Hi, I'm Olga Kuznetsova.
I'm a Senior Principal Scientist at Merck & Co.
I will talk about selection bias in studies with unequal allocation.
I will talk about the origins of selection bias in
clinical trials and derivation of selection bias in trials with unequal allocation.
We will discuss directional guessing strategy and
directional guessing strategy with biasing threshold.
Finally, I will show how
Brick Tunnel Randomization introduced by Kuznetsova and Tymofyeyev,
reduces selection bias when the investigator thinks
that Permuted Block Randomization was used in the trial and applies a biasing threshold.
What are the origins of selection bias in clinical trials?
In an open-label single-center trial,
the investigator knows the sequence of treatment assignments for all allocated patients.
Thus, the investigator might try to guess
the next treatment assignment and allocate
a patient best suited for the guessed treatment.
This introduces selection bias in
the study results by making the treatment groups incomparable.
Selection bias is well studied for equal allocation,
but is less studied for unequal allocation.
We will discuss selection bias in trials with unequal allocation.
Blackwell and Hodges offer the following model for selection bias.
Consider a study with 1:1 randomization to treatments A and B.
The study compares a normal response variable, y,
between the two groups.
Let us assume that there is no treatment effect.
There are two types of patients,
healthier patients with expected response Mu plus Delta and
sicker patients with expected response Mu minus Delta.
Suppose the investigator wants to bias the results in favor of treatment A.
To do that, the investigator allocates a healthier patient,
when their guess for the next treatment is A,
and a sicker patient when their guess is B.
As a result, group A has healthier patients and a better outcome.