91��Ƭ�S�ϼ�

What if clinical trials could be conducted faster, required fewer patients, cost less, and led to improved health? "Our approach is good for patients, companies, and the public health at large, which would come from needed drugs being on the market sooner," states Vishal Ahuja, ITOM professor at SMU Cox. Over many years, Ahuja and his co-author Birge have developed a breakthrough approach that is long overdue in healthcare. "It's not just about big pharma making money, but people's health being potentially saved from a needed treatment."

Given the rapidly rising costs of bringing a new drug to the market, using resources more efficiently in the drug development process is a high priority. Clinical trials, and particularly phase-3 trials, have been cited as a major contributor to these costs, estimated at $300–$600 million.

Traditionally, clinical trials have followed a fixed (non-adaptive) design that randomly assigns patients to treatments in a constant proportion throughout the trial. These designs often result in lengthy trials and poor outcomes, so leading pharmaceutical firms and regulators are searching for more efficient designs.

Recently, the U.S. Food and Drug Administration (FDA) has encouraged the use of adaptive designs, in particular those that follow Bayesian principles, as statistically efficient and ethically sound alternatives to traditional fixed designs. However, the regulation of these types of designs is lagging for various reasons.

New approach

The bulk of work in adaptive design development has been in sequential or “one-person-at-a-time” trials. But, most trials have multiple patients that need to be randomly assigned at the same time. Ahuja and his co-author proposed a design that allows for this randomization of several patients at the same time. However, developing a protocol for implementing adaptive designs (which FDA requires before a trial can begin) means that one must consider all possible scenarios and decide how many patients must be allocated to a treatment drug or a placebo in each scenario. Given the large number of scenarios, often in billions, solving them poses huge computational challenges. This has been a key barrier to the implementation and adoption of adaptive designs in practice.

Ahuja and Birge have developed a computationally efficient approach that they term SLAX,* a heuristic or a rule-of-thumb that allows for implementation of adaptive designs. "One need not consider every possibility; for example, if homework should be done during a two-hour window surrounding dinnertime, a rule-of-thumb, or a heuristic, would be to start approximately an hour before dinner and an hour afterwards," Ahuja explains.

Adaptive designs are hard to solve, says Ahuja. (And this is one reason his research received recognition at a recent operations management conference.) Heuristics are the best way to limit scenarios, he says. Rather than considering every single patient, the authors devised a heuristic whereby they consider patients in blocks, for example, 10 patients at a time.

Even then, substantial computational effort was needed to solve and develop a protocol that specifies how many patients to randomize in each scenario. The authors then simulated the model multiple times on the computer. Ideally the simulation should be run millions of times. Given that it is prohibitively expensive (computationally), they settled on around 5,000 times, which worked well, another rule-of-thumb. Ahuja notes that these types of problems, "are a combinatorial nightmare, and, given the limits of computation, may not have been possible to solve a few years ago."

A big advance

The authors tested their approach retrospectively on three phase-3 clinical trials that studied the safety and efficacy of the drug Rolapitant, under the trade name Varubi. The drug helps with chemotherapy-induced nausea in adult cancer patients. The trials concluded that Rolapitant is a superior treatment. According to Ahuja, the cost of one patient to participate in and complete a phase-3 trial is estimated at $42,000. Further, notes Ahuja, "the average cost to bring a drug to market is $2.6 billion, according to a recent study from Tufts.”

The authors' results showed SLAX to be superior compared to existing heuristics and other popular methods. SLAX could have reduced the number of failures (patients who did not receive the treatment) by 17%. Further, SLAX could have reached the desired conclusion—that is the identification of the most efficacious treatment—quicker and/or with fewer patients than the traditional fixed designs. For the Rolapitant trial, this meant arriving at the conclusion using 21.6% fewer patients than currently used in the trial. This is equivalent to saying the same result could have been found 18 weeks sooner had adaptive designs been used. Bringing a drug to market earlier has potential for health benefits to patients who need this drug, but also potential monetary benefits to the firm.

Regarding their approach, Ahuja says, "On every dimension, the outcomes are better." The drug companies want to get their products to market sooner and at lower cost, so this is a major step forward. The advance is "a big deal," according to Ahuja.

Randomized trials have been used for about 60-70 years, with the FDA trying to figure out how to regulate trials and use adaptive designs. "They opened their use for deadly diseases, such as the I-spy trials in breast cancer," explains Ahuja. "Imagine the possibilities using adaptive designs for HIV or AIDs drugs, with so little work done to date."

The research brings the world of adaptive designs to reality. "It's not a perfect solution, so to speak, but I think we provide a good enough solution that it can be applied to the real world," Ahuja surmises.

The paper " An Approximation Approach for Response Adaptive Clinical Trial Design" by Vishal Ahuja, 91��Ƭ�S�ϼ�'s Cox School of Business, and John Birge of University of Chicago Booth School of Business is under review. The paper was a semi-finalist for the "Best Paper" in healthcare operations at a recent Production and Operations Management Society conference.

* Simulation-based bounded Learning-adjusted ApproXimation

Written by Jennifer Warren.