|
Randomisation is the cornerstone of statistical inference and, as such, scientists go to great lengths to preserve its integrity. Randomisation is particularly important in the pharmaceutical industry, as regulatory agencies worldwide insist on definitive clinical trials conducted in a randomised fashion. In fact, most of these definitive trials are conducted with the clinician and patient unaware of the particular treatment being received by the patient (double-blind trials). This is done to eliminate potential bias in the collection and reporting of information about the patient. Additionally, the statisticians responsible for summarising the information from these trials are also kept ‘blinded’ to the randomisation scheme prior to conducting the final analysis so as not to bias the manner in which the data are scrutinised.
Randomisation can come in many forms. The first of these is simple random sampling (SRS). With SRS, each patient is equally likely to get any of the possible treatments. This presents a problem in multi-centre trials as a balance between treatments is maintained overall but not within a given centre. To better accommodate the multi-centre trial, stratified random sampling can be performed where each centre is considered as a stratum. This greatly improves the balance within centres, but is still not completely effective. In practice, most multi-centre trials use a permuted block design, which allows for the balance between treatments to be maintained within each centre over the course of the trial. There are situations where it would be beneficial to favour one treatment arm over another, based on informative covariates or patient outcomes. In these cases, adaptive randomisation may be useful.
OBJECTIVES OF ADAPTIVE RANDOMISATION
Adaptive randomisation seeks to use cumulative patient information, being gathered on an ongoing basis, to determine the allocation of subsequent patients to the treatment arms.
Adaptive randomisation trials can improve the efficiency of clinical trials by enrolling fewer patients into treatments that are less likely to succeed, by improving the quality of results so that more patients are enrolled in successful treatments, and increasing the chances that patients will receive effective treatments.
TYPES OF ADAPTATIONS
Covariate Adaptive Randomisation When covariates are known to be correlated with the response being studied, it is often desirable to balance on these factors using a stratified approach. In certain situations, however, it is impractical (due to financial considerations or insufficient patient population) to use a stratified approach within each study centre. Covariate adaptive randomisation can be an alternative to stratified random sampling. The system works by balancing important prognostic factors across the entire study, while allowing a balance between treatments to be maintained within each study centre. This is done by examining the balance within the centre, as well as the balance across the study with respect to the covariates, and then using prespecified rules to favour the under-represented treatment group on the factor which is the most out of balance. Several methods have been proposed and they are discussed briefly here. |
