We’ve seen confusion in recent IRT (IVR/IWR) RFPs and study protocols in the use of ‘strata’ and ‘cohort’ when describing desired approaches to randomization. The words are often used in opposition – that is, one is often specified when it’s really the other that’s required.
This confusion in terminology can negatively impact the bidding timeframe, requiring back and forth clarification between vendor and sponsor, or worse yet, can wrongly inflate or deflate the final bid tally. Correcting after the fact also takes additional time. Better by far to specify the strata or cohort requirements correctly from the outset!
Here’s a quick rule of thumb for when to use each:
‘Stratification’ is the correct choice when subject assignment to a treatment group is to be governed not by a randomization schedule that simply lumps all subjects into one large pool but rather one that subdivides randomized assignment by one or more variables. Common examples are stratification by age range, gender, a specific aspect of the subject’s medical history, a screening lab value range, and so on. Stratification by clinical Site is also fairly common, and is correctly called for when the project plan is to assign a different block of randomization numbers (essentially a different randomization sub-schedule) to each Site.
Stratification can also require multiple variables combining to form each strata – for example, Strata 1 could be defined as ‘a moderate condition plus a specified previous treatment’ and Strata 2 as ‘a severe condition plus the same specified previous treatment.’
The general goal of stratification from a biostatistical perspective is to enable easy statistical comparisons of similarities/differences in effects within homogeneous groups of key variables that have been pre-identified as potentially relevant to drug efficacy or safety.
Using cohorts in clinical trials, on the other hand, is the approach of choice when the desire is to control and compare alternate actions imposed upon groups of subjects. The most common example of a good use of cohorts is in dose escalation studies where Cohort A is given Dose Level 1 for X number of subjects and then results are analyzed to determine whether to open Cohort B at Dose Level 2, and so on. Commonality or diversity of subject variables is not at issue here. There’s no attempt to make a homogeneous grouping. Subjects are bound together into cohorts merely by the actions imposed upon them (e.g., dose level).
The use of cohorts is appropriate in models where treatment groups are to be rolled out sequentially rather than concurrently. In the case of stratification, subjects in diverse strata are assigned to treatment groups concurrently, rather like dealing out cards from a full deck into different piles. When the pre-defined number of subjects has been enrolled into the first cohort, it is closed to prevent additional subjects from being randomized until after a DSMB review concludes the next cohort can be opened and subject randomizations can resume.
Things get a bit muddier when multiple open label cohorts are rolled out concurrently – i.e., Cohort A, subjects 1-10, gets Treatment 1 (or placebo), while Cohort B, subjects 11-20, gets Treatment 2 (or placebo), and so on. Using cohorts for this type of design is appropriate, allowing the option to modify group limits and discontinue a treatment group or activate a third Treatment scenario allowable if need be.
A Stratified Cohort?
Yes, Virginia, there are times when a stratified cohort is the correct specification!
In the classic sequential cohort case described above, imagine that the protocol requires Cohort A data to be segmented for analysis not only by Dose Level but by age range. Randomization (to placebo or Dose Level 1) could then be stratified within the cohort
Share this Post