Adaptive trial designs may offer more flexibility than their static design equivalents but, as Andy Grieve of Aptiv Solutions explains, the need to plan a trial in advance remains as pressing as ever
The introduction of statistical thinking into modern medical research can be traced back to 1948 and the famous MRC trial of streptomycin for the treatment of tuberculosis, designed by Sir Austin Bradford-Hill (1). This trial introduced blinding and random allocation to a modern audience, and since that time the randomised controlled trial (RCT) has been seen by many as the gold-standard for the conduct of human studies.
A characteristic of the trials designed following the prescription exemplified by Bradford-Hill is that they are static, in that the key elements driving the designs are specified in advance and do not change. These include the significance level, the primary endpoint, the primary hypothesis, the type of test and test statistic, and the sample size based on a specified effect size and required power. The trial is then run, the outcomes for patients observed, estimates of parameters determined, p-value(s) for specified hypothesis(es) computed and a conclusion drawn.
An obvious problem with this approach is that the planning of such trials is based on a fixed set of assumptions. For example, assumptions are made about the likely effect size, the variability for continuous variables and the control rate for binary data, and each of these is either subject to uncertainty, or at the time of planning is unknown. Incorrect choices for these parameters can lead to trials which, in some cases, are under-powered and in other cases over-powered, and both of these outcomes are ethically questionable. In such trials, the data observed during the trial are not used to guide its course. One reason for this is that the set-up provides a solid, largely accepted, basis for statistical inferential procedures, and that the dangers of bias are minimised by keeping all the parties involved ignorant of the allocation of patients to treatments and the results, until the final unblinding at the end of the study. However, such a prescription does leave considerable scope for improvement in terms of the efficiency of the design. There have been different ways to improve efficiency proposed over time; for example, allowing dynamic modification of a trial’s design during its course, based on accumulating data. This idea has lead to the formation of a broad group of methods known today as ‘adaptive designs’.
ADAPTIVE THINKING: A NEW APPROACH
Recently, pharmaceutical statisticians have begun to question whether the traditional static RCT is always the most appropriate design for use in drug development. In particular, they have investigated the benefi ts and practical feasibility of utilising adaptive designs to increase the chance of a positive outcome and to increase efficiency. Despite a long history of statistical research into adaptive designs, until recently only very few have been reported in medical literature. This article reviews the application of adaptive designs to drug development including their advantages, while not forgetting the challenges that using such designs may involve. It is important to stress that adaptive designs are neither a panacea for the ills in pharmaceutical R&D, nor are they a solution to badly run clinical studies. Adaptive designs need as much, if not more, planning than static designs, and while they allow flexibility, this flexibility is planned in advance. Adaptive designs are adaptive by design, and not randomly adopted to overcome a problem discovered during the course of running a study.
It is generally accepted that the greatest scope for the implementation of adaptive clinical trials is in the ‘learn’ phase of clinical development that is up to the end of Phase 2 (2). Prominent in this phase of drug development are the Phase 2b dose-response studies, whose results drive the selection of the appropriate population, dose and endpoint for the Phase 3 programme. The high failure rate of Phase 3 trials, estimated by many writers to be as high as 40-50 per cent in general, and higher for specific therapeutic areas (for example it has been suggested that it can be as high as 60 per cent in oncology programmes) has often been laid at the door of the Phase 2b design (3).
NOTES FROM A TRIAL
In the mid-1990s I was working for a large pharmaceutical company and was involved in a project looking at the company’s process for dose selection. One important aspect of the project was an investigation of historical programmes and their success or otherwise in preventing the reworking of Phase 2b dose-finding studies. Why should this be an issue?
As an example, consider the development of drug X in indication Y (a real example, anonymised) which started with a study examining three doses of X (80mg, 120mg and 160mg) and a placebo in a large parallel group study. When analysed, the results showed that all three doses were significantly different from the placebo, but each dose was not any better than any other – in all probability the doses lay on a plateau. In the light of these results, a second, larger study, was conducted examining doses of 40mg, 80mg and 120mg, and the placebo again showing similar efficacy among the doses, with all doses exhibiting greater benefit than the placebo. A third study involving doses of 2.5mg, 10mg and 40mg and the placebo was eventually able to establish the rising part of the dose-response curve. This programme, and other similar examples, illustrates an important consideration in dose-response studies – the range of doses. Before the first study was undertaken, the doses were chosen on the basis of preclinical data, but with a relatively tight range of doses – a two-fold range. By the end of the programme a range of 64-fold had been examined.
An obvious question is whether this programme could have been made more efficient. The answer is ‘almost certainly’. For example, initially it would have been more efficient to have chosen a wider range of doses with a smaller numbers of patients per group; as the study progressed, it could adapt to that part of the dose interval for which response was rapidly ascending. Such a study could then have been followed by one large parallel group study, focusing on the doses showing promise in the exploratory study. Of course, this raises challenges, which I will return to later.
This, and similar examples, showed that in those development programmes in which the initial range of doses was narrow – on average four-fold – dose-response studies needed to be repeated. In contrast, programmes in which the initial range was wider – on average eight-fold – a single dose-response study sufficed.
