Show Clinical Cohort Data at Their Best • vernissage

Documentation: https://clement-poupelin.github.io/vernissage/

Put your cohort on show. A vernissage is the opening of an exhibition, when the works are hung and shown at their best. This package prepares clinical cohort data and shows them: overview table, distributions, missing-data maps, paired visits, correlations, PCA, radar and forest plots, with a publication theme, colour-blind safe palettes and polished gt tables.

Data are in wide format (one row per subject, one column per variable and visit, names such as NFL_V0, NFL_V3).

Preparation: French CSV import, name cleaning, visits and long format, bilateral measures, robust outlier rules, log / Box-Cox transformations decided at baseline and applied at follow-up, composite scores.
Description and figures: describe_cohort(), plot_na_map(), plot_distribution(), plot_paired_visits(), plot_trajectories(), plot_cor_heatmap(), plot_pca_*(), plot_radar(), plot_forest().
Reports: number and p-value formatting, gt publication tables with significance highlighting, Quarto tabsets and callouts.

Companion packages:

lodestar: limits of detection and cut-offs (median split, tertiles, reclassification between visits);
statistical modelling lives in a separate package.

Installation

# install.packages("remotes")
remotes::install_github("Clement-Poupelin/vernissage")

Optional features use suggested packages:

install.packages(c("gt", "car", "DT", "ggrepel", "nortest", "stringi", "RColorBrewer"))

Quick start

library(vernissage)

# 1. the cohort at a glance
describe_cohort(cohort_sim, vars = c(AGE, SEX, NFL_V0, MOCA_V0, FALL_V0),
                by = GROUP, id_col = ID)
plot_na_map(cohort_sim, vars = c(NFL_V0:CRP_V3), id_col = ID)

# 2. preparation: decided at V0, applied at V3
choice <- transform_choose(transform_compare(cohort_sim, c(NFL_V0, CRP_V0)))
cohort <- transform_apply(cohort_sim, transform_propagate(choice))$data
outlier_detect(cohort, c(NFL_V0, CRP_V0), id_col = ID, rule = "any")

# 3. figures
plot_distribution(cohort, c(NFL_V0, MOCA_V0, SEX), group = GROUP)
plot_paired_visits(cohort, c(NFL_V0, MOCA_V0), id_col = ID, transitions = TRUE)
plot_radar(cohort, c(AGE, NFL_V0, IL6_V0, MOCA_V0, UPDRS_V0), group = GROUP)

Columns can be given as strings, character vectors stored in a variable, bare names or tidyselect helpers (starts_with("NFL")).

Figures use Unicode symbols (rho, Delta, >=). To keep them in a PDF, save with ggsave("figure.pdf", p, device = cairo_pdf); the classic pdf() device gets ASCII equivalents automatically (see ?vernissage, option vernissage.unicode).

Tutorials

Getting started: the example cohort, a first look, the families of functions
Preprocessing: import, names, visits, outliers, transformations
Describing a cohort: overview table, missing data, correlations, PCA
Figures: distributions, paired visits, trajectories, radar and forest plots
Tables and Quarto reports: formatting, gt tables, tabsets

Coming from the analysis scripts

The former function names (get_label(), paired_visits_plot(), radar_plot()…) still work, with a deprecation warning; see ?vernissage-deprecated. The former limit-of-detection and cut-off functions are in lodestar.