91828 Chap.10 Choosing a Clinical Trial Design
Choosing a Clinical Trial Design
Trial design should solve a scientific or operational problem while preserving an interpretable estimand. This chapter compares individually randomised parallel, cluster, crossover, factorial and feasibility designs.
It makes the allocation, intervention and measurement units explicit, examines contamination and carryover, and shows why the sample-size and analysis model must respect clustering, periods, interactions and real delivery conditions.
What this chapter covers
- 01
Parallel design
- 02
Cluster randomisation
- 03
Crossover design
- 04
Factorial design
- 05
Contamination and carryover
- 06
Pilot and feasibility studies
Worked example · free
Choose a design for a ward program
- +1Individual randomisation creates substantial contamination through shared staff and environment.
- +1Randomising wards aligns allocation with intervention delivery, supporting a cluster design.
- +1Recruitment procedures should limit selection after cluster assignment.
- +1Sample size and analysis must account for correlation within wards and the number of clusters.
- +1The estimand becomes the effect of implementing the program at ward level.
Key terms
- Parallel Trial
- A parallel trial assigns participants to concurrent groups that remain on their allocated strategy.
- Cluster Trial
- A cluster trial randomises intact groups such as wards, practices or communities.
- Crossover Trial
- A crossover trial exposes each participant to multiple strategies in a planned period sequence.
- Factorial Trial
- A factorial trial evaluates two or more interventions through combinations of allocated factors.
Choosing a Clinical Trial Design FAQ
When is cluster randomisation justified?
Use it when the intervention is delivered to groups, individual allocation would create serious contamination, or policy implementation is itself the target. Ensure enough clusters, protect recruitment and adjust both sample size and analysis for within-cluster correlation.
When is crossover unsuitable?
Avoid it when disease changes rapidly, treatment effects are permanent, carryover cannot be controlled, or participants cannot complete multiple periods. A washout needs biological justification, and period and sequence effects belong in the analysis plan.
What does a pilot trial establish?
A pilot or feasibility study can estimate recruitment, retention, delivery fidelity, measurement performance and procedural acceptability. Unless designed and powered for clinical efficacy, it should not be presented as definitive evidence that the intervention works.
What is contamination in a trial?
Contamination occurs when participants or providers assigned to control receive parts of the intervention, reducing separation between strategies. Identify the pathway, measure fidelity in both arms and consider cluster allocation when delivery is genuinely shared. Clustering introduces its own sample-size and analysis requirements.
How should usual care be described?
Specify the services, contact, materials and co-interventions available to control participants rather than using the label alone. Usual care can vary across sites and time. Measuring its delivery makes the treatment contrast interpretable and supports replication in another setting.
When is a factorial design efficient?
It can test two compatible interventions through their combinations when recruitment, delivery and analysis can support both factors. Examine plausible interaction and adherence complexity. If interventions strongly modify each other, main-effect interpretations and sample-size assumptions may need revision.
Assessment move
Compare trial designs by drawing the allocation, delivery and measurement units for a series of interventions. Begin with an individually randomised parallel trial as the reference and identify whether concurrent groups can receive distinct strategies without spillover. If clinicians, wards or communities deliver a shared program, trace the contamination pathway and consider cluster allocation.
Record the number of independent clusters, not only the number of participants, then add intracluster correlation to the sample-size and analysis plan. Examine recruitment timing because enrolling patients after cluster assignment can create selection bias. For crossover candidates, test four conditions: stable disease, reversible effect, justified washout and repeatable measurement.
Draw both sequences, period timing and carryover pathway. Reject the design when treatment permanently changes the condition or the first period alters later behaviour. For factorial proposals, state the factors and combinations, then consider whether interaction is scientifically plausible and how it changes the interpretation of main effects.
Separate pilot or feasibility objectives from clinical efficacy; recruitment, retention, fidelity and measurement performance can be valuable endpoints without claiming that the intervention works. Write a design-choice paragraph containing the problem being solved, selected structure, principal assumption, main vulnerability and protective procedure or analysis.
Add feasibility: sites, participant flow, adherence and delivery capacity are part of validity when failure would erase the contrast. For every design, name the comparator precisely because placebo, attention control and usual care estimate different effects. Define outcome timing from the biological or behavioural window rather than scheduling convenience.
Translate the method into an AT2 flow diagram showing eligibility, randomisation unit, groups, delivery, outcome and primary comparison. Ask a peer to infer the design without reading the textual abstract. Finally, create a rejection table for attractive but unsuitable designs, explaining why small sample size or visual elegance does not compensate for contamination, carryover, few clusters or an incoherent estimand.
The best choice is the simplest structure that protects the intended comparison under real conditions.
