Cohorts + Shared Resources

For the class example, we use Strategus test assets built around the GI bleed question.

• Cohorts.csv + SQL cohort definitions (JSON optional)
• No negative control outcomes required for today’s workflow

These become sharedResources in the specification.

Core Question in the Example

Example framing in the guidance:

• New users of celecoxib vs diclofenac
• Outcome: GI bleed

This gives us concrete IDs and module-ready assets.

Quick Note: R6 Classes

Strategus modules use R6, R’s object-oriented class system. If you haven’t seen this before, a few things to know:

• You create an object with $new() instead of calling a plain function
• The object has methods you call with $ (like a named list)
• Each module object is a self-contained toolkit for that analysis type

This is different from most R code but the pattern is always the same.

R6 in Practice

# Create an object (like a factory for that module type)
cdModule <- CohortDiagnosticsModule$new()

# Call methods on it with $
cdModule$createModuleSpecifications(...)
cdModule$createCohortSharedResourceSpecifications(...)

# Inspect what methods are available
ls(cdModule)

You never need to define your own R6 classes. Just use the ones Strategus provides.

High-Level Module Pattern

For each module:

1. Instantiate module object with $new()
2. Call $createModuleSpecifications(...) to build settings
3. Add to analysis specification

Know Your Defaults

Every createModuleSpecifications() call has defaults. Inspect them before accepting:

formals(cdModule$createModuleSpecifications)

For your assignment, you must list all parameters explicitly and annotate which are defaults vs overrides.

See the Strategus reference docs and underlying HADES package docs (CohortDiagnostics, CohortIncidence, Characterization).

Modules Used in Today’s Example

• CohortGeneratorModule
• CohortDiagnosticsModule
• CohortIncidenceModule
• CharacterizationModule

This aligns directly with Part 1 assignment scope.

CohortGenerator + Diagnostics

We include:

• Cohort generation stats
• Inclusion stats
• Source concepts and orphan concepts
• Incidence rate and cohort relationship diagnostics

Incidence + Characterization

We also add:

• Incidence design (targets, outcomes, TAR windows, strata)
• Characterization settings for selected target/outcome cohorts

Build the Full Analysis Specification

Composition order in code:

1. Add cohort shared resources
2. Add module specs (generator, diagnostics, incidence, characterization)
3. Save with ParallelLogger::saveSettingsToJson()

Scope Note for Class + Assignment

For this course workflow, negative controls are optional and not required.

Deliverable Artifact

Main design output:

• inst/settings/studyAnalysisSpecification.json

This is the primary object you submit for design readiness.

Execution Is Separate

After the analysis specification is built:

• Create execution settings JSON
• Connect to Eunomia
• Run execute()

For assignment work, execution is optional; specification quality is required.

In-Class Script

modules/06_strategus-gibleed-demo/strategus-gibleed-demo.R

Run in sections and inspect each object as it is created.

Please note that I created a project file in this directory, so you should open the project to run this code to ensure relative paths are correct.

Open directly with:

• strategus-gibleed-demo.Rproj (RStudio) or
  
• strategus-gibleed-demo.code-workspace (VS Code/Positron)

From Demo to Part 1 Submission

Use this translation:

1. Replace example cohorts with your study cohorts
2. Keep the same module assembly pattern
3. Save your Part 1 analysis specification JSON
4. Include clear rationale for diagnostics/characterization/incidence choices

Part 1 Minimum Checklist

Before submission, verify:

1. JSON compiles and saves cleanly
2. Cohort assets are organized and consistent
3. Diagnostics, characterization, and incidence specs are present
4. Repository is easy to review

Reminder on Dataset Choice

Do not use GiBleed for your assignment submission. Choose a different Eunomia dataset (i.e., MIMIC or Synthea) or your own study cohort definitions.