Design of clinical trials with failure-time endpoints and interim analyses: An update after fifteen years

Abstract

Time to event is the clinically definitive endpoint in Phase III trials of new treatments of cancer, cardiovascular and many other diseases. Because these trials involve relatively long follow-up, their protocols usually incorporate periodic interim analyses of the data by a Data and Safety Monitoring Board/Committee. This paper gives a review of the major developments in the design of these trials in the 21st century, spurred by the need for better clinical trial designs to cope with the remarkable advances in cancer biology, genomics and imaging that can help predict patients' sensitivity or resistance to certain treatments. In addition to this overview and discussion of related issues and challenges, we also introduce a new approach to address some of these issues.

Adaptive design of confirmatory trials: Advances and challenges

“…Hybrid resampling is especially suited for primary and secondary analysis, in a regulatory environment, of complex data in adaptive clinical trials. A detailed discussion is given in [66] for the case of censored survival data, the complexity of which has led to inefficient adaptive designs, as reviewed in the second paragraph of Section 2.2. A new approach is also developed in [66] that can adapt the choice of the test statistics to the observed survival pattern.…”

Innovative designs of point-of-care comparative effectiveness trials

“…Hence a more convincing and robust analysis is to use the logrank statistic that is the efficient score statistic of the proportional hazards model, in which random effects can also be readily incorporated [36]. As explained in [6] and [37], the Bayesian approach does not have type I error guarantees even after “frequentist twists” of the type used in [25] and [38]. Section 4.1 of [6] points out that mainstream statistical methods have well-established theories, efficiency properties, and implementation details/software and that many of the advanced techniques already provide powerful tools for developing efficient and flexible adaptive designs that also have valid type I errors or confidence levels.…”

“…Section 4.1 of [6] points out that mainstream statistical methods have well-established theories, efficiency properties, and implementation details/software and that many of the advanced techniques already provide powerful tools for developing efficient and flexible adaptive designs that also have valid type I errors or confidence levels. Although Bayesian methods are part of mainstream statistics, “so are parametric, semiparametric, and nonparametric (empirical) likelihood methods.” The more complex case of time-to-event endpoints is discussed in [37] which also reviews semiparametric/nonparametric and Bayesian (MCMC-based) survival analysis in mainstream statistics. The data-driven adaptive designs of POC-CER trials that we propose to get around the prohibitively large sample sizes under standard trial designs can use these advances in statistical methodologies together with some refinements and modifications, as shown in Section 4.2 and in the closely related paper [39].…”

Health Analytics, Economics and Medicine toward a 21<sup>st</sup> Century Health Care System

“…Despite the sequential nature of Phase I-III trials, trials are often planned separately, treating each trial as an in-dependent study whose design depends on results from studies in previous phases. The need for innovative study designs to better integrate the phases is now widely recognized and major advances, particularly in the area of adaptive designs, were made in the past decade; see [5]- [14]. In particular, one of the major hurdles in bringing the new genomic-guided and risk-adapted personalized treatments, which are devised to attack specific targets in individual patients, to the market is the exorbitant cost and time needed to conduct confirmatory Phase III trials using standard randomized clinical trial designs for drug approval in industry.…”