CONTRACT RESEARCH
and then weighting the outcome in favor of the treatment assignment that will result in the lowest imbalance score. There are many methods for calculating such a score and we will illustrate two common concepts. Consider the example in Table 1, where 15 patients have already been randomized and now the 16th arrives at the site – a male, non-smoker in the “low” age group (sub-group MNL for Male/ Non-smoker/Low age group).
The first method that could be used is to sum the number that would now be in the MNL group for treatment A and the same sum for treatment B. In this example, assigning the patient to A would result in a score of 6+6+4=16. Assignment to B would score 4+5+5=14. Since the score for assignment to treatment B is lower, we would weigh the coin in favor of that treatment group. Another method is to sum the differences within each stratification factor. In this example, if the patient is randomized to A, the differences total as follows: ABS(6- 2)+ABS(2-6)+ABS(4-4)=8. The differences for B would be: ABS(4-5)+ABS(4-5)+ABS(4-4)=3. Again, assignment to treatment B results in the lower imbalance score and we would weigh the coin in favor of that treatment group.
The imbalance calculations presented so far have been rather straightforward. Suppose, though, that maintaining balance in gender
is far more important than smoking status or age group. Then we could apply a weighting or multiplier to that factor in our calculation to help ensure that we continue to have approximately equal numbers of males and females in each treatment group.
Simulation
As randomization methods grow more com- plex, how can we be certain that they will maintain balance and lead to the desired ratios in treatment groups? Biostatisticians typically run a number of simulations through the study design and then analyze the behavior of the randomization schema. These simulations are based on the expected patient characteristics while also introducing some randomness to account for real-world unknowns. Using our previous example from Table 1, a simulation
run may begin randomly creating patients with a 50/50 probability that the next patient will be male or female, 30/70 probability for smoker/ non-smoker and 40/60 for low-age/high-age group. Further, some “what if” scenarios can be modeled to see what would happen if more or less randomness is introduced into the al- gorithms. Simulations give the statistical team insights into the likely outcomes and also give clinical teams strong evidence that the algo- rithms will lead to the desired results in terms of balance and treatment ratios.
Regardless of the randomization method used, it is vital to have a provider who has the tools to monitor how the study is performing once it goes live. While simulation results provide a solid reassurance that the randomization algorithm is set up according to the needs of the study, even the best simulation techniques cannot account for every real-world factor.
Table 2. Permuted Blocks
Vials Distribute
Dose 10 mg 20 mg 30 mg 40 mg 50 mg Placebo 2 x 20 mg
1x10 mg 1 Placebo
Alternative n/a
1 x 20 mg 1 Placebo
2 x 10 mg
1 x 30 mg 1 Placebo
1 x 20 mg 1 x 10 mg
Pharmaceutical Outsourcing | 18 | November/December 2016
1 x 30 mg 1 x 10 mg
1 x 30 mg 1 x 20 mg
n/a 2 x Placebo n/a
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54