HSM: Estimation of intervention effects for adaptive enrichment design RCTs that incorporate identification of predictive biomarkers
Lead Research Organisation:
University of Warwick
Department Name: Warwick Medical School
Abstract
Advances in medicine have led to the acknowledgement that differences in patient characteristics may lead to differences in how patients respond to therapies such as drugs. Patient characteristics that have had much medical research interest recently are those based on the genetic make-up. For example, patients may be categorised as those who express a certain genetic make-up and those who do not. The term generally used is biomarker, with the patients who express a certain genetic make-up referred as being biomarker positive and those who do not referred as biomarker negative. The term is usually not restricted to genetics and so could be used, for example, to categorise patients into those below and those above a certain age.
This project is concerned with settings where the biomarker may interact with a therapy, that is, patients' responses to a therapy depends on whether they are biomarker positive or negative. If a patient's response depends on whether he/she is biomarker positive or negative, the biomarker is said to be a predictive biomarker.
Stratified medicine is the branch of medicine where testing of new therapies allows for the possibility of differences in therapy effects in different subpopulations defined by biomarker status. One strategy is to have two separate trials. The first trial consisting of patients from the full population and its aim is to use a predictive biomarker to select the subpopulation that will benefit from the new therapy. The second trial recruits patients from the selected subpopulation and its aim is to use the data collected to get a definitive estimate of the size of the effect (benefit) of the new therapy in this subpopulation.
An efficient strategy is to have a single trial that includes an interim analysis partway through the trial to select the subpopulation that benefits from the new therapy. After the interim analysis, more patients are recruited from the selected subpopulation and their data, together with data used in the interim analysis, are used is to get a definitive estimate of the size of the effect of the new therapy in the selected subpopulation. Such designs are commonly referred to as adaptive enrichment designs. They are efficient because fewer patients would be required to test a new therapy.
Estimating the size of the effect of the new therapy when an adaptive enrichment design is used needs to adjust for the fact that the interim data used to select the subpopulation are also used in estimating the effect. If this is not done and the standard methods are used, the effect will be overestimated because the larger effect observed in the selected subpopulation using the interim analysis data may have occurred by chance. This is undesirable because definitive estimates of the size of the effect of new therapies are used by many stakeholders to make a decision on whether to adopt a therapy.
Only one appropriate method for estimating effects of therapies, and which is for a specific design, has been developed for adaptive enrichment designs. This is one of the barriers of using these designs. The aim of this project is to remove this barrier by developing new methods for estimating size of effects while adjusting for subpopulation selection made using a predictive biomarker.
The difference between the existing method and this work is that we will consider different forms of biomarkers, such as having more than one biomarker, and we will also consider a common type of patient outcome data in stratified medicine: time to event data such as overall survival for patients.
This work will increase the use of adaptive enrichment designs. Consequently, this will save resources while testing new therapies and lead to more rapid development of safe effective treatments.
This project is concerned with settings where the biomarker may interact with a therapy, that is, patients' responses to a therapy depends on whether they are biomarker positive or negative. If a patient's response depends on whether he/she is biomarker positive or negative, the biomarker is said to be a predictive biomarker.
Stratified medicine is the branch of medicine where testing of new therapies allows for the possibility of differences in therapy effects in different subpopulations defined by biomarker status. One strategy is to have two separate trials. The first trial consisting of patients from the full population and its aim is to use a predictive biomarker to select the subpopulation that will benefit from the new therapy. The second trial recruits patients from the selected subpopulation and its aim is to use the data collected to get a definitive estimate of the size of the effect (benefit) of the new therapy in this subpopulation.
An efficient strategy is to have a single trial that includes an interim analysis partway through the trial to select the subpopulation that benefits from the new therapy. After the interim analysis, more patients are recruited from the selected subpopulation and their data, together with data used in the interim analysis, are used is to get a definitive estimate of the size of the effect of the new therapy in the selected subpopulation. Such designs are commonly referred to as adaptive enrichment designs. They are efficient because fewer patients would be required to test a new therapy.
Estimating the size of the effect of the new therapy when an adaptive enrichment design is used needs to adjust for the fact that the interim data used to select the subpopulation are also used in estimating the effect. If this is not done and the standard methods are used, the effect will be overestimated because the larger effect observed in the selected subpopulation using the interim analysis data may have occurred by chance. This is undesirable because definitive estimates of the size of the effect of new therapies are used by many stakeholders to make a decision on whether to adopt a therapy.
Only one appropriate method for estimating effects of therapies, and which is for a specific design, has been developed for adaptive enrichment designs. This is one of the barriers of using these designs. The aim of this project is to remove this barrier by developing new methods for estimating size of effects while adjusting for subpopulation selection made using a predictive biomarker.
