Bias:
key points
i)
What is bias ?
This
means that the difference in treatment group means will not have expectation
equal to the genuine treatment effect.
In this sense the idea of bias is the same as that encountered widely in
mathematical statistics. However, in the
context of RCTs a rather wider view of bias is taken in the sense that the
problem is usually due to some aspect of the data collection yielding data that
cannot provide an unbiased estimator. In
general, bias arising this way cannot be calculated and no simple correction
can be computed.
ii)
Selection bias
If the treatment the next patient is to receive is known before this allocation
is made, then this can influence the patient and/or recruiting doctor’s
decision to admit that patient to the trial and this can lead to unbalanced
groups.
iii)
Allocation bias
Patients will differ in their characteristics, including some (often know as prognostic factors) which affect how
they might respond to treatment.
Randomization should lead to treatment groups that are balanced with
respect to all prognostic factors.
However, especially in small trials, this may not occur, and a bias due
to uneven allocation can arise
iv)
Assessment bias
This can arise if the person assessing the outcome variable(s), on which the
results of the trial will be based, knows which treatment the patient has
received. Problems can also arise if the
patient knows which treatment they were allocated.
v)
Publication bias and stopping rules
These will not be considered further in the course. The former can arise if journals decide
whether or not to publish reports based on the results they contain. The latter can arise if the way the trial is
stopped could be influenced by the results obtaining at that time.