When planning a conventional trial, one can anticipate the drug supply necessary for the trial by determining how the number of patients reflected in the sample size will distribute across the trial sites. Implementing an adaptive trial, by contrast, raises many challenges for predicting the necessary drug supply. It can require planning for different sample sizes depending on the outcome of an interim look; or preparing different dosages if certain arms of a multi-arm trial are to drop after the interim look. In the case of a biomarker-driven adaptive design, determining adequate drug supply may require the ability to predict which doses are necessary for different subpopulations at particular trial sites.
Optimizing over all of these eventualities can prove difficult. In the attached slides Nitin Patel, Cytel Founder and Chief Technology Officer, explains how to plan for drug supply for an adaptive dose-ranging trial.
The Safe Approach
Nitin offers the following scenario: A Phase 2a trial requires 120 patients with 40 patients randomized into the placebo group and 80 patients randomized into the active arm. The purpose of this dose-ranging trial is to test seven possible doses.
A standard trial would then require drug supply with 40 kits for placebo and 80 kits distributed across seven doses. By contrast, we might design an adaptive trial as follows:
The sample size of 120 is split into cohorts of 12 (40 placebo, 80 drug)
The first cohort is randomized to doses in equal proportion, with 3 in the placebo group and 9 into active arms with doses labeled dose 1, dose 2,…, dose 7
Each subsequent dose is assigned in response to data acquired during the trial
In this example, we still need 40 kits of placebo. Let us say that for the first cohort we also need 2 kits for Dose 4, and 1 kit for the other six possibilities. However, we cannot know at this point, what drug supply we will need for the remaining nine cohorts of patients, as the doses assigned will depend upon data gathered during the trial. Nitin writes that a ‘safe approach’ is then to provide 72 kits for each of the seven doses (i.e. 9 cohorts with 8 subjects for each cohort.) This would come to a total of 552 kits prepared for an adaptive trial, where a conventional trial would only require 120 kits.
Thankfully, there are techniques available to adaptive trial planners that can reduce this number. Both require simulations and assume that drug supply is distributed using some form of inventory control. By way of establishing preliminaries:
Simulations: Instead of utilizing the Safe Approach, it is possible to use simulation software to optimize drug supply. Such simulation techniques are an extension of those tools already used for adaptive trial design.
Inventory Control: The Inventory Control approach begins by sending a certain amount of drug supply to each trial site. A ‘trigger’ or ‘floor’ level of stock is specified, such that when the inventory at a particular site dips below the trigger or floor, new stock is sent to the site. (Note that the inventory includes the available stock at the site, plus the stock that the site has already ordered.)
The Adaptive Bayesian Design: There are two methods one can employ for the Adaptive Bayesian trial design. The more conservative one provides sufficient drug supply no matter what scenarios play out during the trial. (The other method predicts what scenarios will play out using prior probabilities, and plans for drug supply accordingly.) Using the more conservative method, Nitin's simulations reveal 193% excess (or overage) in stock. The conventional trial, by contrast, had a 113% overage.,
The Adaptive Drop Arms Design: As a result, Nitin suggests adopting an adaptive ‘arm-dropping’ design. The simulations in the attached slide show that a slightly different trial design can improve logistics substantially while still improving trial performance when compared to a conventional trial. Simulations of the Adaptive Drop Arms Design shows drug supply overage at 130% compared to 113% in a conventional trial and 193% in the Adaptive Bayesian Design.