Is minimization true randomization

Minimization as applied above is NOT random, and DOES depend on the patient characteristics and so appears to contradict the basic definition of randomization. However, it does meet the requirement to avoid systematic bias in allocating treatments, and where it is impossible for the person considering randomizing a patient to have full knowledge of the characteristics and treatment allocations of all previous patients (and this would usually be the case in multi-centre trials) minimization fulfils the concealment aspect of randomization too in that the next treatment allocation is not predictable.

Bearing in mind the earlier statement that randomization must not only be correct, but must be seen to be correct, this form of minimization may not be an appropriate method to use in a single centre study. However, where predictability is a concern, it is possible to include a random element into minimization. To do this, instead of always allocating the treatment which minimizes the imbalance, that treatment is allocated with a certain probability p(p < 1), for example p = 0.75.

To do this in practice, one can prepare a random number list on which 0 to 5 ='allocate the treatment which minimizes imbalance' and 6 and 7 = 'allocate the other treatment' (ignore 8 and 9 and take the next digit <7). This ensures, on average, that the allocation which minimizes the imbalance is chosen 75 per cent of the time. Having worked through the 'summing' procedure as described above, consult the list to see if the next patient should be allocated the treatment which minimizes the imbalance or not, remembering to strike through each number on the list as it is used.

Minimization is widely used in oncology, perhaps because of the multi-centric nature of many cancer trials - stratifying by centre and other factors does, as described above,

Stage = I/II

Stage

= III/IV

Treatment 1

Treatment 2

1. 1002

1. 1001

2. 1005

2. 1004

3. 1006

3.

4.

4.

5.

5.

6.

6.

7.

7.

8.

8.

9.

9.

10.

Age >50

Age <50

Treatment 1

Treatment 2

1. 1002

1. 1004

2. 1006

2.

3.

3.

4.

4.

5.

5.

6.

6.

7.

7.

8.

8.

9.

9.

10.

Hospital = Y

Hospital = X

Treatment

1

Treatment 2

1. 1005

1. 1001

2. 1006

2.

3.

3.

4.

4.

5.

5.

6.

6.

7.

7.

8.

8.

9.

9.

Fig. 4.8 Minimization cards for manual allocation become impractical and may not provide the overall balance required. However, it is not accepted universally, and those conducting licensing trials should ensure that the relevant licensing body accepts minimization as a valid method of randomization. Some organizations (of which the Food and Drug Administration (FDA) in the US and the Medicines Control Agency in the UK are perhaps the ones of most note), will certainly require that every allocation method, including minimization, has some form of random element.

0 0

Post a comment