Comparative cohort using existing data
How do outcomes compare between exposure or treatment groups in existing data?
When to use this project
Use a comparative cohort when existing longitudinal data can identify alternative exposures or treatment strategies, baseline covariates and outcomes. State whether the aim is an adjusted association or a causal effect.
How it differs
A descriptive cohort summarizes one population’s experience. A comparative cohort adds an explicit contrast and requires attention to baseline comparability. Target trial emulation makes the hypothetical trial protocol and its mapping to observational data explicit; it is a framework that can strengthen a comparative cohort, not an unrelated data source.
Decisions before you start
- Align eligibility, exposure classification and follow-up start.
- Consider an active comparator and new-user design when they fit the clinical question.
- Prespecify confounders, overlap, outcome ascertainment and handling of treatment changes; adjustment cannot recover unmeasured information.
How the flows work together
Each flow below links to its diagram, steps and tables. Flow order organizes the project; it does not imply that all work is strictly sequential. Data assessment, analysis planning and protocol refinement may inform one another.
1. Comparative study overview
Overview · Question through comparative effect interpretation
Use this overview to state the purpose and link to detailed decisions. Keep a versioned protocol and decision register, and add evidence references for the rationale.
2. Sources and data preparation
Data · Assess coverage linkage and quality
A large dataset can still omit the clinical variables or follow-up needed for the question. Review source coverage, linkage and changes in coding before finalizing eligibility and analysis. A recorded absence may mean no capture rather than no disease.
3. Groups and follow-up
Selection · Align eligibility exposure and observation
Define eligible alternatives at a common decision point. Consider new-user and active-comparator restrictions when clinically appropriate, and avoid immortal time created by classifying people using future treatment or survival.
4. Outcomes and estimands
Design · Define outcomes and comparison targets
An estimand specifies the quantity the study aims to estimate. Define the population, strategies, outcome and horizon, including treatment changes and competing events. Safety outcomes are included when relevant, not because the study is assumed to be a trial.
5. Confounding adjustment plan
Statistics · Specify covariates adjustment and diagnostics
Select adjustment variables using their causal and temporal roles. Avoid indiscriminate adjustment for post-exposure mediators or colliders. Examine overlap and measured balance; neither establishes absence of unmeasured confounding.
6. Estimation and sensitivity plan
Statistics · Specify effects and bias analyses
Use an estimator and uncertainty method compatible with the target contrast, adjustment and follow-up. Report absolute measures where useful. Hazard ratios, rate ratios, risk ratios and odds ratios are not interchangeable.
Equations and statistical interpretation
These explanations describe notation and assumptions. Flow does not calculate these values.
Risk contrast at a fixed horizon
R₁(t) and R₀(t) are risks by the same horizon t for the two specified groups or strategies. State the target population and whether risks are crude, standardized or weighted.
A causal interpretation requires justified identification assumptions. Hazard ratios and rate ratios are different quantities and cannot simply be substituted for a risk ratio; handle censoring and competing events consistently.
Published example
Examine new-user eligibility, propensity matching, claims-based outcomes and follow-up. Matching measured covariates does not establish that unmeasured causes of treatment choice are balanced.
This publication illustrates a study design, not an endorsement of every method or a reproduction of its results. Follow its protocol, supplement and update history when studying the example.
Hypothetical scenario
A health system wants to compare one-year hospitalization risk among eligible patients starting two alternative diabetes medicines.
- Comparative study overview: Define initiation of medicine A versus B and the target patient population.
- Sources and data preparation: Confirm treatment, covariate, hospitalization and mortality capture in linked records.
- Groups and follow-up: Apply a common eligibility assessment and start follow-up at treatment initiation in both groups.
- Outcomes and estimands: Specify hospitalization definitions, follow-up horizon and treatment-change strategies.
- Confounding adjustment plan: Select baseline confounders, inspect overlap and assess balance after the chosen adjustment.
- Estimation and sensitivity plan: Estimate the prespecified contrast and uncertainty, with sensitivity analyses for residual bias and missing capture.
Deliver a populated protocol, decision log, data specification, and planned outputs. Enter numerical findings only after analysis and verification. Keep decisions made before seeing results distinguishable from later amendments.
Sources
- STROBE checklist for cohort studies
- RECORD checklist for studies using routinely collected health data
- Hernán et al. (2016): specifying a target trial and preventing immortal-time bias
- HMA–EMA Catalogue of RWD studies: registration and protocol guidance
- Hernán and Robins: Causal Inference—What If
Source check: 2026-09-17. Record the version of each reporting or appraisal tool used in your actual protocol.
Compare all Observational and RWE studies project types · Open editor in a new tab