Clinicians must regain control over phase 1 cancer combination-therapy trials


April 16, 2019

This is a lay explainer for my bioRxiv preprint Costing ‘the’ MTD … in 2-D.

What is it about?

Combination therapies have long been important for treating cancer. Figuring out how to dose even a single drug can be challenging enough; figuring out how to dose two drugs in combination presents an entirely new level of difficulty! This paper suggests ways to think about and confront the special challenges presented by ‘phase 1’ dose-finding trials for combination therapies.

The rationale for giving two drugs in combination usually comes from a hoped-for synergy between them. Two drugs may sometimes act together in the body, so that they complement each other. When this happens, the drugs could have stronger beneficial effects when given together as combination therapy (A + B) than separately as sequential monotherapies (first A, then B). Unfortunately, the toxic effects of drugs may also compound each other in a similar way. Therefore, the right doses for A and B when given in combination are not necessarily the same as the right doses of A and B when given separately. Somehow, we must strike a balance between synergistic benefits and compounded toxicities.

In this paper, I advance two ideas as crucially important for confronting this problem. First, as in my earlier work on monotherapy dose-finding, I stress the importance of dose individualization as opposed to 1-size-fits-all dosing. Second, I argue that this 2-dimensional problem (finding 2 doses simultaneously!) is so difficult that we must take full advantage of clinicians’ detailed understanding of toxicity. Specifically, when clinicians can assign blame for a particular toxic effect to one of the two therapies A or B, that knowledge should be used to guide dose titration rationally.

Figure 3 in the paper serves to make these two ideas more concrete:

This figure shows a scenario with 24 different patients’ hypothetical experiences of 3 distinct toxicities: neutropenia, mostly attributable to drug A; diarrhea, chiefly due to drug B; and fatigue, a ‘nonspecific’ toxicity that can’t be blamed solely on either drug. The colored regions in Figure 3 indicate which doses are intolerable from the perspective of each of these toxicities. Some of these colored regions overlap (like colored cellophane), showing that more than one toxicity is intolerable. (For example, the points inside brown=pink+green areas correspond to combination doses where both diarrhea [pink] and fatigue [green] are intolerable.) To understand Figure 3 better, consider yourself as one of the patients plotted. Working within ‘your’ box in the Figure, start with a very small dose of (A+B) near the lower left corner, and slowly increase the dose moving upward and to the right. Which toxicity do you ‘hit’ first? If you are patient #17 or #20, then you’ll hit (green) fatigue first. But if you are patient #2 or #16 you will hit either intolerable neutropenia or intolerable diarrhea first—or possibly, both together—before your fatigue becomes intolerable.

The methodologists who design today’s 1-size-fits-all combination dose-finding trials typically ignore the kind of detail and clinical realism elaborated in Figure 3. Instead of forthrightly recognizing these realities, and collaborating with clinicians to confront them, these methodologists prefer to tinker with technical novelties that sweep such realities under the rug. (In the paper, I refer to the attitude driving this behavior as “abnegation of clinical realism.”) One example of this approach is the POCRM method, which joins a technical idea called Partial Orders (PO) to a quarter-century-old dose-finding technique called the Continual Reassessment Method (CRM). In the second half of this paper, I use the scenario of Figure 3 to criticize POCRM.

This criticism has 2 parts. In the first part, I run a simulation showing that 1-size-fits-all dosing based on POCRM can be much less efficient (in terms of patients’ outcomes) than individualized dosing based on a titration procedure. This comparison is shown in the ‘Pragmatic’ column of Figure 7, with a 38% loss of efficiency:

To be clear, I am not claiming that this one simulation definitively proves anything, or even that it yields a representative or typical efficiency loss. But I do think this simulation shifts the burden of proof to the 1-size-fits-all dose-finding methodologists, who now ought to demonstrate under what conditions—if any—their methods perform well. (Lay readers may safely ignore the ‘Oracular’ column of Figure 7, and the theoretically-minded discussions of this in the main text. But the intention of my concept of ‘oracular method’ can easily be understood by analogy with energy efficiency. Sometimes, basic physics can establish a theoretical upper limit for the efficiency of a technology—such as an internal combustion engine, or a light bulb. This limit provides an absolute standard, against which we can say, for example, that modern automobile engines are up to 35% efficient in converting gasoline to motion, or that incandescent bulbs are only 2% efficient in converting electricity to light. The top, left corner of Figure 7—the optimal, individualized, oracular method—provides just such a theoretical standard for the other methods examined.)

The second part of my criticism of POCRM appears in the final paragraph of the Discussion section, where I underscore my point about “abnegation” mentioned above. I point to the “barren artifice” of Table 2 from the paper, noting that this Table essentially depicts POCRM’s ‘worldview’, since it contains all the key inputs needed to run a POCRM study. I close by quoting the great Lewis Sheiner, the father of pharmacometrics, as he urged his fellow clinicians to insist on more realism in clinical trials—advice that remains as important today as it was when first issued more than a quarter-century ago.

Why is it important?

Today, the Biostatistics profession dominates the design of phase 1 dose-finding trials. Unfortunately, the standard training of biostatisticians omits ideas, attitudes and techniques that are fundamental for understanding and developing dose-individualization methods. Progress toward tomorrow’s dose individualization designs will therefore depend on increasing the awareness and influence of other professions implicated in the design and conduct of today’s 1-size-fits-all dose-finding studies. This paper reaches out directly to clinician-investigators as important partners who hold the critical insights and natural interest needed to promote dose individualization in early-phase trials.