Choose the cohort type
- Prospective cohort: define participants/exposure now and follow outcomes forward.
- Retrospective cohort: reconstruct exposure and follow-up from existing records where both exposure and outcomes may already have occurred.
- Ambidirectional cohort: use historical data and continue prospective follow-up.
Approval and governance
Usually needs formal review / authorisation
- Prospective recruitment/contact and collection of research measurements typically require ethics/governance review and consent unless an authorised body approves another route.
- Retrospective access to identifiable clinical records generally requires ethics/governance and data-access authorisation, with any consent waiver granted only by the competent body.
- Record linkage, genomic data, biospecimens or transfer outside approved systems may require additional approvals.
May follow a lighter or different route
- Fully anonymised public datasets may follow a different route if no re-identification is possible and terms permit the work.
- Some routine-data analyses may qualify for waiver of consent, but that is an ethics/governance decision, not an investigator decision.
Do not do this
- Do not start chart abstraction “just to see what is there” before access is authorised.
- Do not redefine exposure or outcome after seeing which definition gives a stronger result.
- Do not ignore patients lost to follow-up.
Mediclinic Middle East publicly states that research projects carried out at MCME are to receive approval from its internal Research and Ethics Committee and applicable local regulatory authorities before initiation. The exact route varies by project, facility and emirate. Dubai projects may involve DSREC depending on applicability. This hub must therefore route users to the Research Office and current local forms rather than declaring a project “ethics exempt.” Institution-specific forms and contacts will be inserted after verification.
Step-by-step workflow
Define time zero
Specify the point at which each participant becomes at risk and exposure status is assigned. Time zero must be aligned across comparison groups to avoid immortal-time and selection biases.
Define exposure and comparison groups
Use objective, reproducible rules. If exposure can change over time, decide whether it is baseline, time-updated or cumulative and plan analysis accordingly.
Define outcomes and ascertainment
Use validated definitions where possible. Specify whether outcomes come from records, laboratory values, imaging, registries, interviews or adjudication. State the follow-up window and censoring rules.
Set eligibility criteria independent of future outcome
Eligibility should be knowable at or before cohort entry. Excluding participants because of later missing data can introduce bias.
Identify confounders and causal structure
List plausible common causes of exposure and outcome. Distinguish confounders from mediators and colliders. Prespecify the adjustment set using clinical knowledge and, where appropriate, a causal diagram.
Calculate sample size / events needed
Base the calculation on the primary outcome, expected event frequency, effect size, follow-up and planned model. For regression models, ensure the number of outcome events is adequate for the model complexity.
Plan follow-up and retention
For prospective cohorts, define contact intervals, permissible windows, retention methods and procedures for withdrawal. For retrospective cohorts, define data availability and end-of-follow-up consistently.
Obtain approvals and data permissions
Submit protocol, variable list, recruitment/consent materials if prospective, record-query logic if retrospective, linkage plan, analysis plan and security controls.
Build the dataset with a participant flow log
Track assessed, eligible, included, excluded, lost, withdrawn and analysed participants. Store direct identifiers separately from the analytic dataset.
Check baseline comparability without “testing” it mechanically
Describe important baseline characteristics by exposure group. Differences should be interpreted substantively; baseline p-values are usually less informative than the magnitude of imbalance.
Estimate incidence and effects
Depending on the question, report risk/cumulative incidence, incidence rates, risk differences/ratios, rate ratios or time-to-event estimates such as hazard ratios, each with confidence intervals.
Handle time correctly
Use person-time when follow-up differs. For survival analysis, define censoring, assess proportional-hazards assumptions if using Cox models, and display Kaplan–Meier curves when appropriate.
Address confounding and missing data
Use the prespecified model/weighting/stratification approach. Report missingness and perform sensitivity analyses where assumptions could affect the result.
Evaluate loss to follow-up and competing events
Compare follow-up completeness and assess whether loss may be related to exposure or outcome. For outcomes where competing risks matter, choose methods that match the estimand.
Write using STROBE
Report cohort entry, exposure ascertainment, outcomes, follow-up time, losses, confounding strategy, unadjusted and adjusted estimates, missing data, sensitivity analyses, limitations and generalisability.
Final checklist
- Time zero clearly defined.
- Exposure and outcome definitions prespecified.
- Follow-up and censoring rules explicit.
- Confounders chosen using subject knowledge.
- Sample size/events justified.
- Approval/data access documented.
- Loss to follow-up quantified.
- Appropriate effect measure and confidence interval reported.
- STROBE checklist completed.