 |
| April 12, 2010 Part 2 As you’ll recall from last week, we introduced the concept of adaptive trial design, how it can serve drug development investigators and set the stage with a top-line summary of a virtual clinical trial – a compilation of several real life studies - in which an adaptive design was utilized. Click here if you would like to link back to review Part 1. In this virtual case study, real-time learning about the dose-response was deployed in a large, multicenter Central Nervous System (CNS) study. The adaptive design permitted early determination of the failed nature of the study, i.e. the inability to separate the active reference compound from placebo. This early determination led to significant conservation of research resources; whereas, a non-adaptive approach would have involved enrolling over twice the number of patients. The adaptive approach exposed fewer patients to the unnecessary potential risk inherent in the study of an investigational compound. At the same time, the adaptive design allowed more efficient learning about the dose-response of the investigational drug than would have been possible in a conventional fixed dose study. The absence of any indication of a dose-response curve for investigational drug further reinforced the conclusion that the drug was without efficacy in acute exacerbation of the target illness. This finding coupled with the results of previous fixed dose studies lead to the decision to discontinue the development of this agent for the indication being studied. In the previous trials with the compound the data suggested the possibility of a curvilinear dose-response curve. That finding could have been either a chance effect or could have suggested a complex dose-response relationship for the drug. In regard to the latter possibility, an additional advantage of the adaptive design, which used a normal dynamic linear model (NDLM) to learn about the dose-response, was its ability to efficiently learn about non-monotonic dose-response curves. This virtual study examined the effect of multiple doses of the investigational drug with a positive (i.e., established comparator drug) and negative (i.e., placebo control). The study would have enrolled over 500 patients if it had gone to completion but instead only 250 had to be enrolled to detect study failure. To those not familiar with adaptive designs, the relatively small number of subjects on any given dose of the investigational drug might raise a question about the degree of confidence that could be put on the findings with any one dose. However, the strength of the modeling approach is that it uses all of the information from all doses in estimating the dose-response curve rather than simply doing a pair-wise comparison between study drug dose arms and placebo. Allocation to placebo and to the active compound initially were fixed, so that they could serve as reference points for the investigational drug just as they traditionally do in non-adaptive trials. However, after an initial “burn-in”, the adaptive algorithm permitted deviation from the initial fixed proportion of patients allocated to the controls, should the variability of the placebo and active control data, as observed in the real trial, be different from the variability that was assumed when planning the trial. Running a trial with many treatment arms can be challenging, both conceptually and logistically. In this specific virtual study, population pharmacokinetics provided indirect evidence for successful administration of the different doses employed in the study. Compliance was not a problem given that patients were confined and medication was prepared for them by nursing staff. Steps were taken to ensure maintenance of blind and to assure proper dosing once assignment to a dose arm was done. In the final installment, we’ll look at why this virtual study failed, the lessons learned and comments on trial design. |
 |