Already a DIA Member? Sign in. Not a member? Join.

Sign in

Forgot User ID? or Forgot Password?

Not a Member?

Create Account and Join

Menu Back to Poster-Presentations-Details

M-16: Impact of Different Randomization Techniques on The Statistical Efficiency in Clinical Trials





Poster Presenter

      Jackline Jepkorir Kemboi

      • Student
      • African Institute For Mathematical Science (aims)-Rwanda
        Kenya

Objectives

The objective of this study is to explore the allocation imbalance, power loss and predictability in simple, block and stratified randomization for two-treatment superiority trials and explore reasons why some one of the mentioned techniques are more efficient than others against the other.

Method

A Monte-Carlo-Simulation was used to generate clinical trials with the outcome and 4 predictive binary variables. For each randomization technique, the absolute allocation imbalance (AAI) and absolute stratified allocation imbalance (ASAI) was explored and their impact on the statistical power.

Results

We have defined the following variables to quantify the allocation imbalance: AAI is the is the absolute difference between the number of patients in the two treatment groups while the ASAI is the absolute sum of the difference between the treatment group in each subgroup. We have demonstrated that these two metrics are important because they have a significant impact on the statistical power. Small AAI results in insignificant loss of power. However, more pronounced AAI (i.e. >60%), the power may be reduced by up to 24%-points. For example, a trial which was originally powered with 90%, would only have 66% when the AAI is large, which would be largely underpowered. To compensate for such a power loss, the sample size would need to be increased. When simple randomization is used, the AAI can be quite large with values of up to 90%, in particular when the required sample size is low (less than 100). Therefore, simple randomization should generally not be used in clinical trials. On the other hand, block randomization (with block sizes of up to 8) ensures that that the size of the groups are almost equal, and the AAI is very low.To account for prognostic variables (aiming for balancing these variables across the treatments), stratified block randomization could be used. While the ASAI would be can be ensured to be low, the AAI could be larger because typically numerous blocks will not be used completely, which could lead to potential imbalance and hence power loss. The risk for larger AAI increases with the number of prognostic factors.

Conclusion

The simple randomization technique, technique is easy to use. However, there is large probability of getting unequal samples sizes which in return affect the statistical power of the trial. Although, treatment imbalances affect power, the effects are unremarkable unless the inequalities are substantial. Larger sample sizes (>200) reduce the risk for large relative allocation imbalance between the treatment groups and lead to more statistical power. In block randomization, the absolute allocation imbalance is almost zero regardless of the sample size, (except when there is an incomplete block at the end of the trial. However, block randomization is susceptibility to selection bias and accidental bias in the estimate of treatment effect. The probabilities of both biases are a direct function of the predictability of the randomization sequences. Therefore, it is essential that possible bias in the study is eliminated, the researcher should increase the block sizes or use random blocks in the assignment of treatment and secondly, double blinding should be used to ensure that the researcher nor the patients know the kind of treatment they are administering and receiving respectively. Unlike block randomization, stratified randomization ensures that treatments groups are comparable with respect to important prognostic factors. The results showed that the total ASAI doesn't depend on the frequency of patients in each stratum. Moreover, the researcher should ensure that only important factors considered in stratification as too many strata result to increase in ASAI. Moreover, further work should be done to quantify the effect ASAI has on power. As the 4 variables are predictive for the Overall outcome, an imbalance within the randomized strata might lead to more severe power loss.

Be informed and stay engaged.

Don't miss an opportunity - join our mailing list to stay up to date on DIA insights and events.