The difference between the existing method and this work is that we will consider different forms of biomarkers, such as having more than one biomarker, and we will also consider a common type of patient outcome data in stratified medicine: time to event data such as overall survival for patients.
This work will increase the use of adaptive enrichment designs. Consequently, this will save resources while testing new therapies and lead to more rapid development of safe effective treatments.
Technical Summary
The aim of this work is to develop point and interval estimators following randomised clinical trials (RCTs) that are conducted using adaptive enrichment designs in the context of stratified medicine.
In stratified medicine, it is acknowledged that patients with different characteristics such as different genetic biomarkers, may respond to treatments in different ways so that an intervention may only benefit a subpopulation of the full patient population. Testing of new interventions allow for this possibility.
We will consider adaptive enrichment designs where a RCT includes an interim analysis. Using a predictive biomarker, that is a biomarker that determines how a patient responds to a treatment, the interim analysis is used to select the subpopulation that benefits from the new therapy. After the interim analysis, patients are recruited from the selected subpopulation. Definitive analysis includes data collected before and after interim analysis.
Interim analysis data induce selection bias and so point and interval estimators need to adjust for this. Very little work exists that adjust for selection bias and so more research is required. We will develop point estimators for three settings: (1) Patient data are normally distributed and the biomarker is continuous with several candidate threshold values, (2) Same setting as above but for time to event data, and (3) A panel of biomarkers is used to make selection. Data collected after the interim analysis are unbiased and we will use them to get unbiased estimators, which we then Rao-Blackwellise to obtain unbiased estimators that have smaller variances.
We will also develop two interval estimators for the first setting; one based on p-value inversion and the second based on the density of the statistic in the selected subpopulation.
By enabling unbiased point and interval estimation, we will increase the use of these efficient adaptive enrichment designs.
In stratified medicine, it is acknowledged that patients with different characteristics such as different genetic biomarkers, may respond to treatments in different ways so that an intervention may only benefit a subpopulation of the full patient population. Testing of new interventions allow for this possibility.
We will consider adaptive enrichment designs where a RCT includes an interim analysis. Using a predictive biomarker, that is a biomarker that determines how a patient responds to a treatment, the interim analysis is used to select the subpopulation that benefits from the new therapy. After the interim analysis, patients are recruited from the selected subpopulation. Definitive analysis includes data collected before and after interim analysis.
Interim analysis data induce selection bias and so point and interval estimators need to adjust for this. Very little work exists that adjust for selection bias and so more research is required. We will develop point estimators for three settings: (1) Patient data are normally distributed and the biomarker is continuous with several candidate threshold values, (2) Same setting as above but for time to event data, and (3) A panel of biomarkers is used to make selection. Data collected after the interim analysis are unbiased and we will use them to get unbiased estimators, which we then Rao-Blackwellise to obtain unbiased estimators that have smaller variances.
We will also develop two interval estimators for the first setting; one based on p-value inversion and the second based on the density of the statistic in the selected subpopulation.
By enabling unbiased point and interval estimation, we will increase the use of these efficient adaptive enrichment designs.
Planned Impact
Who will benefit from this research?
This research will benefit UK and global regulatory bodies and individuals who want to use estimates from clinical trials to make decisions on whether to adopt or not adopt therapies. The research will also benefit any institution that conducts and analyses clinical trials such as the pharmaceutical industry and public sector research institutions. Finally, the ultimate aim of this research work is intended to benefit patients and the general public.
How will they benefit from this research?
The aim of this research is to develop methods for obtaining unbiased estimates following clinical trials conducted using adaptive enrichment designs in the context of stratified medicine. Currently, appropriate methods for hypothesis testing are available but methods for obtaining unbiased estimates are lacking. As well as well as hypothesis testing, unbiased estimates are important to decision makers as they quantify the size of effect for therapies. This research will fill the current gap associated with lack of methods for obtaining unbiased estimates. This will enable regulatory bodies and individuals who want to make decisions to have better understanding of the benefits of therapies tested using adaptive enrichment designs.
Pharmaceutical industry and public sector research institutions will benefit because they will be able to obtain unbiased estimates and confidence intervals with right coverage probabilities when they use adaptive enrichment designs to conduct their clinical trials. This will increase uptake of adaptive enrichment designs. Consequently, it will become cheaper for institutions to conduct clinical trials.
The patients will benefit because adoption of adaptive enrichment designs means that effective and safe therapies will become available more quickly. The general public will also benefit because clinical trials that are funded through taxes will become cheaper. The money saved may be used to conduct more trials or to provide other services.
