Stratified randomization is used to ensure balance of product assignments across 1 or more prognostic factors. A prognostic factor is an aspect of the disease or a characteristic of the subject that may influence product effect. The prognostic factors used to stratify the randomization are specified in the protocol. As a simple example, suppose age group (<50, >=50) and gender (male, female) are considered important prognostic factors. When a subject is deemed eligible for randomization, their individual values of these factors are determined at the site and used as input to the randomization process to determine their product assignment. The situation may occur where the value of a factor used for randomization is later discovered to be in error. For example, suppose a subject was randomized according to the age group of <50 and male. Later, it was discovered that the subject was actually 54 and therefore should have been randomized according to the age group of >=50 and male. If this situation happens too often, the balance in product assignments across these factors is in question, which may then result in the use of sensitivity analyses. Therefore, there is an analysis need to have 2 sets of values to describe the stratification factors. In this document, these 2 sets of values are referred to the “as-randomized” values and the “as-verified” values. As-verified values are derived using source documentation.

At present, there is no standard method for representing the randomization strata factors and values in SDTM-based datasets. Depending on the randomization process, it might be unnecessary to represent variables and values specific to stratification in SDTM-based datasets if the information can be found within an appropriate domain. For example, if age and sex were used as stratification factors, then the Demographics (DM) variables AGE and SEX should appropriately reflect values used for randomization. However, more sophisticated randomizations or more complicated derivations of prognostic factors, such as whether a subject had ever used a particular concomitant medication for a given length of time, may be harder to identify or document in SDTM-based datasets. If using an interactive voice response system (IVRS), the values used for randomization would be captured by the system and would correspond to the values that are represented on the randomization schedule. As-verified values are typically derived by comparing the values used for randomization against the data that is in the SDTM dataset, whether it be a simple match with a single data point such as sex or the reprogramming of more complex factors such as previous products.

The following table provides a set of variables to allow maximum flexibility in representing the description of the prognostic factors. To illustrate the interrelationships of the variables, the examples for every variable in the CDISC Notes column use the combination of 3 stratification factors: age group (“<50” or “ >=50”), prior product status (“Product naïve”, “Product experienced”), and hypertension (“Y” or “N”).

Variable Name

Variable Label

Type

Codelist/ Controlled Terms

Core

CDISC Notes

STRATAR

Strata Used for Randomization

Char


Perm

STRATAR contains the combination of values of the individual stratification factors used for randomization. The exact format should be determined by the applicant.

This variable is intended for studies that use stratified randomization.

For example, ">=50, Product experienced, N"

STRATARN

Strata Used for Randomization (N)

Num


Perm

Numeric representation of STRATAR. For example, STRATARN=3 when STRATAR=">=50, Product experienced, N". There must be a one-to-one relationship between STRATARN and STRATAR within a study.

STRATARN cannot be present unless STRATAR is also present. When STRATAR and STRATARN are present, then on a given record, either both must be populated or both must be null.

STRATwD

Description of Stratification Factor w

Char


Perm

STRATwD is a full text description of the stratification factor "w". This text description will remain constant for all subjects. These descriptive variables are included to quickly and clearly communicate critical study design information as well as to facilitate integration.

For example, STRAT3D="Hypertension"

STRATwR

Strat Factor w Value Used for Rand

Char


Perm

STRATwR is the subject-level value of the "w'th" stratification factor used for randomization.

For example, STRAT3R="N"

STRATwRN

Strat Factor w Value Used for Rand (N)

Num


Perm

Numeric representation of STRATwR. For example, STRAT3RN=0 when STRAT3R="N". There must be a one-to-one relationship between STRATwRN and STRATwR within a study.

STRATwRN cannot be present unless STRATwR is also present. When STRATwR and STRATwRN are present, then on a given record, either both must be populated or both must be null.

STRATAV

Strata from Verification Source

Char


Perm

STRATAV contains the entire string value represents the combination of values of the individual stratification factors that should have been used and represents the "as verified" value. The STRATAV variables are based on the source documentation and are determined after randomization. If the values used for the randomization of a given subject were all correct, then STRATAV will equal STRATAR. Otherwise, one or more components of the text string for STRATAR and STRATAV will be different.

The exact format should be determined by the applicant.

For example, ">=50, Product experienced, Y"

STRATAVN

Strata from Verification Source (N)

Num


Perm

Numeric representation of STRATAV. For example, STRATAVN=4 when STRATVR=">=50, Product experienced, Y". There must be a one-to-one relationship between STRATAVN and STRATAV within a study.

STRATAVN cannot be present unless STRATAV is also present. When STRATAV and STRATAVN are present, then on a given record, either both must be populated or both must be null.

STRATwV

Strat Factor w Value from Verif Source

Char


Perm

STRATwV is the "as verified" subject-level value of the "w'th" stratification factor. If the value based on randomization was correct, then STRATwV will equal STRATwR.

For example, STRAT3V="Y"

STRATwVN

Strat Fact w Val from Verif  Source (N)

Num


Perm

Numeric representation of STRATwV. For example, STRAT3VN=1 when STRAT3V="Y". There must be a one-to-one relationship between STRATwVN and STRATwV within a study.

STRATwVN cannot be present unless STRATwV is also present. When STRATwV and STRATwVN are present, then on a given record, either both must be populated or both must be null.


  • No labels