Methods in Practice
01
irrCAC (v1.4, 2026-04-27)
An R package for analysing agreement among multiple raters using several chance-corrected coefficients, including Fleiss’ and Conger’s kappa, Gwet’s AC1/AC2 and Krippendorff’s alpha.
Fleiss' kappa vs Gwt's AC
We used irrCAC in a simulated multi-reader study to explore how prevalence can produce surprisingly low kappa values despite high observed agreement.
02
Fairmodels (V1.2.2., 2025.11.30)
fairmodels is an R package for evaluating and visualizing fairness in predictive models. It builds on the DALEX explainability framework and compares model behaviour across protected and privileged groups using multiple parity-based fairness metrics.
Fairness assessment in clinical prediction
We used fairmodels to compare subgroup-level behaviour of logistic regression and random forest models in a clinical prediction example. The analysis examined multiple parity metrics simultaneously and explored how changing the classification threshold altered the fairness profile.
03
Boot (V1.3-32, 2025.08.29.)
boot is an R package for bootstrap-based statistical inference. It provides flexible tools for generating bootstrap resamples, estimating sampling distributions and quantifying uncertainty without relying solely on analytical approximations.
Fairness assessment in clinical prediction
We used boot to generate 3,000 bootstrap estimates of prediabetes prevalence in the analysed NHANES sample. The resulting empirical distribution was used to illustrate sampling uncertainty and to derive a 95% bootstrap interval around the prevalence estimate.
04
limma (3.68.5)
limma is a Bioconductor package for analysing high-dimensional gene-expression data using linear models and empirical Bayes methods. Its moderated statistical approach can improve the stability of inference when many features are analysed simultaneously.
Multiplicity in high dimentional data
We used limma in our TCGA-BRCA multiplicity analysis to compare conventional feature-wise testing with an empirical Bayes approach. The example showed how the statistical framework used for large-scale testing can substantially affect the number and ranking of detected signals.
05
TCGAbiolinks (2.40.0)
TCGAbiolinks is a Bioconductor package for accessing, preparing and analysing data from the NCI Genomic Data Commons. It supports reproducible retrieval and preprocessing of TCGA datasets and provides tools for several common downstream genomic analyses.
Multiplicity in high dimentional data
We used TCGAbiolinks to retrieve and prepare TCGA-BRCA RNA-sequencing data for our multiplicity case study. The resulting dataset provided a realistic high-dimensional setting for exploring false positives, multiple-testing correction and empirical Bayes modelling.
06
survival (v3.8-12, 2026.09.09.)
survival is one of the core R packages for time-to-event analysis. It provides functions for survival objects, Kaplan–Meier estimation, Cox proportional hazards regression and more advanced extensions such as stratified and multi-state models.
From subgrouops to precision medicine
We used survival in both our legacy-data harmonization and subgroup-analysis examples. Cox models were used to estimate treatment effects across individual studies, pooled patient-level data and clinically defined subgroups, and to model continuous treatment-effect modification.
07
rtables (v0.6.17, 2026.10.01)
rtables is an R framework for constructing complex, structured reporting tables with flexible layouts and cell-level formatting. It was designed with clinical-trial reporting requirements in mind and supports programmatically accessible values together with presentation-ready output.
Beyond TLG generation
We used rtables in our Beyond TLG Generation example to translate a predefined laboratory table shell into a reproducible reporting object. The package allowed the reporting specification—rows, columns, summaries and formatting—to be separated from the underlying analysis data and regenerated consistently.
08
xportr (v0.6.0, 2026.09.01)
xportr provides tools for applying metadata to clinical datasets and exporting them as SAS XPT transport files. It supports attributes such as variable types, lengths, labels and formats and is designed for clinical and submission-oriented data workflows.
Beyond TLG generation
We used xportr in our executable clinical-reporting example to export the derived ADLB-style laboratory dataset to XPT format. This demonstrated how an R-based reporting workflow can extend beyond TLG generation to include submission-oriented data artifacts.