This research will benefit UK and global regulatory bodies and individuals who want to use estimates from clinical trials to make decisions on whether to adopt or not adopt therapies. The research will also benefit any institution that conducts and analyses clinical trials such as the pharmaceutical industry and public sector research institutions. Finally, the ultimate aim of this research work is intended to benefit patients and the general public.
How will they benefit from this research?
The aim of this research is to develop methods for obtaining unbiased estimates following clinical trials conducted using adaptive enrichment designs in the context of stratified medicine. Currently, appropriate methods for hypothesis testing are available but methods for obtaining unbiased estimates are lacking. As well as well as hypothesis testing, unbiased estimates are important to decision makers as they quantify the size of effect for therapies. This research will fill the current gap associated with lack of methods for obtaining unbiased estimates. This will enable regulatory bodies and individuals who want to make decisions to have better understanding of the benefits of therapies tested using adaptive enrichment designs.
Pharmaceutical industry and public sector research institutions will benefit because they will be able to obtain unbiased estimates and confidence intervals with right coverage probabilities when they use adaptive enrichment designs to conduct their clinical trials. This will increase uptake of adaptive enrichment designs. Consequently, it will become cheaper for institutions to conduct clinical trials.
The patients will benefit because adoption of adaptive enrichment designs means that effective and safe therapies will become available more quickly. The general public will also benefit because clinical trials that are funded through taxes will become cheaper. The money saved may be used to conduct more trials or to provide other services.
Publications
Stallard N
(2018)
Uniformly minimum variance conditionally unbiased estimation in multi-arm multi-stage clinical trials
in Biometrika
Stallard N
(2019)
On the need to adjust for multiplicity in confirmatory clinical trials with master protocols.
in Annals of oncology : official journal of the European Society for Medical Oncology
Kimani PK
(2020)
Point and interval estimation in two-stage adaptive designs with time to event data and biomarker-driven subpopulation selection.
in Statistics in medicine
Kimani PK
(2018)
Point estimation following two-stage adaptive threshold enrichment clinical trials.
in Statistics in medicine
| Title | New statistical methods in three methodology papers |
| Description | The award was to develop new statistical methods to analyse efficient clinical trials with interim analysis to adapt the trial partway through the trial. Using the funding we published four medical statistics methodology papers with three corresponding to three new techniques to analyse data for three different clinical trial designs. The fourth provided a proposal of how to handle statistical multiplicity for different forms of efficient master protocols. |
| Type Of Material | Data analysis technique |
| Year Produced | 2018 |
| Provided To Others? | Yes |
| Impact | One of the statistical techniques we have developed has been used to compute point estimates in the primary analysis of a randomised trial published in a peer reviewed journal. Our papers have been cited by several publications as a reference of how analysis should be performed. |
| Description | Collaboration with Dr Lindsay Renfro from University of Southern California in the US |
| Organisation | University of Southern California |
| Department | Keck School of Medicine |
| Country | United States |
| Sector | Academic/University |
| PI Contribution | This collaboration was proposed in the grant application, when Dr Renfro was working at Mayo clinic. The collaboration has been successfully established. Our contribution to the collaboration is several years expertise in developing statistical techniques that can be used to analyse trials that include selection of patients based on biomarkers. Dr Renfro also has a position at Children's Oncology Group. Through this, we brought in another collaborator from Children's Oncology Group based at the university of Florida who identified real trial data from their research that we have reanalysed to demonstrate how to use new statistical methods that we have developed with real data in a paper that was published in Statistics in Medicine journal. |
| Collaborator Contribution | The partner provided input into the design that we considered while developing new statistical methods in a published paper. The partner uses the designs and patient outcomes that we assumed for new statistical methods that were published in two of our papers. The partner also contributed to practical aspects in developing those statistical methods. Also, as described above, the partner provided real clinical trial data that we reanalysed to demonstrate the work in a paper that was published in a peer reviewed journal. |
| Impact | (1) I had a research visit in October 2017 at Mayo clinic where Dr Renfro was based then and I presented methods developed in work package 1. (2) Research partner provided input in terms of statistical expertise and application of methods in clinical trials on a paper that has been published. (3) I visited Dr Renfro at Children's Oncology Group in August 2019. During the visit I worked with Dr Renfro to reanalyse some data from the institution which we used to show a statistical method we have developed can be used in a real clinical trial. The paper that includes this work was published in the Statistics in Medicine Journal. (4) We have discussed collaboration on further grant applications to USA and UK funding bodies. (5) We submitted one grant that was unsuccessful and we intend to submit more grants. |
| Start Year | 2016 |