13 Research methods, statistics, critical review and evidence-based practice13.3 Experimental studies
13.3.6 Other experimental study designs
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.
Match the design to the intervention
Design | How it works | Main limitation |
|---|---|---|
Crossover | Each participant receives the treatments in turn, in different periods | Carry-over from one period to the next, and the illness may change over time |
Factorial | Participants receive combinations of two or more interventions | Interactions can complicate the main effects |
Cluster randomised | Groups such as wards or practices are allocated, not individuals | Outcomes within a group are correlated |
Stepped-wedge cluster trial | Groups switch from control to intervention at randomised times | Time trends can look like intervention effects |
Adaptive trial | Rules set in advance allow design changes as data accumulate | Changes need appropriate inference and control of error |
Quasi-experimental | An intervention is introduced without random allocation | Confounding and secular trends (background change over time) |
Parallel-group trials keep each participant in one allocated group. This avoids crossover carry-over, but the comparison is between different people.
Factorial interaction means the effect of one treatment depends on whether the other is given. A single averaged main effect may then mislead.
Contamination happens when controls receive parts of the intervention. Allocating whole services reduces it, but does not remove every other bias.
Planned adaptation
Adaptive rules are set in advance. They may change the sample size, the allocation probabilities or which arms continue.
Unplanned changes made after seeing favourable results can cause selection and multiplicity problems (many looks at the data raise the chance of a false positive).
Independent monitoring keeps interim results away from the study team and protects the intended comparison.
Non-randomised interventions
Interrupted time series compares repeated measurements before and after an intervention. It models the existing trend and any change in level (a sudden step) or slope (the rate of change).
A simple before-and-after comparison cannot separate the intervention from natural recovery, regression to the mean or other service changes happening at the same time.
A concurrent control series helps separate time trends that both groups share. You still need to check how comparable the groups are.
Analysis must match how people were allocated and measured. Where relevant, allow for clusters, repeated observations, time trends and planned interim analyses.