Professor Turnbull Comments on the Use of Adaptive Design in Clinical Trials

Clinical trials, especially those in the late stage of the testing of new pharmaceutical products,  medical procedures, and medical devices, can cost tens to hundreds of millions of dollars.  Traditionally such trials entail administering the drug or procedure on a randomized basis, with some patients receiving the innovation while others do not, the choice being made at random and without knowledge of the clinicians and patients ("double-blind trials").  Many trials have several 'arms' where, in addition to those receiving a placebo, subsets of patients are given different dosing levels or other treatment variations.  At the end of the trial, the data are 'unblinded' and statisticians analyze the level of effectiveness.  

A new approach is increasingly employed by companies such as Wyeth, Eli Lilly and Novartis, who allow designated individuals to see the interim data and recommend changes in the trial design, such as dropping a clearly unproductive arm -- or stopping the trial altogether.  But in an article in The Scientist, a "magazine of the life sciences,"  statisticians such as ORIE's Professor Bruce Turnbull are quoted as expressing caution about such approaches, which require novel statistical methods and entail possible pitfalls. 

"If you're always tweaking [the trial] to get the best result possible," says Turnbull, "then you will get the best result possible."  He notes that the statistical maneuvers required to counteract that possibility are far from straightforward.  

The Food and Drug Administration, which regulates pharmaceutical trials, already requires that companies use an independent data monitoring board made up of people uninvolved in the company or trial to evaluate and oversee the trial, empowered for example to stop the trial if there is excessive risk to participants.  As part of this oversight, such a board can 'unblind' the data without revealing it to the company.  This is also a step which is inherent in the use of adaptive design.  Hence the purview of monitoring boards is being extended to cover responsibility for aspects of adaptive trials.  But "it's a tough pill for companies to swallow, putting decisions about a trial into the hands of totally independent bodies," says Turnbull.

Turnbull has written a number of articles on adaptive trials, which are discussed in his 2000 book "Group Sequential Methods with Applications to Clinical Trials," coauthored with Christopher Jennison of the University of Bath in the United Kingdom.  Turnbull was also quoted in a Wall Street Journal "Marketplace" article about adaptive trials in 2006 that reported on the initial FDA acceptance of such trials.

Adapting analytical approaches as results are discovered is not new to statistics.  Abraham Wald, with the late Cornell professor Jacob Wolfowitz, pioneered sequential analysis for industrial quality control in World War II.   After the war Wald and Wolfowitz were at Columbia, where the late ORIE professor and director Robert E. Bechhofer earned his Ph.D.   Bechhofer was one of Turnbull's thesis advisors.