By Bill Byrom and Graham Nicholls
August 18, 2008 | Adaptive clinical trials use the analysis of accumulating subject data to make changes to the study without undermining its inherent validity or integrity. Over the years, the industry has made use of such design types, including Phase I cohort studies, group sequential designs and mid-study sample size re-estimation.
More recently, interest has focused on study designs that incorporate decision rules enabling individual treatment groups to be dropped or added, or the treatment allocation ratio adjusted, as the trial progresses. These features enable, for example, more treatment groups to be investigated at dose finding stages – making identification of optimal dose more accurate without large increases in sample size or timeline – and seamless Phase II-III designs that may shorten the overall duration of drug development.
Such designs bring with them new implementation challenges: rapid access and cleaning of response data, seamless execution of randomization changes and ensuring sufficient supplies are at site following a randomization change.
Accessing and Cleaning Response Data
A key concern for researchers when performing interim analyses is how clean the data needs to be. Simon Day, in his investigation of the impact of erroneous data, showed that analyses are insensitive to random errors, and simple range checks enable the main errors that affect analysis conclusions to be corrected [see ref. 1]. Applying this to adaptive trials, we recommend that you use a process that enables point-of-entry range checking and rapid data cleaning so that errors that may affect analysis conclusions can be eliminated and data cleaned efficiently. This ensures that while you are striving for the cleanest possible data at all times, all data are used, whether cleaned or not, in each mid-study analysis.
Data that are modified after an analysis run can be replaced by the updated values in subsequent analyses. Although this can be accomplished using paper case report forms (CRFs), where data collection and cleaning can be geared up around scheduled interim analysis timings, it is not as easy to do with response-adaptive designs, where data are fed into the algorithms continually. In both cases, electronic solutions such as electronic data capture or electronic patient-reported outcomes are optimal in providing clean data quickly. However, other solutions such as interactive voice or web response (IVR/IWR) and fax with optical character recognition can also be successfully used. In these cases, however, because the data are collected via an additional mechanism to the main data management process, it is essential to include checks and balances to ensure the adaptive dataset is always consistent with the clinical database.
Implementing Randomization Changes
Central randomization solutions are essential components of adaptive designs, as these provide rapid and error-free changes to the randomization algorithm. Such solutions (e.g. IVR/IWR systems) provide minimal interruption to the ongoing recruitment and implement changes without the knowledge of site personnel. (This is important in demonstrating the avoidance of selection bias, which is possible if investigators know when sub-optimal treatments are dropped.) Among the numerous ways to accommodate adaptations are: switching between pre-generated randomization code lists; modifying an existing list; automatic or manual generation of new lists; and use of dynamic allocation methods where the treatment assignment is determined by comparing computer-generated random numbers to treatment assignment probabilities that change over time.
When a design includes scheduled interim analyses, the switch to a new randomization scheme can be made using an IVR call performed by a designated user – perhaps a member of the data safety monitoring board. When using a response-adaptive algorithm, automated integration between the algorithm output and the randomization system is recommended. Other points to consider include: (i) the procedure for handling subjects requiring randomization while an interim analysis is in progress, (ii) whether to withdraw or continue ongoing subjects randomized to a dropped treatment group, and (iii) making sure sites have sufficient supplies to accommodate the randomization change.
Ensuring Site Supplies Are in Place
When a change to the study randomization is made, it’s vital to ask whether each site needs to be re-supplied with study medication in light of the design change and whether there is enough medication currently at sites to ensure dispensing to new and existing subjects in the time required to ship additional supplies. This question is often more challenging when a design incorporates a small number of scheduled interim analyses, as the resulting design changes can be more dramatic.
In addition to making an immediate assessment of site supply inventories prior to changing randomization, IVR/IWR systems should adjust site supply strategies when appropriate. For example, if the proportion of subjects expected on a specific treatment group increases significantly, the supply scheme should accommodate a higher supply level of these packs moving forward. Mid-study supply simulations and forecasts are also valuable in understanding ongoing requirements following a major adaptation.
Adaptive trials provide potentially exciting enhancements to drug development. Well-considered use of integrated technology solutions facilitates their effective implementation.
Reference:
[1] Day, S. et al. (1998) Double data entry: what value, what price? Controlled Clinical Trials; 19:15-25
-------------------
Bill Byrom, PhD, is vice president of product strategy, and Graham Nicholls, MSc, is product manager, ClinPhone plc, Nottingham, U.K. Email: [email protected].
_______________________________
This story first appeared in eCliniqua,one of Bio-IT World’s free e-newsletters. Subscribe here.