaml, jamovi/psychopdaROC.a.yaml, jamovi/enhancedROC.a.yaml, vignettes/, plus git ls-files/git log. Earlier content derived from jamovi/meddecide.qmd.