Adaptive protocol design challenges researchers to think outside the box

New generation of clinical trials on horizon

Imagine conducting clinical trials that provide a greater possibility of personal benefit to participants and reach conclusions in a faster and more efficient manner.

Thanks to technological advances and the efforts of investigators who made an effort to think outside the box, such a possibility can be a reality today. The only thing holding it back is the research industry’s reluctance to change the old methods and processes, says Donald Berry, PhD, a professor and chair of the department of biostatistics and applied mathematics at the University of Texas M.D. Anderson Cancer Center in Houston.

"There are conflicts here with people who have been doing something this way all of their lives and to say that what they’re doing is not the best way is a hard sell," Berry says.

Sometimes radical new changes only come about when someone new enters the field and looks at the old process in a different way. That’s what happened when Berry entered medical research from a statistician background.

Berry wanted to come up with a way to improve clinical trial research and improve patient care, but his ideas quickly met resistance. "It was like I was from Mars," Berry says. "They thought I was naïve and didn’t have an appreciation for what clinical research was about."

So Berry began working on clinical trials, following convention until he was well-enough established to discuss the possibility of change. A funny thing happened when he began to discuss changing the trial process: other people also began to discuss change, and more people began to listen.

"Other people have had these kinds of notions, these revolutionary ideas like we ought to be looking at the data that we collect during the course of the trial to understand where we’re going and to try to modify where we’re going based on what we see," Berry says. "In the past five to 10 years, these ideas have started to gather momentum in the pharmaceutical industry, and certainly in the medical device industry, and within some institutions."

Bayesian method vs. frequentist approach

Essentially, Berry promotes what he calls Bayesian statistical methods in clinical trials, which allow investigators to know the research answer much earlier than the standard approach. "If you know the answer you can stop the trial," Berry says. "Or if the answer is looking interesting you can modify the trial, and in some cases you might have to make the trial bigger because you are on the cusp."

The chief objection in the clinical trial industry is the fear of bias, Berry notes. "There are potentials for bias, but it is possible to accommodate that and get over those biases," he says. "Sometimes you have to view it as a cost benefit where the benefits are so great that you might have to sacrifice a bit of bias now and then."

This approach also could be called adaptive designs, says Jane Perlmutter, PhD, a breast cancer survivor who is on the national board of the Y-ME National Breast Cancer Organization in Chicago.

"The Bayesian approach and adaptive designs are all things that I think can help us more efficiently and effectively speed up the process of clinical trials," Perlmutter says.

"Adaptive designs modify the proportion of patients getting one treatment versus another as the treatment progresses," Perlmutter says. "I think that is advantageous to patients."

Traditional clinical trials are designed in the frequentist approach which expects a single experiment to lead to an answer, although in practice it typically will take converging evidence from a number of trials before an answer is considered complete, Perlmutter says. "In Bayesian statistics, the basic premise is we have a lot of knowledge and any experiences we have change our beliefs, and these have to be taken in the context of previous beliefs," Perlmutter says.

Perlmutter uses the analogy of an eye exam to explain how adaptive design works: "When you go to get glasses, it’s always frustrating when the optometrist says, Which is better: A or B?’ and you don’t know," she says. "But what the optometrist does is give you two extremes, and he rapidly realizes where your problem area is, and then he can adapt the eye test until you’re clearly within this small range."

In Bayesian statistics, clinical trials are put into that small range and kept there until there’s a clear conclusion, Perlmutter explains.

"Sometimes there will be a short test because it’s clear where the problems are," she says. "And sometimes you’ll keep testing because you’re not sure which the right one is."

Berry published a paper about Bayesian clinical trials in Nature Reviews journal, and it was accompanied by an editorial that called for further investigation of the method in order to solve the problems of long lag time in studying and approving new molecular entities (NMEs).1,2

Berry has been involved in clinical trials that use the Bayesian design approach, and he explains how the process works, using examples from his own work.