UNDERSTANDING THE IMPLICATIONS
These results raise a number of issues and challenges. First, we know that there is a relationship between the dose interval studied and the number of doses studied. For example, the data presented in Figure 1 show the dose range and number of doses studies in a random sample of approximately 200 studies recently extracted from http://www.clinicaltrials.gov/ using the search criterion ‘(dose finding OR dose selection OR dose response OR dose ranging) AND Phase 2 AND adults’ between 2002 and 2011. The relationship is clear. Therefore, to increase the dose range will, in general will mean the number of doses will need to increase.
These data also suggest that the designs of the majority of Phase 2b dose response studies involve only two or three doses. That said, there is evidence that over the last 10 years the dose range has been increasing slightly. Figure 2 shows the same data in three time cohorts: 2002 to 2004, 2005 to 2007 and 2008 to 2011, revealing a trend showing an increase in the dose range in the last five or six years.
The second issue is that if we are to use more dose groups, how do we tackle the added cost involved in increasing the number of doses, since this will lead to an increased number of patients and larger studies in general? There are two ways in which this can be tackled. In the first instance, we need to educate clinical researchers that the most efficient way to conduct dose response studies is not the traditional way of comparing each individual dose to placebo, and secondly by prospectively planning to randomise patients to doses which show the appropriate efficacy in an adaptive fashion (4,5).
The third issue is the question of how to persuade pharmaceutical development colleagues to increase the number of doses. In a study I was involved in some time ago we chose to design a study with 15 doses and placebo. This was only made possible because the study drug was delivered by infusion, which allowed us to create multiple doses by dilution of a relatively small number of individual doses (6). The use of intravenous formulation is not a limiting factor in implementing designs with a large number of doses. If treatments are delivered as a tablet, or capsule, many more doses than is normal under current practice could be considered. For example, if dose strengths of 0, 1, 3, 4 units are available, any dose for 0 to 8 units can be studied using only two tablets/capsules: 0 + 0, 0 + 1, 1 + 1, 0 + 3, 0 + 4, 1 + 4, 3 + 3, 3 + 4, 4 + 4.
The fourth issue relates to the complexity of such studies and their cost both in terms of the infrastructure associated with clinical research, and in terms of the cost in their implementation and running. The infrastructure required involves randomisation systems including IVRS and IWRS, and data capture systems (EDC), as well as drug supply management systems, since using the traditional approach of sending large numbers of individual doses to sites will not be feasible with an increased number of dose, because it would lead to a serious wastage of expensive material. The problem is that the investment required may be beyond the majority of individual sponsors.
Figures 3 and 4 whose data are extracted from http://www.clinicaltrials.gov/ show that in the last 10 years the vast majority of sponsors involved in Phase 2b dose-response studies have only been involved in sponsoring a single study. Moreover, fewer have been involved in sponsoring two studies and even fewer three studies, and so on. Interestingly, this graph is an example of a ‘long tail’ distribution first exemplified by Chris Anderson (7). It is only large pharma companies who have been involved in multiple studies. Clearly the large pharma organisations may be able to justify and afford the expenditure in the appropriate infrastructure to support complex, adaptive dose response studies, but it is unlikely that the remainder, largely biotech companies, will be able to do so.
A common mantra in recent years is that adaptive designs will be quicker, faster and cheaper. Seen from the perspective of a single trial, this is unlikely to be the case. Indeed, my experience has been that, if anything, adaptive design in Phase 2b will be larger than an originally planned static design. But the information per resource invested in such designs will be greater and at the development programme level there are likely to be savings, as the likelihood of repeating studies will be reduced.
- Medical Research Council Streptomycin in Tuberculosis Trials Committee, Streptomycin treatment for pulmonary tuberculosis, British Medical Journal 2: pp769-782, 1948
- Sheiner LB, Learning versus confirming in clinical drug development, Clinical Pharmacology and Therapeutics 61: pp275-291, 1997
- Kola I and Landis J, Can the pharmaceutical industry reduce attrition rates? Nature Reviews Drug Discovery 3: pp711-716, 2004
- Bornkamp B, Bretz F, Dmitrienko A, Enas G, Gaydos B, Hsu CH, König F, Krams M, Liu Q, Neuenschwander B, Parke T and Pinheiro J, Innovative approaches for designing and analyzing adaptive dose-ranging trials, Journal of Biopharmaceutical Statistics 17: pp965-995, 2007
- Dragalin V, Bornkamp B, Bretz F, Miller F, Padmanabhan SK, Patel N, Perevozskaya I, Pinheiro J, Smith JR, A simulation study to compare new adaptive dose-ranging designs, Statistics in Biopharmaceutical Research 2(4): pp487-512, 2010
- Krams M, Lees KR, Hacke W, Grieve AP, Orgogozo J-M, Ford, GA, Acute stroke therapy by inhibition of neutrophils (ASTIN): An adaptive dose-response study of UK-279,276 in acute ischaemic stroke, Stroke 34: pp2,543-2,548, 2003
- Anderson C, The long tail: Why the future of business is selling less of more, 2006