| Design |
Implement strategies to minimise the risk of stratification errors, such as clearly defining stratification variables, avoiding stratification variables that are likely to be subject to high levels of misclassification, training staff to carefully check the stratification variables before randomising participants, and designing randomisation systems that are easy to use. Pre‐specify how stratification errors will be addressed in the primary analysis and any planned secondary or subgroup analyses in the statistical analysis plan.
If the discovery of errors is unlikely to be related to treatment group (eg, in a blinded trial), then (i) adjust for the updated strata
a
in the primary analysis, (ii) adjust for the randomisation strata
a
in a sensitivity analysis, and (iii) use the updated strata
a
to perform any subgroup analyses. If the discovery of errors could plausibly be related to treatment group (eg, in an unblinded trial involving more contact with intervention participants), then (i) adjust for the randomisation strata
a
in the primary analysis, (ii) adjust for the updated strata
a
in a sensitivity analysis, and (iii) use the randomisation strata
a
to perform any subgroup analyses.
|
| Conduct |
-
3
Implement systematic strategies to identify stratification errors across all participants, such as cross‐checking stratification variables against alternate data sources (where available).
-
4
Thoroughly document any stratification errors that are identified, including how they occurred, to assist with staff training and inform the design of future trials.
-
5
Record the updated strata
a
in a new field in the trial database and leave the randomisation strata
a
unchanged as a record of how the randomisation was performed.
|
| Analysis |
-
6
Examine the number of stratification errors that were identified overall and broken down by treatment group, stratum and their combination.
-
7
Address stratification errors in the planned analyses using the approach(es) pre‐specified in the statistical analysis plan.
-
8
Consider whether any additional, unplanned analyses should be performed, based on the number and pattern of stratification errors that were identified in the trial (eg, a sensitivity analysis excluding participants affected by stratification errors).
|
| Reporting |
-
9
Report the number of stratification errors that were identified by treatment group and stratum, potentially in a Supplementary Appendix S1, or state that no such errors were identified.
-
10
Indicate whether the randomisation strata
a
or the updated strata
a
was used in each analysis involving the stratification variables.
|
