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.

 

 

 

Return to Main Outline document