study-design

name: study-design description: > Design an epidemiological study using observational data. Use when planning a new analysis, defining research questions, identifying exposures, outcomes, and confounders, or structuring a study protocol. Covers cohort, cross-sectional, and ecological designs common in infectious disease epidemiology. disable-model-invocation: true user-invocable: true

Epidemiological Study Design

Work through each section with the user, asking clarifying questions at each step.

1. Research question

Frame the question using the PECO structure:

• Population: Who or what is being studied?
• Exposure: What factor, intervention, or characteristic is of interest?
• Comparison: What is the reference group?
• Outcome: What is being measured?

Clarify the study aim:

• Descriptive: Characterise patterns (no hypothesis)
• Exploratory: Identify associations (hypothesis-generating)
• Confirmatory: Test a pre-specified hypothesis

Distinguish:

• Prediction: Best forecast of outcome (variable selection by predictive accuracy)
• Association: Relationship after adjustment (variable selection by confounding structure)
• Causal inference: Effect of intervention (requires DAG, consider /dag-reasoning)

Draft a one-sentence research question.

2. Study design classification

Classify using standard epidemiological designs:

For forecast evaluation: frame as "retrospective observational study of forecast performance" where the unit of observation is individual forecasts or forecast-target pairs.

See references/study-designs.md for detailed guidance on each design.

State inherent limitations of the chosen design.

3. Variable roles

List all variables and classify each:

• Outcome (Y): What is measured. Define precisely, including scale and scoring rule if applicable
• Exposure (X): The factor of interest. Define categories and classification criteria
• Confounders (C): Common causes of X and Y. Must satisfy: (1) associated with exposure, (2) associated with outcome, (3) NOT on the causal pathway
• Effect modifiers (M): Variables that change the X-Y relationship
• Mediators: On the causal pathway; do NOT adjust unless decomposing direct/indirect effects
• Precision variables: Predictive of outcome but not of exposure; improve precision without affecting bias

For each confounder, justify inclusion with a brief causal argument. Consider using /dag-reasoning to formalise the causal structure.

4. Data requirements

Define:

• Unit of analysis: What each row in the dataset represents
• Inclusion criteria: Who/what is in the study (with explicit justification)
• Exclusion criteria: Who/what is excluded (with explicit justification)
• Data sources: Where each variable comes from
• Sample size: How many units are available; any power considerations

Consider:

• Selection bias from inclusion/exclusion criteria
• Measurement error in exposure classification
• Missing data: expected patterns and handling strategy
• Temporal alignment of exposure, confounders, and outcome

5. Analysis plan skeleton

Pre-specify:

• Primary analysis: Model family, link function, adjustment set
• Unadjusted analysis: Same model without confounders (for comparison)
• Subgroup analyses: Any planned stratifications
• Sensitivity analyses: Alternative specifications to probe assumptions
• Multiple comparisons: How handled if applicable

6. Output

Generate a structured study protocol in markdown covering sections 1-5. Flag decisions that will need explicit justification in the manuscript Methods section.