Clinical Research - Making adaptive clinical trials mainstream

Making adaptive clinical trials mainstream

Tom Parke looks at industry’s adoption of adaptive clinical trials and highlights where the advances, and operational difficulties, lieTom Parke looks at industry’s adoption of adaptive clinical trials and highlights where the advances, and operational difficulties, lie

The pharmaceutical industry has been running into trouble for several years now as the cost of R&D has soared and the rate of successful introduction of new drugs has fallen. In recognition of this looming crisis the US Food and Drug Administration launched its Strategic Path Initiative to improve the way drugs are discovered, developed and brought to market. One plank in that initiative, and the areas that is getting increasing attention from the industry, is the adoption of adaptive clinical trials.

An adaptive clinical trial is one where the data collected during the trial is monitored and analysed as the trial is running in order to make pre-planned adjustments to the trial as it progresses. With careful planning the monitoring, analysis and adjustments can be made without invalidating the results of the trial. This approach can improve the ethics, the science, and the efficiency of both the trial and the drug development process.

Several types of adaptive trial are already in use. For example, Phase I safety trials have always been adaptive, monitoring the early subjects on low doses to see if it is safe to proceed to higher doses. Sometimes trials have had an ‘ad hoc’ review and analysis of the data at some interim point in order to see if any mid-trial adjustment is necessary. And ‘minimisation’ has been used to ensure balanced randomisation occurs within sub-populations of the subjects being treated in the trial, as well as across the subject population as a whole. This is of particular concern in trials of rare diseases where balanced randomisation cannot be assured through sheer numbers.

In the 1970s, Group Sequential Designs were introduced that allowed pre-planned interim looks at the data to determine whether a trial could be stopped early – either for efficacy or futility. And in 1990, an improved Phase I design – the Continuous Re-assessment Method (CRM) – was introduced that gave better control of the amount of toxicity observed during the trial, was flexible in the target level of maximum tolerated toxicity, and gave a more reliable estimate of the maximum tolerated dose. This has given rise to a whole family of CRM designs.

In addition, the late 1990s heralded two landmark developments: the introduction of a completely general frequentist statistical method for controlling the ‘type-1 error’ in trials where the data has been looked at and an adaptation has been made – confusingly some people mean precisely this sort of trial when they refer to an adaptive clinical trial – and secondly, the design of the first Phase II adaptive dose finding trial using Bayesian statistics.

In Phase II Bayesian adaptive dose finding designs a Bayesian statistical model is used to analyse the data as it is collected. This allows estimates of how the trial is progressing to be used to modify the trial as it goes along. The information gathered in the remainder of the trial is therefore maximised, for instance by dropping unsuccessful treatments or biasing the randomisation towards the dose with the best efficacy/safety profile. The Bayesian paradigm seems to be the best suited for constructing robust, complex and novel designs that take into account multiple doses, multiple endpoints, biomarkers and so on, although understandable caution from regulatory agencies means these are currently confined mainly to pre-confirmatory trials.

The advances in frequentist analysis have allowed trial designers to come up with seamless trial designs, where a Phase II trial transitions into a Phase III trial without pause, saving considerable drug development time. Two sorts of seamless design are possible: operationally seamless, which simply seeks to exploit the saving in time; and inferentially seamless, which uses the new statistical methods to benefit from combining the relevant data from the Phase II part with the Phase III data in the final analysis. This latter method yields greater power in the overall design (or the option to have a smaller Phase III). For the moment, seamless Phase II-Phase III design will probably be limited to well understood indications (low risk of the Phase III trial design having to be changed in light of the Phase II data) or indications where sufficient subjects are difficult to recruit, such as in follow-on paediatric studies.

Attractive properties

The new wave of adaptive clinical trials have sprung from these developments, driven by the growing difficulty of developing new drugs, and enabled by the increasingly early availability of clinical data resulting from electronic data capture. Not surprisingly, these new adaptive designs have many attractive properties.

Most compounds in drug development eventually fail, but these designs allow earlier detection and the killing of a failed compound.

Moreover, trial subjects are used more efficiently, and fewer are given ineffective compounds, ineffective doses, or doses that are unnecessarily high. At the same time, pharmaceutical companies waste less on unsuccessful compounds and can more quickly re-assign resources to alternative drugs within their pipelines.

