Assessing allocation bias in stratified clinical trials with multi-component endpoints evaluated using the stratified Wei-Lachin test

Abstract

Background: A common issue in rare disease stratified clinical trials with multi-component endpoints is allocation bias, as they frequently lack blinding. Allocation bias arises when future treatment allocations can be predicted from prior ones, potentially leading to patients with specific characteristics being preferentially assigned to either the treatment or control group. Despite its potential impact, the effect of allocation bias on inference in these trials remain unstudied. Methods: To model biased patient responses, we derived an allocation biasing policy tailored to stratified trials with multi-component endpoints. Using this policy, we assessed the impact of allocation bias by evaluating type I error rates of a stratified version of the Wei-Lachin test, integrating Fleiss’s stratified test with the Wei-Lachin test, when allocation bias was present but ignored during inference. Results: Ignoring allocation bias when applying the stratified Wei-Lachin test results in an inflation of the type I error rate, exceeding the 5% significance level. The amount of inflation depends on the number of strata, number of endpoint components and the chosen randomization procedure. Less restrictive randomization procedures, such as the stratified Big Stick Design, exhibited the lowest type I error inflation, while stratified Permuted Block Randomization, results in highest inflation. The inflation of the type I error increases with the number of strata included. An increasing number of independent endpoint components is also associated with higher inflation. Conclusion: Allocation bias threatens the validity of stratified clinical trials with multi-component endpoints evaluated using the stratified Wei-Lachin test and should be mitigated through careful study planning. Ensure the number of patients in each stratum is not smaller than the number of strata, restrict the number of endpoint components to those essential for the study’s objectives, and use randomization procedures that allow for some imbalances, such as the Big Stick Design, to reduce allocation predictability.