DIA: Adaptive designs
| Peter Thomas of Novartis spoke in the theme on adaptive study designs about the involvement of sponsor senior management with Data Monitoring Committees (DMC). He started by saying that this involvement is widely discouraged, eg in the EMEA discussion paper. However, sponsors often feel uncomfortable with the notion of not having access to interim confirmatory data during the study, and also if the data evolves in a way not predicted as a possibility in the pre-trial modelling. He cited an ongoing Novartis study in which phase IIb and phase III are combined, with certain dose and comparator arms being continued into the later phase while others are dropped. To do this, the DMC require a number of rules to be defined to guide them in determining which doses to drop. This is done by defining a threshold for efficacy, with the lowest dose producing this effect is retained. Of course, safety issues would be weighted against this. Peter set out a number of scenarios that were considered, from the ideal to the less than ideal (eg, none of the doses meet the threshold) and the unexpected, which cannot be parsed according to the preset rules. One might expect that the DMC would want to consult with the sponsor in the latter case. Peter then set out the standard composition and process of a DMC. He proposed that sponsor involvement could be permited if a clear rationale existed, based on the complex unexpected situation, and given that the individuals involved would be suitably distanced from the study (eg, senior clinician and statistician) and only be involved at the point where their decision was required. In some cases, Peter stated, sponsor involvement might be required. The sponsor must make a strong case that appropriate procedures are in place and follows, and these will need to demonstrate that the integrity of the study is intact, and paramount. The final presentation in this session was given by Jerry Schindler of Merck. He gave a simple and easy strategy for implementing adaptive trial design. He summarised this concept as designing a study using information that wasn’t available when you started designing it! His role is maximising benefit (eg, earlier decision-making or reduced sample size) while minimising risk (eg, not being able to do the trial). Having written off an entirely hybrid phase I-III approach, he looked at the different information points discovered in the different phases, to explore the potential to group them. He thought that the division should come just before the pivotal phase II study, to combine phase I with phase IIa, and late phase IIb with phase III. As a product moves from early development to pre-registration development, the number and variety of adaptive options are constrainted. These constraints reflect the fact that you’ve learned something from the earlier studies. Once you’ve constrained these options, the process becomes simpler. In pre-pivotal studies, he argued that statisticians could be able to see unmasked data as it comes in, enabling a flexible decision process to consider many options that could allow options to be eliminated or even new ones to be added. Conversely, in the pivotal phase, masking and use of an independent DMC are often necessary. Decisions should be driven by algorithms (contrary to the view of the previous speaker) with limited and decreasing options and futility analysis of individual dose arms during the study. Jerry closed his presentation by summarising his 8 key points. |
