Retrospective cohort study
Why this level?
Specify the research question: intervention, diagnostic accuracy, prognosis, aetiology, or screening intervention. Other question types are outside this hierarchy. NHMRC 2009 sourceFor an intervention question with concurrent controls, retrospective and prospective cohorts share a level. For aetiology or prognosis questions, timing and inception criteria can change the classification; the question must be specified first.
Where the NHMRC 2009 hierarchy names this design, it places it here:
| Research question | Level | When |
|---|---|---|
| Intervention | III-2 | An observational comparison uses concurrent controls. Statistical adjustment does not change this design level. |
| Intervention | III-3 | The comparison uses historical controls or separate single-arm studies, without concurrent controls. |
| Prognosis | III-3 | A retrospective cohort examines prognosis. |
| Prognosis | IV | A case series or cohort includes people at different stages of disease. |
| Aetiology | III-2 | A retrospective cohort examines an exposure that cannot be experimentally allocated. |
What question does this help answer?
“Among people who differed in a prior exposure, did the later risk or rate of an outcome differ?”
When to use it
This design is useful when reliable longitudinal data already exist, exposure occurs before outcome, follow-up can be reconstructed, and a prospective study would be slower or more expensive.
When not to use it
Avoid relying on a retrospective cohort when exposure timing is unclear, the source population cannot be defined, follow-up is too incomplete, important confounders were never measured, or outcome ascertainment differs substantially by exposure group.
How it works in plain language
Researchers define an eligible cohort at a historical baseline, measure exposure status using records available at or before baseline, then follow the record forward to determine outcomes. Although the researcher is looking backward from the present, the analytic direction still runs from exposure to later outcome.
What the timeline looks like
What a cohort's Table 1 looks like
| Characteristic | Group 1 (n = 250) | Group 2 (n = 250) | P value |
|---|---|---|---|
| Age, mean (SD), y | 64.2 (12.1) | 63.0 (13.1) | .07 |
| Female, No. (%) | 142 (56.8) | 138 (55.2) | .74 |
| Comorbidity index, mean (SD) | 2.6 (1.7) | 2.5 (1.9) | .54 |
| Insurance type, No. (%) | 62 (27.2) | 56 (25.2) | .64 |
| Baseline severity score, mean (SD) | 5.4 (2.2) | 5.2 (2.4) | .75 |
Why Mallard may recommend it
Mallard may recommend this design for EHR, claims, registry, or administrative-data questions in which a historical index date and longitudinal outcomes are available.
Key assumptions and design requirements
- Exposure is defined before the outcome risk period.
- Eligibility and time zero are aligned across comparison groups.
- Outcome ascertainment is comparable.
- Confounding is addressed using subject-matter knowledge and an appropriate causal strategy.
- Censoring and loss of observable follow-up are handled appropriately.
What to watch out for
Immortal time bias, misclassification, confounding by indication, selection into the dataset, changes in coding over time, and adjustment for variables measured after exposure can all distort results.
Clinical example
Among adults with obesity who initiated GLP-1 therapy versus a comparator therapy in 2022, compare subsequent cardiovascular events over two years using EHR data.
- STROBE statement for observational studies (opens in a new tab) · EQUATOR Network
- Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies (opens in a new tab) · BMJ · doi 10.1136/bmj.39335.541782.AD
- Observational studies: cohort and case-control studies (opens in a new tab) · Plast Reconstr Surg · doi 10.1097/PRS.0b013e3181f44abc
- Overview of clinical study designs (opens in a new tab) · Clin Exp Emerg Med · doi 10.15441/ceem.23.036