Skip to contents

Feature Comparison

Source: vignettes/articles/feature_comparison.Rmd
feature_comparison.Rmd

This article compares selecta with other R packages for generating clinical study flow diagrams. The R ecosystem includes several packages targeting specific EQUATOR guidelines, as well as general-purpose flowchart tools. The following comparison identifies areas of overlap and distinction.

Similar Packages

Package Primary Focus
selecta Multi-guideline enrollment diagrams (CONSORT, STROBE, STARD, PRISMA, MOOSE)
consort CONSORT diagrams from disposition data
flowchart Tidyverse-style participant flow diagrams
dtrackr Data pipeline tracking with CONSORT/STROBE output
ggconsort CONSORT diagrams with ggplot2
PRISMA2020 Interactive PRISMA 2020 flow diagrams
PRISMAstatement Simple PRISMA flow charts (pre-2020 template)
stard STARD diagnostic-accuracy flow diagrams
prismadiagramR PRISMA diagrams via DiagrammeR
metagear Meta-analysis toolkit with plot_PRISMA() function
consortr Shiny app for interactive CONSORT creation

Core Feature Matrix

Feature selecta consort flowchart dtrackr ggconsort PRISMA2020
CONSORT (randomized trials) ✓ ✓ ✓ ✓ ✓ —
STROBE (observational cohorts) ✓ ◐ ◐ ✓ — —
STARD (diagnostic accuracy) ✓ — — — — —
PRISMA (systematic reviews) ✓ — — — — ✓
MOOSE (observational meta-analysis) ✓ — — — — ◐
Data-driven counts ✓ ✓ ✓ ✓ ✓ —
Manual counts ✓ ✓ ◐ — ✓ ✓
Multi-arm layouts (3+) ✓ ◐ ✓ ✓ ◐ —
Factorial (nested-split) designs ✓ ◐ — — — —
Multi-source convergence ✓ — — — — ✓
Split-and-recombine topology ✓ — — — — —
Phase labels ✓ ✓ — ◐ — ✓
Exclusion sub-reasons ✓ ✓ ◐ ✓ ◐ ✓
Per-arm exclusion reasons ✓ ◐ — ✓ — —
Hierarchical (nested) sub-reasons ✓ — — — — —
Cohort extraction ✓ — — ✓ ✓ —
Count-first display mode ✓ — — — — —
Auto-sized file export ✓ — ✓ — — —
Graphviz/DOT output ✓ ✓ — ✓ — —
Interactive HTML output† — — — — — ✓
Pipe-friendly API ✓ — ✓ ✓ ✓ —

Legend: ✓ Full support | ◐ Partial support | — Not available

† Currently, selecta DOT engine output is embeddable as an HTML widget via DiagrammeR::grViz(), but widget content is a static SVG without native tooltips or clickable hyperlinks.

Feature Definitions

Feature Description
CONSORT Flow diagrams for randomized controlled trials with allocation and per-arm follow-up
STROBE Flow diagrams for observational cohort studies with exposure stratification
STARD Diagnostic accuracy diagrams with inverted assessment labels and cross-classification
PRISMA Systematic review diagrams with multi-source identification and convergence
MOOSE Observational meta-analysis diagrams (structurally similar to PRISMA)
Data-driven counts Participant counts computed automatically from a dataset
Manual counts Participant counts supplied directly by the analyst
Multi-arm layouts (3+) Support for three or more parallel arms with automatic layout
Factorial (nested-split) designs Cross-classification by two factors via chained allocate()/stratify(), with parent boxes centered over their sub-arms
Multi-source convergence Parallel identification streams merging into a single screening flow
Split-and-recombine topology Fanning a population into strata for independent characterization, then converging to a single stream before the endpoint
Phase labels Vertical labels (e.g., Enrollment, Allocation, Follow-up, Analysis)
Exclusion sub-reasons Itemized counts within exclusion boxes (e.g., “Declined: 12, Other: 8”)
Per-arm exclusion reasons Different sub-reason breakdowns for each treatment arm
Hierarchical (nested) sub-reasons Two-level exclusion breakdowns grouping itemized sub-reasons under broader categories
Cohort extraction Return the analysis-ready dataset after all exclusions
Count-first display mode Bold count before label (e.g., “450 Drug A” instead of “Drug A, n = 450”)
Auto-sized file export Automatic dimension calculation when saving to PDF/PNG/SVG/TIFF
Graphviz/DOT output Export diagram as a Graphviz DOT string for external rendering
Interactive HTML output Clickable HTML widgets with tooltips and hyperlinks
Pipe-friendly API Pipeline construction with |> or %>% operators

