Sample Size Re-Estimation as an Adaptive Design

Laurence Collette;

Sample Size Re-Estimation as an Adaptive Design

September 8, 2021

Laurence Collette

Publication

Article

Applied Clinical TrialsApplied Clinical Trials-09-01-2021

Volume 30

Issue 9

Using the promising-zone approach can salvage an underpowered trial.

An emerging trend for developers looking for ways to increase the likelihood of success and their return on investment is adaptive trial designs. Defined by FDA as a “clinical design that allows for prospectively planned modifications to one or more aspects…based on accumulating data,”¹ an adaptive trial can help “de-risk” the arduous process of drug development.

According to a paper in the Journal of the National Cancer Institute in 2017, the approach is motivated by “both ethical and resource considerations.” An integral feature of an adaptive trial is the existence of an independent data-monitoring committee (IDMC) or alternatively of a separate dedicated independent adaptation committee, to carry out interim review of accruing outcome data that allow pre-specified adaptations to be made while the trial is ongoing.

There are various types of adaptive trials and of adaptations that can be proposed, each with their risks and benefits. Sample-size re-estimation (SSRE) is a family of designs that allow for increasing the sample size based on blinded or unblinded interim data analysis of randomized trials. This is important because research has shown that underpowered clinical trials, when the sample size is too small to reject the null hypothesis, is one of the common reasons a potentially efficacious drug can fail to demonstrate its worth during a study.²

Sample-size calculations performed before the start of a trial require assumptions about the hypothesized treatment effect and the outcomes in each arm. Incorrect assumptions can result in a trial being underpowered or even overpowered. Both situations pose ethical and scientific concerns: either recruiting an insufficient number of patients without being able to fully answer the research question or subjecting more patients than necessary to a potentially inferior treatment when there’s already sufficient evidence to answer the research question.³ Unblinded SSRE methods assess the validity of the assumptions and enable an increase of the sample size based on the observed interim treatment effect, if necessary. Unblinded SSRE methods inherit the group sequential capacity to stop the trial early as soon as the scientific question addressed by the trial is answered.

The promising-zone approach

The promising-zone SSRE method was first described by Mehta and Pocock in 2011, based on earlier work by Chen et al.⁴ It allows to recalculate the required sample size based on interim data, while minimizing the inevitable risks of bias and type-I error and yet, use a conventional test statistic at all stages of the study.

At the last interim analysis for efficacy, the IDMC will use the observed data and the assumption that future treatment effect will follow the same trend to re-calculate the conditional power. This value will fall into one of three pre-defined zones:

Unfavorable: The observed treatment effect is not sufficient to warrant the increase in sample size needed to maintain study power.

Promising: The observed treatment is effect is lower than expected, but assuming the observed trend exactly reflects the true effect and that the trend continues, the power can be recovered, by increasing the sample size.

Favorable: The observed treatment effect is sufficiently favorable, and no sample-size increase is needed to maintain power.

It is a relatively simple approach to SSRE as it allows using a conventional test statistic at the end of the study, and, in some cases, can be very beneficial. However, it is worth noting that the promising-zone approach is not always the most optimal. It can, as all SSRE methods, introduce the risk of operational bias, and it creates additional regulatory hurdles. Moreover, the promising zone is defined on a statistical basis to enable the use a conventional test statistic, and not on clinical basis, and has been criticized for that.

SSRE in practice

When used appropriately, this approach can confer several benefits, not least of which is the ability to salvage a slightly underpowered trial, by adjusting the sample size when interim results show a treatment effect that is somewhat smaller than anticipated.

Unlike other SSRE methods, it allows the use of a standard test statistic without type-I error inflation. The promising-zone method is particularly useful for small organizations with limited upfront resources. Starting with the smallest sample size and reasonably optimistic assumptions regarding the treatment effect, they can then attract additional investment to expand the study based on promising interim data.

However, it is not a one-size-fits-all solution. At IDDI, we recommend sponsors to compare the risks and benefits of different forms of adaptive designs, including the more classical group-sequential designs, and select the one that best suits their individual circumstances and likely total sample size. In many cases a group-sequential design can be more advantageous than a trial with unblinded sample-size adaptations. Sponsors must also keep in mind that sample size re-estimation will be inefficient if the initial study was largely underpowered.

Operational challenges

When unblinded SSRE is the right approach, sponsors and CROs must ensure they consider and respond to some added challenges.

First, the method can increase the risk of operational bias if the announcement of the SSRE conveys the anticipation of an effective treatment. To minimize that risk, it’s important to tell sites to continue recruitment until asked to stop, without announcing the magnitude of the sample-size increase. Moreover, knowledge of that increase can allow “back-calculation” of the treatment effect, which undermines the blinded nature of the interim analysis outside the circle of the IDMC.

The method also comes with increased regulatory hurdles. These are similar for FDA and EMA,⁵ and we expect similar requirements to be included in the upcoming ICH E20.⁶ To comply, teams need to:

demonstrate, through extensive simulations, that they have controlled for type-I error and the chance of erroneous conclusions.
show that the methods used can correct for the potential bias introduced by the adaptation and ensure an unbiased treatment-effect estimation.
complete a pre-specification of the plan that includes the timing of the interim analysis, and stipulates the SSRE method used (i.e., the promising-zone rule), and test statistics.
ensure unplanned changes are not influenced by knowledge of the data, and ensure the homogeneity of effects before and after adaptation.

Conclusion

SSRE is a relatively new concept, but, thanks to its simplicity, SSRE using the promising zone is one that has grown in popularity in recent years. It comes with a host of benefits: it can increase study power, reduce upfront investment, and has the potential to shorten trial duration, for example. But it is important to note that it is not suited to
all situations.

Despite its appeal, the promising-zone method, like other SSRE methods, comes with additional logistical and regulatory hurdles, and organizations must consider how each of these relate to their own individual development program. Ultimately, organizations should choose the design that best suits the needs of their study, because there is no panacea in the world of adaptive clinical trials.