Synthetic Test Data Generator • gsm.datasim

Overview

gsm.datasim generates synthetic clinical trial data for testing and development of clinical monitoring applications. It produces multi-snapshot longitudinal datasets across a configurable set of clinical domains (SDTM-style and custom) and can run configurable analytics and reporting pipelines on the generated data via the workr workflow engine.

Installation

You can install the latest release of gsm.datasim from GitHub with:

# install.packages("pak")
pak::pak("Gilead-BioStats/gsm.datasim@*release")

You can install the development version of gsm.datasim from GitHub with:

# install.packages("pak")
pak::pak("Gilead-BioStats/gsm.datasim")

Quick Start

library(gsm.datasim)

# Generate a standard 6-month study with analytics + reporting
study <- quick_longitudinal_study(
  study_name       = "DEMO-001",
  participants     = 200,
  sites            = 15,
  months_duration  = 6,
  study_type       = "standard"
)

# Access raw data, analytics, and reporting results
names(study)          # $study_id, $config, $raw_data, $analytics, $reporting
names(study$raw_data) # one entry per snapshot date

# Inspect the first snapshot
snap <- get_snapshot_data(study, 1)
nrow(snap$Raw_SUBJ)
nrow(snap$Raw_AE)

# Track a domain across all snapshots
ae_counts <- sapply(get_domain_timeline(study, "AE"), nrow)

Key Functions

Study generation

Function	Description
`quick_longitudinal_study()`	Single-call entry point: raw data + analytics + reporting
`create_longitudinal_study()`	Full control over domains, intervals, and pipelines
`create_multiple_longitudinal_studies()`	Generate a portfolio of studies in one call
`create_study_config()`	Build a config object for low-level control
`generate_study_data()`	Generate raw data from a config object

Study access helpers

Function	Description
`get_snapshot_data()`	Extract data for a specific snapshot
`get_domain_timeline()`	All snapshots for a single domain
`get_available_domains()`	List all domains present in the study
`summarize_longitudinal_study()`	Print a study summary

Pipelines

Function	Description
`run_longitudinal_analytics()`	Run (or re-run) the analytics pipeline (powered by `workr`)
`run_longitudinal_reporting()`	Run (or re-run) the reporting pipeline
`generate_analytics_layers()`	Run analytics on already-generated raw data
`generate_reporting_layers()`	Run reporting on analytics results

Export

Function	Description
`export_study_data()`	Write study to structured folder hierarchy (CSV or Parquet)

Common Patterns

Data generation only (no pipeline)

data_only <- create_longitudinal_study(
  study_id      = "DATA-001",
  participants  = 200,
  sites         = 15,
  snapshots     = 6,
  interval      = "1 month",
  domains       = c("AE", "LB", "VISIT", "QUERY"),
  run_analytics = FALSE,
  run_reporting = FALSE
)

Increased outlier prevalence

study <- quick_longitudinal_study(
  study_name        = "DEMO-OUTLIER-HIGH",
  participants      = 200,
  sites             = 15,
  months_duration   = 6,
  outlier_intensity = 2.5   # default is 1
)

Stepwise pipeline execution

# Step 1 — raw data
study <- create_longitudinal_study(
  study_id      = "STEP-001",
  participants  = 100,
  sites         = 10,
  snapshots     = 4,
  interval      = "1 month",
  domains       = c("AE", "LB", "VISIT"),
  run_analytics = FALSE,
  run_reporting = FALSE
)

# Step 2 — add analytics
study <- run_longitudinal_analytics(study)

# Step 3 — add reporting
study <- run_longitudinal_reporting(study)

Low-level config API

config <- create_study_config(
  study_id          = "CUSTOM-001",
  participant_count = 300,
  site_count        = 20,
  analytics_package = "gsm.kri"
) |>
  set_temporal_config(start_date = "2023-01-01", snapshot_count = 12, snapshot_width = "months") |>
  add_dataset_config("Raw_AE",    enabled = TRUE) |>
  add_dataset_config("Raw_LB",    enabled = TRUE) |>
  add_dataset_config("Raw_VISIT", enabled = TRUE)

raw_data  <- generate_study_data(config)
analytics <- generate_analytics_layers(raw_data, config)
reporting <- generate_reporting_layers(analytics, config)

Multiple studies

studies <- create_multiple_longitudinal_studies(
  study_names   = c("TRIAL-001", "TRIAL-002", "TRIAL-003"),
  participants  = 200,
  sites         = 12,
  snapshots     = 6,
  domains       = c("AE", "LB", "VISIT", "PD"),
  run_analytics = TRUE
)

Export

export_study_data(
  study      = study,
  output_dir = "./output",
  format     = "parquet",   # or "csv" (default) or "both"
  overwrite  = TRUE
)

Examples

Full worked examples are in the Examples section of the website:

example_demo.html — complete walkthrough of all major features
example_longitudinal.html — longitudinal data generation patterns
example_domain_registry.html — extending the Domain Registry

Domain Registry

The Domain Registry is an extensible system for per-domain data generation. Each entry defines how data should be generated for a single Raw_* dataset — independently testable and overridable without modifying core generation code.

registry <- get_domain_registry()
cat("Registry-backed domains:", paste(names(registry), collapse = ", "), "\n")

Domains not yet in the registry are handled by a legacy dispatcher and can be migrated incrementally.