For instance, this approach was used in a three-arm trial in acute myeloid leukaemia. The study involved using troxacitabine (T) combined with standard therapies: first with idarubicin and separately with cytarabine. These arms were compared with an arm that used the two standard therapies in combination.1

"I said to the investigator, Let’s not randomize patients equally; let’s assign patients according to how they do on the various drugs and drug combinations, so if one arm was doing better we’ll assign patients to it with a higher probability, and if it is doing badly then we’ll drop the arm,’" Berry recalls.

In the study’s final results, the troxacitabine and idarubicin arm was dropped after it had 24 patients, and the troxacitabine and cytarabine arm was dropped after 34 patients, because both of the experimental arms had a lower complete remission rate than did the standard treatment, Berry says.

"Ten out of 18 patients had a complete remission in the standard arm, which is similar to historical results," Berry says. "In the troxacitabine/cytarabine arm there were only three out of 11 remissions, and in the other arm there were zero complete remissions."

When Berry and co-investigators sent an article about the trial to one journal, the response was that it was a lousy trial because an arm with complete data from only five patients does not have enough information to draw any conclusions, Berry says.

The second journal where the article was submitted loved the design and published it, Berry notes. "We’ve gotten really quite positive reactions," Berry says. "There are still some people who say, Maybe you made a mistake,’ and I say, Maybe we did, but if we made a mistake we didn’t make much of one because the Bayesian probability of benefit from the troxacitabine/idarubicin arm is quite small.’"

Berry asks people who question the trial’s design whether they would want to try troxacitabine/idarubicin if they got the disease.

One area of clinical research that has used the Bayesian statistical method more than others is the medical device field, Berry notes.

"They do the adaptive kinds of design we’re talking about," he says.

Also, drug companies are beginning to use adaptive design in dose-finding trials, Berry says.

"The standard dose-finding trial has you assigning the drug equally to a number of doses," Berry says. "So you close your eyes and then open them a year or two later and say, Oh rats! I wish I would have done something else.’"

With adaptive design, investigators look at data as the trial continues and fine tune the trial to where the dose response curve seems the most interesting, Berry says.

This may result in the trial being stopped or a particular dose arm being dropped.

With some trials already moving into the direction of adaptive design, this type of clinical trial design is the wave of the future, Berry says.

Pharmaceutical companies and the FDA are changing, partly due to necessity, he notes.

"It’s a novel kind of idea, and there’s no question that it will have to come, and it will come faster in AIDS than say in heart disease," Berry says. "It ties efficiency to treating patients better, and it saves patients on average and it saves time, which is something very important to pharmaceutical companies."

With the Bayesian approach, clinical trials can reach a conclusion more rapidly, and they have the potential of coming up with a better solution and giving more trial participants a better treatment, Perlmutter says.

"The Bayesian theory says, If while I’m running the trial I look at the data and see that most patients are doing better with one drug, then I should change the odds," Perlmutter explains. "If I give more patients that apparently better treatment than one of two things will happen: either I’m helping the patient and I more rapidly run to the conclusion that this is the better drug, or the arms will start to converge again and look alike."

Either way, the patients are treated better or no worse than they would be under the clinical trial with a traditional design, Perlmutter adds.

"But I think people are very skeptical and are concerned that the FDA won’t approve things that use those approaches, and so drug companies that would like to get their studies out faster and less expensively are saying, It’s no good if the FDA will send it back,’ and scientists are saying, I’m not sure this is rigorous,’" Perlmutter says.

Critics of the change might question why the Bayesian design hasn’t been used for decades if it indeed provides all of the reported benefits, Perlmutter says.

"The answer is we didn’t have the tools years ago to do the Bayesian statistical method because we didn’t have computational power," Perlmutter explains. "Traditional statistics can be solved by purely mathematical techniques, but with the Bayesian method, the mathematics is beyond computational approaches, so you have to use a simulation approach."

Perlmutter recalls her graduate school years pre-desktop computers when running a simulation would take all night. "Now a desktop computer can do the same thing in 30 seconds," she says.

References

  1. Berry DA. Bayesian clinical trials. Nat Rev Drug Discov. 2006;5:27-36.
  2. Editorial: Chance for change? Nat Rev Drug Discov. 2006;5:3.