New Methods For Clinical Trials Of Investigational Agents

The likelihood that many combinations and permutations of LIs, with each other and with HIs, will be investigated in the future leads to consideration of desirable elements to incorporate in the relevant statistical designs. These elements include the need to (1) randomize early in the investigative process; (2) formally monitor multiple outcomes; (3) account for the possibility that the effects of a given treatment may depend on the treatments given before or after; and (4) develop strategy that views drug development as a process rather than as a series of unconnected trials. Topics (2) to (4) are examples of the general problem of "multiplicities." Most statistical designs underlying LI protocols have not focused on these issues.


The great majority of LIs are tested in single-agent phase II trials. This practice reflects the conventional phase II ^ phase III paradigm. Specifically, phase II trials are "exploratory," designed to establish activity, with the idea that comparative trials (phase III) should be conducted only after activity has been observed. A fundamental problem with this formulation is that phase II trials are inherently comparative. In particular, patients are vitally interested in whether a particular therapy is superior to another. The comparative nature of phase II trials is implicit in the designs governing their conduct, as these designs specify minimally acceptable response rates,49 derived via comparison with other available therapies.

Although phase II trials are thus inherently comparative, the current emphasis on single-arm, nonrandomized phase II trials provides an unreliable basis for treatment comparison because of treatment-trial con-founding.50 Indeed, it is logically inconsistent that the need to avoid confounding trial and treatment effects is addressed by randomizing in phase III, yet is ignored in the evaluation of phase II data that determines whether the phase III trial will be conducted in the first place. These considerations emphasize the desirability of randomization among various treatments and strategies in the early stages of their development. Designs for this purpose have been described.50 In general, these designs call for randomization of a relatively small number of patients among a relatively large number of therapies. Enrolling fewer patients on each trial permits a larger number of treatments to be investigated. This ability is useful because preclinical rationale is an imperfect predictor of clinical success. Thus, although interferon is the only drug currently known to prolong survival in chronic myeloid leukemia (CML), its mechanism of action remains unknown. The development of all-trans-retinoic acid (ATRA) for APL is now cited as an exemplar of the "bench-to-bedside" paradigm. However, it is important to remember that the initial Chinese reports of high CR rates were greeted with skepticism in the West; presumably, there would have been less skepticism if a bench-based rationale were readily apparent. Indeed, with the exception of imatinib, most of the drugs that have improved outcome for patients with leukemia (2-chlorodeoxya-denosine in hairy cell leukemia, interferon in CML, ATRA and arsenic trioxide in APL) are examples of "bedside-to-bench" development. If preclinical rationale cannot yet replace empiricism, we should examine a larger number of therapies. Thus, rather than randomizing 240 patients between a standard and an investigational induction regimen, we might be better served by randomizing the same 240 patients among a standard and three investigational induction regimens. It is true that such trials will be nominally "underpowered." However, this argument ignores the false negative rate inherent in the selection of which investigational regimen to study. For example, if there are three potential regimens that could be investigated and if preclinical rationale is, as argued above, a poor predictor of clinical results, then limiting ourselves to one regimen in effect potentially entails a false negative rate of 67%. Simply put, the most egregious false negative results when a treatment is not studied at all.

0 0

Post a comment