Unique Strengths of selecta

The following features distinguish selecta from comparable packages:

Multi-Guideline Support

Most existing packages target a single EQUATOR guideline—typically CONSORT or PRISMA—requiring different tools for different study types. selecta provides a unified API across five guidelines (CONSORT, STROBE, STARD, PRISMA, MOOSE), with guideline-specific functions (allocate() vs. stratify(), assess(), sources(), combine()) that share common conventions for exclusions, phases, and rendering.

# Same rendering pipeline, different study types
enroll(n = 500) |> allocate(...) |> endpoint("Analyzed") |> flowchart()  # CONSORT
enroll(n = 500) |> stratify(...) |> endpoint("Analyzed") |> flowchart()  # STROBE
enroll(n = 500) |> assess(...) |> endpoint("Final") |> flowchart()       # STARD
sources(...) |> combine(...) |> endpoint("Included") |> flowchart()      # PRISMA

Declarative Pipeline Syntax

The pipeline reads top-to-bottom, matching the visual layout of the diagram and the narrative order in which investigators describe their enrollment process. Each function corresponds to a single structural element (enrollment, exclusion, phase, allocation, endpoint), making the code self-documenting.

Factorial and Recombination Topologies

selecta renders flow shapes that single-split tools cannot express. Two allocate() or stratify() calls in sequence produce a factorial (nested-split) design, cross-classifying each arm by a second factor with parent boxes centered over their sub-arms. combine() reconverges parallel streams, supporting both split-and-recombine layout (stratify, characterize, recombine) and the stepwise pooling of a crossed design; it may be applied more than once to collapse a factorial into a single analysis cohort.

# 2x2 factorial, then pool the second factor back into the first
enroll(n = 480) |>
    allocate(labels = c("Drug A", "Drug B"), n = c(240, 240)) |>
    allocate(labels = c("Vaccine", "Placebo"), n = c(120, 120, 120, 120)) |>
    combine("Pooled by antiviral") |>
    endpoint("Analyzed") |>
    flowchart()

Hierarchical Exclusion Reasons

Exclusion reasons may be itemized one level deep or grouped two levels deep, with broad categories subdividing into specific sub-reasons. A named list (manual mode) or a pair of columns (data mode) records the hierarchy, which both rendering engines display with sub-reasons indented or bulleted beneath each category.

enroll(n = 1000) |>
    exclude("Excluded", n = 250,
            reasons = list(
                "Ineligible"     = c("Out of age range" = 70, "Comorbidity" = 55),
                "Declined"       = c("Time commitment" = 40, "Travel burden" = 20),
                "Administrative" = 65)) |>
    endpoint("Enrolled") |>
    flowchart()

Cohort Extraction

The specialized cohort() function returns the dataset remaining after all exclusion criteria have been applied, enabling a seamless transition from diagram construction to statistical analysis.

flow <- enroll(data, id = "patient_id") |>
    exclude("Ineligible", criteria = eligible == FALSE) |>
    allocate("treatment") |>
    endpoint("Analyzed")

# Render the diagram
flowsave(flow, "consort.pdf")

# Extract the analysis-ready cohort
final_data <- cohort(flow)

Performance

Built on data.table with memoized text measurement, vectorized box and edge drawing, and single-pass layout computation. Diagram rendering completes in under one second for typical clinical trial configurations.

Additional Resources