The designs also allow more doses to be tested in Phases I and II. The modelling of the dose-response and dose-toxicity, and the optimisation of the allocation of subjects to the most useful doses, results in a better understanding of the effect of the compound on patients in the doses that are clinically relevant. This leads to better decisions concerning the drug’s development and, for successful compounds, a better design of the Phase III trial, thereby helping to reduce the failure of a drug at this stage, or worse, later in its development.

New applications

These richer trial designs could also be used to combine other traditionally separate phases. A seamless Phase I and Phase IIa trial would allow efficacy and toxicity to be studied at the outset, with the safety study in patients not healthy volunteers. Moreover, seamless Phase IIa and IIb trials would either allow the trial to be stopped early if the compound is ineffective, or allow it to be continued with additional dose arms.

Trial designers have only just started looking at the ways adaptation and modelling can be used to improve how clinical trials are performed. But a number of exciting ideas are circulating. These include: the use of biomarkers and measures of drug exposure in the modelling to yield a more accurate picture of a compound’s behaviour; or the consideration of optimisation issues other than just dose, such as optimising between different compounds, optimising the patient population, or optimising how the new compound is combined with existing therapies.

Equally encouraging is that initial concerns about the statistical validity and regulatory acceptance of adaptive trials are dwindling. There is continuing discussion about which approaches are best, but the published methods have now had considerable public review. And while the regulators are still cautious about innovation in Phase III they are largely supportive, and even encouraging, of innovation in early phases.

Operational difficulties

However, there are still operational difficulties in deploying these new trial designs as they run counter to many established procedures and objectives in clinical trial operations. Changes to accommodate adaptive clinical trials have to be made across a range of functions, in particular trial design, data capture, randomisation, trial monitoring and trial supply.

For the first attempts at adaptive trials, an organisation can minimise these problems by careful selection of the development projects that use adaptive designs and outsourcing. However in the long term, an organisation needs to embrace the need for change from the top down. Teams will need motivation, training and support just as for any other carefully implemented business process change.

Adaptive designs are more complex and a central team is required to help select the trials that will most benefit from being adaptive, form a repository of knowledge on adaptive designs, and advise on the tools, infrastructure and processes changes required to get the most from them.

Clinical operations will need to balance fast patient recruitment with allowing adaptive designs time to adapt. Perhaps they will use fewer centres to decrease the variability in the measurement of subjects’ responses, meaning fewer subjects are required and easing the problems of supply that adaptive designs can create.

To be able to adapt a trial it is necessary to have swift and reliable data collection. Centres need to be encouraged to enter and correct data quickly and the data cleaning process needs to be streamlined so the data are available early on. This will have the beneficial side-effect of improving the data quality overall, but to achieve it clinical teams may need to be more selective in the data they collect. There is currently a tendency to err on the safe side and collect more than is necessary, adding costs and delays in data entry and cleaning.

To implement the adaptations, the data monitoring committee (DMC) needs to meet regularly and quickly. It may need to review high-level interim reports every two weeks and either approve adjustments in the randomisation or schedule a more detailed interim, not at some pre-specified point in the trial but when there is sufficient data to make a decision. The more frequent unblinding of the DMC, the use of more in-house staff on DMCs, and more people being made aware of adaptive decisions means firewalls to maintain the blind of the study team need to be strengthened.

For some types of adaptive trial, either automatically or once the DMC has approved an adaptation, it needs to be possible to adjust the randomisation quickly, by replacing the unused portion of the current master randomisation list with a new one.

Adaptive trials will also make drug supply during the trial harder. The trial may use more doses or dose combinations, the ratio of allocation to the different treatment arms changes and the size and duration of the trial is not known ahead of time. These difficulties can be ameliorated by simulation of the trial beforehand to estimate supply needs accurately, regular reporting (possibly unblinded) from the DMC to the supply team, and having an adaptive re-supply regime. However in the end it may still be necessary to accept there will be additional supply costs but that they are more than offset by the other savings and gains of adaptive trials.

Finally, adaptive trial designs are a major new tool for drug development, redefining how drug development is performed. Their full-scale adoption will require change across all departments involved in running clinical trials. For some companies that process of change has already started.

Tom Parke is the Head of Clinical Trials Solutions at Tessella