Observational and RWE studies

Start with the question, the available measurements and when events occur. The studies below progress from population snapshots and conventional designs to more specialized sampling and causal comparisons. Complexity also depends on the data and question.

Which study should I use?

Study type	When to use it	How it differs
Cross-sectional study	Use a cross-sectional study to estimate prevalence or describe characteristics and associations at a specified time or over a short assessment window. Specify the population that the sample is intended to represent.	A cohort follows outcomes over time; a cross-sectional study describes a snapshot. Repeated cross-sectional surveys sample populations at successive times and need not follow the same people. Simultaneous measurement usually cannot establish whether an exposure preceded an outcome.
Retrospective descriptive cohort	Use a retrospective descriptive cohort to describe care, events or clinical trajectories over a historical follow-up period. The question is about the experience of an eligible population, rather than the causal effect of choosing one treatment over another.	Unlike a cross-sectional snapshot, this design requires a time origin and follow-up. Unlike a comparative cohort, it does not require a causal treatment contrast. Retrospective describes when the records are examined, not whether the outcome preceded cohort entry.
Prospective observational cohort	Use a prospective observational cohort when participants can be enrolled and measurements collected forward under a protocol while exposures and care are observed. Plan measurement quality and follow-up before the outcomes of interest occur.	Compared with a retrospective cohort, prospective collection can tailor measurements but requires recruitment and retention. It is not a randomized trial: the study does not randomly allocate the exposure. Causal comparisons still require explicit assumptions and confounding control.
Case-control study	Use a case-control study when selecting people by outcome status makes investigation efficient, especially for an uncommon outcome or costly exposure ascertainment. Controls must represent the exposure experience of the population that produced the cases.	A cohort starts with eligibility and follows events; a case-control study samples cases and controls then reconstructs relevant exposures. A nested case-control study additionally draws controls from a defined cohort’s event-time risk sets. Control sampling determines the interpretation of the odds ratio.
Nested case-control study	Use a nested case-control study when a defined cohort and event dates are available but measuring exposures or biomarkers for everyone is costly. For the incidence-density starter, sample controls from those still at risk when each incident case occurs.	Unlike a general case-control study, the source cohort and risk-set membership are explicit. Unlike a case-cohort design, controls are selected at event times rather than as a baseline subcohort. A person can be a control more than once and later become a case.
Comparative cohort using existing data	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.	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.
Interrupted time-series study	Use interrupted time series when an intervention has a defensible implementation time and repeated outcome measurements before and after it. Specify an expected immediate, gradual or delayed response and consider a suitable comparison series.	A single before–after comparison cannot estimate the prior trend. Interrupted time series uses that trajectory to construct a counterfactual. Difference-in-differences centers on changes relative to a comparison group; controlled time-series designs can combine both ideas.
Difference-in-differences study	Use difference-in-differences when outcomes are observed before and after an intervention in exposed and comparison groups, and a credible argument can be made that their untreated trends would have changed similarly without the intervention (parallel untreated trends).	Unlike a simple before–after study, it subtracts change in a comparison group. Unlike a basic interrupted time series, its identifying argument is about comparative untreated trends. Staggered adoption requires methods suited to treatment timing and heterogeneous effects.
Self-controlled case series	Use a self-controlled case series for a transient exposure and an acute event when reliable exposure, event and observation dates are available among people who experience the event. Compare defined risk and baseline periods within each person.	A cohort or case-control comparison uses differences between people. SCCS controls factors constant within a person during observation, but time-varying factors still need attention. It differs from case-crossover sampling and from a descriptive clinical case series.
Target trial emulation	Use target trial emulation to structure a causal comparison of treatment strategies with observational data. First specify the hypothetical trial, then map each element to what the data can actually support.	It is a design framework for observational causal studies, often implemented as a comparative cohort. Explicit eligibility, strategy assignment, time zero and follow-up prevent avoidable design errors, but do not create randomization or eliminate unmeasured confounding.

Data source and study design

Real-world data are routinely collected information about health or care, such as health records, claims and registries. Real-world evidence is evidence produced by analyzing such data. These describe the data and resulting evidence, not a single study design. An observational study may also collect measurements specifically for research.

A disease registry stores information about patients or events. A study registry records a study’s protocol and administrative details. Neither term means that treatment was randomized.

Decisions shared across these studies

State whether the question is descriptive, associational or causal. Specify the target population, outcome and quantity to estimate.
Define calendar dates, time zero, exposure windows and follow-up. Keep people, episodes, events and person-time as distinct counting units.
Assess source coverage, measurement validity, selection and missingness before choosing an analysis. For causal comparisons, state the assumptions that connect the observed data to the intended effect.
Keep a dated protocol and record amendments. Where registration is appropriate, identify the registry, study identifier and timing. The HMA–EMA Catalogue of RWD studies is one option for relevant studies; it is a study catalogue, not a clinical trial allocation system. Confirm eligibility and applicable requirements for the particular study.

Terms used in these guides

Estimand: The precise quantity the study aims to estimate, defined by its population, outcome, timing and any comparison or strategy.
Time zero: The start of follow-up. Eligibility and exposure or strategy assignment must use timing compatible with this origin.
Confounding: Distortion of an exposure–outcome comparison by factors that help explain both exposure and outcome. Adjustment requires a justified choice of variables and their timing.
Censoring: Follow-up ends before the event or horizon of interest is observed. Analysis must address how that end of observation relates to the outcome.
Competing event: An event that prevents the outcome of interest from subsequently occurring, such as death before a first readmission.
Person-time: The sum of eligible time at risk contributed by people. It is a denominator for incidence rates, not a count of participants.

Using the guides

Each guide explains the steps used in its New project starter, shows the tables to complete, defines relevant equations and follows a hypothetical scenario. Published examples show how the design appears in research. Reporting checklists support transparent reporting; they do not replace study-design decisions or establish validity.