Randomization and Comparative Designs

1
Randomization and Comparative Designs

• Oncology Journal Club
• April 5, 2002

2
Comparative Designs

• Compare need more than one group
• Different types
• historical control
• two treatment groups
• treatment and placebo groups
• Phase III

3
Was this study comparative?

• What are the groups that are being compared?
• Treatment 1 vs. treatment 2?
• Was it randomized?
• What was were they randomized to?
• Did they show a difference in the two groups
  under consideration?
• Did they show that the groups being compared were
  comparable with regard to pertinent factors?

4
Randomization

• Why? Whats the big deal?
• Reduces potential for bias
• Ensures that groups being compared are likely
  to be similar to each other.
• Example of violation of randomization bias
• selection bias the physician decides which
  patients are assigned to which treatment
• i.e. physician decides which patients get high
  versus low radiotherapy!

5
Randomization

• What if physicians tend to give sicker patients
  less radiotherapy?
• Now, there is a correlation between being sick
  and treatment.
• Is it so strange to imagine that the sicker
  patients would tend to have shorter survival?
• Now that they have confounded sick status with
  treatment, they CANNOT conclude anything about
  treatment.

6
Randomization

• Idea of Confounders many variables may be
  associated with outcome. By randomly assigning
  individuals to treatment groups, we decrease
  likelihood of making an error due to a confouding
  variable

7
Randomization

• Randomization to low versus high radiotherapy
  WOULD have made illness and treatment
  independent.
• How could this have been helped?
• Inclusion/exclusion criteria so that only kids
  who were healthy enough could receive full dose
• Stratify by stage ensure that comparable
  numbers of sick and less sick kids are in each
  arm.

8
Final Comments on Randomization

• It does not guarantee that groups are the same,
  but the principle is that for large numbers of
  patients, the groups will even out.
• For small studies, might be a good idea to
  stratify to really ensure balance.
• Randomization isnt always truly random
• blocking
• stratification

9
Final Comments on Comparative Trials

• Selection bias not just physician choice
• center (e.g. multi-center study)
• patient (think about ITT vs. actual received)
• Blinding/Masking
• when possible, it is generally a good idea for
  patient (blinded) or patient and physician
  (double-blinded) to not know which group patient
  is assigned to
• avoids sub-concious effects
• avoids cross-over