Using Surrogate Endpoints for Decision-Making in Adaptive Seamless Designs
Lead Research Organisation:
University of Warwick
Department Name: Warwick Medical School
Abstract
Clinical trials are used to evaluate the safety and effectiveness of new drugs and other novel therapies prior to general use. Conventionally, a major part of this evaluation is a trial to compare one new drug (at a single prespecified dose/formulation) with an existing standard or control treatment. Statistical methods have been developed over many years to allow valid interpretation of the data from such trials.
A recent innovation in clinical trial design has been the development of new approaches in which several new treatments are compared simultaneously with the control in a single clinical trial, with the less effective treatments dropped from the trial on the basis of early results. Such methods, called adaptive seamless designs, have the potential to reduce the number of patients required. A particular challenge arises in chronic diseases when the data indicating treatment effectiveness may be observed only after long-term follow-up. For example, in Multiple Sclerosis the primary outcome considered might be a change in health status after three years of treatment. Often some other earlier outcome can be observed that it is believed is predictive of the primary long-term outcome. Such an outcome is called a surrogate endpoint. In the Multiple Sclerosis case, for example, imaging of the brain can give an early indication of disease progression. In this case it would seem sensible to use this surrogate endpoint information in addition to any available primary endpoint data to decide which treatments should be dropped from the trial. The development of statical methods that allow the use of surrogate endpoints for treatment selection in adaptive seamless designs has been an area of active research in the last few years. The difficulty comes when trying to ensure that the final comparison of the selected treatment (or treatments) is a fair one, since they could have been selected because they appeared superior to other treatments by chance.
The aim of this project is to develop a new statistical method for using a surrogate endpoint in an adaptive seamless design. By making the most efficient use of the surrogate endpoint information, this new approach will lead to further reduction in the number of patients required in the trial. This will reduce the number of patients exposed to less effective treatments and ensure that effective new drugs are made available for general use, on the basis of thorough evaluation, as quickly as possible.
A recent innovation in clinical trial design has been the development of new approaches in which several new treatments are compared simultaneously with the control in a single clinical trial, with the less effective treatments dropped from the trial on the basis of early results. Such methods, called adaptive seamless designs, have the potential to reduce the number of patients required. A particular challenge arises in chronic diseases when the data indicating treatment effectiveness may be observed only after long-term follow-up. For example, in Multiple Sclerosis the primary outcome considered might be a change in health status after three years of treatment. Often some other earlier outcome can be observed that it is believed is predictive of the primary long-term outcome. Such an outcome is called a surrogate endpoint. In the Multiple Sclerosis case, for example, imaging of the brain can give an early indication of disease progression. In this case it would seem sensible to use this surrogate endpoint information in addition to any available primary endpoint data to decide which treatments should be dropped from the trial. The development of statical methods that allow the use of surrogate endpoints for treatment selection in adaptive seamless designs has been an area of active research in the last few years. The difficulty comes when trying to ensure that the final comparison of the selected treatment (or treatments) is a fair one, since they could have been selected because they appeared superior to other treatments by chance.
The aim of this project is to develop a new statistical method for using a surrogate endpoint in an adaptive seamless design. By making the most efficient use of the surrogate endpoint information, this new approach will lead to further reduction in the number of patients required in the trial. This will reduce the number of patients exposed to less effective treatments and ensure that effective new drugs are made available for general use, on the basis of thorough evaluation, as quickly as possible.
Technical Summary
A recent innovation in confirmatory clinical trial design is the development of adaptive seamless designs (ASDs) that combine phases II and III in a single trial. In such designs, patients are initially randomised between a control and a number of experimental treatments. Part-way through the trial, based on the results of an interim analysis, more promising treatments are selected to continue with the control, with other treatments dropped from the trial. At the end of the trial selected treatments and the control are compared using all data from these groups. The use of a common control group and the early dropping of ineffective treatments in the ASD leads to considerable gain in power, or saving in number of patients required, relative to the conventional approach of separate phase II and phase III trials. The inclusion of data used for treatment selection in the final analysis leads, however, to inflation of the type I error rate if a conventional analysis is conducted and special statistical methods have been developed to avoid this.
The use of ASDs is challenging when the primary endpoint is observed only after long-term follow-up, since recruitment may have closed before observation of sufficient primary endpoint data for the interim analysis. The savings associated with dropping treatments may thus be lost. Often in such a setting, an endpoint that is predictive of the primary response, possibly a validated surrogate, is observed more rapidly. It is therefore appealing to use these data, along with available primary endpoint data, at the interim analysis for treatment selection. Although this leads to additional statistical challenges, we have previously developed two methods that enable such an approach.
The two methods proposed both lead to increased power from the use of the surrogate endpoint data. They have been shown to have very different properties, however, with the gain in power arising respectively primarily from individual-level and trial-level surrogacy. This suggests that a new method superior to either of these approaches could based on estimation of the relationship between the surrogate and primary endpoints from the interim analysis data. The aim of this project is a method that utilises the surrogate endpoint data as efficiently as possible. This will reduce the number of patients required, particularly those on ineffective treatments, and also accelerate the drug development process.
The use of ASDs is challenging when the primary endpoint is observed only after long-term follow-up, since recruitment may have closed before observation of sufficient primary endpoint data for the interim analysis. The savings associated with dropping treatments may thus be lost. Often in such a setting, an endpoint that is predictive of the primary response, possibly a validated surrogate, is observed more rapidly. It is therefore appealing to use these data, along with available primary endpoint data, at the interim analysis for treatment selection. Although this leads to additional statistical challenges, we have previously developed two methods that enable such an approach.
The two methods proposed both lead to increased power from the use of the surrogate endpoint data. They have been shown to have very different properties, however, with the gain in power arising respectively primarily from individual-level and trial-level surrogacy. This suggests that a new method superior to either of these approaches could based on estimation of the relationship between the surrogate and primary endpoints from the interim analysis data. The aim of this project is a method that utilises the surrogate endpoint data as efficiently as possible. This will reduce the number of patients required, particularly those on ineffective treatments, and also accelerate the drug development process.
