Often in clinical trials for the validation of a drug, one has to compare a test and a control arms. Although setting up and implementing a traditional clinical trial is complex, the aim is ultimately to perform a simple hypothesis test to prove superiority (or at least non-inferiority) of the test arm with respect to the control arm. Patients that are recruited for the trial are generally randomized to one or the other arm, and most of the time the investigators as well as the patients are “blind” to this randomization. Sometimes, there are factors related to the patient characteristics that are known -or suspected- to have an impact on treatment outcome (weight, age, smoking or not, etc.). These are called confounding factors. In large (Phase III) trials, the distribution of the patients between arms with respect to these confounding factors is supposed uniform. In smaller trials, typically Phase I or II, this is really not a probable scenario. As a consequence, even observing a statistically significant difference between the arms of the trial does not allow to conclude that this difference is a consequence of the tested treatment: it might come from the confounding factor’s imbalance between arms.
It is possible to dynamically allocate patients as they enter a trial to various arms in such a way that a good balance of one confounding factor is maintained. It is however much more complicated to reach this objective for several confounding factors at the same time. For a project run in partnership with Tools4Patient, we were asked to design a web service directly available to the clinician involved in one of their trials. The service would stratify the patients for four confounding factors at the same time.
What we did
We developed a procedure able to reach this goal of multi-factorial balancing in a dynamic way. Although a perfect balance cannot be guaranteed, we showed that our procedure divided by 15 the risk of a non-conclusive trial. On top of that, a very convenient web interface allowed both distant investigators and local sponsor representatives to access the stratification application and manage the cohort right from their laptop or tablet.
“DNAlytics proposed to develop and implement this online solution for multifactorial balanced patient stratification for our common project. We have been convinced by the approach and its statistical motivation. The application integrated neatly in the clinical recruitment practice. It allowed different people in the project to have different role/permissions in the application. The stratification strategy could even adapt dynamically to the loss of drop-out patients.”
Dominique Demolle, CEO Tools4Patient.
Do you have to deal with randomization in presence of potential confounding factors?
Do not risk being unable to draw conclusions from your study.