- Published
- Updated
Stratified randomisation can account for important baseline characteristics. Combining too many factors and categories, however, can produce sparse cells and an allocation process that is difficult to operate reliably at participating sites.
Define the question and assumptions
Assess each proposed factor for prognostic importance, measurement reliability and availability at the moment of randomisation. Make the number of possible combinations visible. Agree with the statistician how factors used in allocation will also be handled in analysis, rather than leaving that decision until results are known.
Hypothetical study example
In a hypothetical study, country, site, age group and disease category are all proposed as strata. Expected enrolment leaves many combinations empty. The team prioritises the most defensible factors and separately explains how other characteristics will be addressed analytically or descriptively.
Design review checks
- Confirm each factor can be measured consistently before treatment allocation occurs.
- Inspect expected enrolment across combinations rather than examining factors only separately.
- Define correction and analysis handling for an incorrectly entered stratification category.
Record the decision and its limits
The decision should not promise perfect balance on every characteristic; chance imbalance remains possible. Acceptance testing of the allocation system includes boundary categories and entry mistakes. Reducing the number of strata does not justify omitting clinically important heterogeneity from the interpretation of results.
Frequently asked questions
- Should every prognostic variable define a stratum?
- No. Study size, operational practicality and the analysis strategy must be considered together.
- Does an incorrect stratum require reversing allocation?
- Not automatically. Document the event and apply the prospectively defined trial procedures.