At a joint workshop of the FDA and the American Association for Cancer Research (AACR) last week, industry and regulators had a frank conversation about how the search for optimal dosing of cancer drugs needs to change.

At first blush, it might seem like the better question is what doesn't need to change. At the "Dose-finding for Small Molecule Oncology Drugs," workshop, for example, pretty much all major aspects of the current dose-finding process were criticized or acknowledged to be in need of a fresh approach.

The problems start with preclinical research, which remains, Eli Lilly and Co.'s Thomas Jones said, "still the quintessential black box."

While toxicity testing is fairly good at picking frequent toxicities, such as bone marrow and GI toxicity, it is much worse at predicting rarer toxicities. Jones described it as barely better than a coin flip for liver toxicities, and worse for other, rarer events.

Once a drug makes it, partly on the basis of those toxicology studies, into the clinic, the current default is to determine the maximum tolerated dose (MTD) by using 3+3 expansion cohorts. In many cases, that approach amounts to using a study you don't want to get data you don't really want, either.

In the still widely used 3+3 expansion cohorts, changes in dosing are driven solely by the reactions of the most recent patients.

The alternative approach, the Bayesian linear regression models (BLRM), is more complex and needs a statistician on the trial team, but it can explore the biologically effective dose range that goes from minimum effective dose (MED) via biologically efficacious dose (BED) to MTD.

For targeted therapies, MTD is less critical information than it is for chemotherapies. Indeed, about half of all targeted agents do not reach their maximum tolerated dose in clinical trials. And broader efficacy explorations are much more likely to uncover the sweet spot where "probably effective" and "very likely to be safe" overlap.

THE CLOCK IS TICKING

The simplest answer to why those more complex but also information-richer approaches are not universally used is that finding the optimal dose for a cancer medication suffers from the same tension as just about any task: The more thoroughly you do it, the longer it takes.

And time is something cancer patients don't have.

As one of the roundtable participants put it, a drowning person doesn't want to be thrown the best of all possible ropes – any rope will do.

Under such circumstances, a 3+3 trial design that comes up with a quick answer that is good enough exerts a powerful pull. Massachusetts General Hospital's Alice Shaw said the 3+3 design was "very successful" for identifying best dose for Pfizer Inc.'s ALK inhibitor Xalkori (crizotinib).

Increased pressure can come from the sense of having a highly effective agent, especially when that agent receives breakthrough status.

Novartis Inc.'s Don Howard described the company's situation when its ALK inhibitor Zykadia (ceritinib) received breakthrough designation. The drug's phase I trial suddenly turned into its pivotal trial, and "a lot of decisions were made in a very, very short period of time."

Adding to the sense of urgency was the fact that the drug was active in patients in whom the first ALK inhibitor, Xalkori, had stopped working. That led to "a great sense of urgency" in the Zykadia trial, said Massachusetts General Hospital's Shaw, who led the Zykadia trial.

As a result, many of the aspects of the trials were chosen because they provided "the justification we need in order to provide for the registration of this compound fast," Novartis' Howard said.

Zykadia is approved at a dose of 750 mg, and during the trial, which used Bayesian statistics to determine doses, the MTD was not reached, something Howard "still regrets" from a clinical pharmacology perspective.

"MTD is not the enemy," he said. "The enemy is using MTD as default condition to select your dose."