13 Research methods, statistics, critical review and evidence-based practice13.2 Epidemiological studies
13.2.3.3 Self-controlled case series
2 min read1 figure
Written by Resident Doctor Committee · Published
New in the 2027 syllabus
The College added this topic for exams from January 2027.
What it is
A self-controlled case series (SCCS) includes only people who have had the event. It compares how often the event happens in set risk periods with how often it happens in control periods in the same person. Each person is their own control.
Exposure changes over the observation time, for example the period after starting a medicine. Investigators do not randomise it.
Fixed characteristics are controlled by the within-person comparison. This covers genetic factors and anything else that stays the same during observation.
Time-varying factors such as age, season and changing illness severity can still confound the association, so they need proper modelling.
Setting up the timeline
Element | What it means |
|---|---|
Observation period | Time during which events and exposure can be recorded |
Risk period | Time after exposure, set in advance, when an effect is thought to occur |
Control period | The rest of the eligible time, used for comparison |
Pre-exposure period | A separate interval when an event may delay or prompt exposure |
Base risk windows on a plausible mechanism and the protocol. Do not choose them because they give the largest association.
Conditional Poisson analysis estimates a relative incidence (the event rate in risk periods compared with control periods). It allows for each person's own fixed baseline event rate.
Assumptions and limitations
Exposure must not depend on the event. In the standard model, an event should not decide later exposure. Stopping treatment after an event breaks this assumption.
The event must not end observation. Under the standard model, observation time should not stop because of the event. Event-related death needs a suitable extension of the model.
Recurrent events must meet the model's assumptions about how events depend on each other. Closely linked recurrences may need a modified analysis.
Dates must be accurate. Wrong timing of exposure or events can move events between risk and control periods.
Relative incidence is not absolute risk. The design also does not compare exposed cases with a separate group of healthy controls.
Worked example of the rate arithmetic
These made-up aggregate figures show the arithmetic only. A fitted SCCS estimate also conditions on each person and adjusts for time effects.
Period | Events | Person-days |
|---|---|---|
Risk | 12 | 600 |
Control | 18 | 1,800 |
Risk-period incidence = 12 / 600 = 0.020 events per person-day.
Control-period incidence = 18 / 1,800 = 0.010 events per person-day.
The crude rate ratio is 0.020 / 0.010 = 2. On its own, this does not show that the treatment causes the event.