Xiaofei Wang and Jianwen Cai
In the era of personalized medicine, an increasing number of cancer drugs are being developed to target specific genetic variations among patients. Such cancer drugs are often referred to as target agents. Unlike traditional standard chemotherapy, the clinical benefit of a target agent is expected to depend on a putative biomarker, which measures the status of or the extent of the genetic variation. Biomarker-integrated randomized clinical trials (RCTs) are essential tools for evaluating overall drug effect, drug effects of specific patient groups with different biomarker status, as well as the utility of the putative biomarker for treatment selection.
Often, the genetic variation that defines a therapeutic target occurs only in a small proportion of the underlying patient population. If the drug is expected not to work as effectively in biomarker-negative patients as in biomarker-positive patients, it would be costly or even unethical to recruit and treat many biomarker-negative patients. The central question is how to design a biomarker-integrated RCT that achieves study objectives with reduced cost, improved efficiency, and high ethical standards.
Patient enrichment has been used to achieve these goals and to maximize trial efficiency in biomarker-integrated RCTs. Patient enrichment can be generally defined as a design feature or strategy that funnels patients who meet eligibility criteria into different treatment arms or trial cohorts. Choosing an effective strategy for patient enrichment depends on the answers to several questions: (1) What are the primary and the secondary objectives of the trial?; (2) Is the targeted agent expected to benefit both biomarker-positive and biomarker-negative patients?; and (3) What is the accuracy of the biomarker assay in identifying truly biomarker-positive patients? If a drug is expected to only benefit the biomarker-positive patients and the putative biomarker has compelling credential in accurately classifying patients, a biomarker-positive-only design, which is often referred to as an enrichment or target design, would be a sensible approach for enriching the patient cohort. Biomarker-positive-only design can achieve better trial efficiency in terms of a smaller sample size of treated patients, but it comes at the expense of not being able to make inference for the overall drug effect among all patients or the drug effect among biomarker-negative patients.
Source: National Cancer Institute
When the putative biomarker has no compelling credential in classifying patients with high accuracy or the drug has an off-target effect on biomarker-negative patients, an inclusion of biomarker-negative patients in the trial cohort allows one to evaluate both the drug effect among biomarker positives and the drug effect among biomarker negatives. The Biomarker-stratified design or all-comer design is one such design that recruits and treats all patients regardless of the status of the putative biomarker. One significant advantage of this design is that it enrolls all four patient subgroups, i.e., positive/experimental, positive/control, negative/experimental, negative/control. This design makes statistical inference on a broad range of clinical questions possible, including the drug effects of specific biomarker groups, the interaction between drug and biomarker, the overall drug effect, and the clinical benefit of utilizing the biomarker for treatment selection. When the biomarker-positive patients represent a relatively small proportion of the patient population, a large number of biomarker-negative patients will be enrolled in order to recruit the required number of biomarker-positive patients. To improve trial efficiency and to reduce unnecessary monetary and human cost, a variety of patient enrichment strategies could be used to oversample (i.e., enrich) biomarker positives and undersample biomarker negatives. For example, while recruiting a fixed number of biomarker-positive patients, one can enroll selectively a comparable number of biomarker-negative patients. This design will achieve comparable statistical power for testing the drug effect among negatives or the overall drug effect among all patients. In a slightly different scenario, the cost of biomarker assays is expensive and screening a large number of patients to identify sufficient positive patients poses a roadblock for conducting a biomarker-stratified trial. If there exists a low-cost surrogate biomarker of the true biomarker and it is positively correlated with the true biomarker, one could enrich the trial cohort by accepting all patients who are identified positively by the surrogate biomarker and a small percentage of the patients who are identified negatively by the surrogate biomarker. Indeed, one can tune the design parameters such that the trial efficiency in terms of the number of randomized patients and the study cost are minimized. These strategies of patient enrichment can be considered as variants of the biomarker-stratified design. A common feature of these designs is that they allow testing multiple hypotheses to address multiple study objectives. In these designs, the issue of multiplicity can be managed by splitting alpha between hypothesis testing on drug effects for positives, negatives, and all patients, or by engaging a sequential testing procedure.
In cases where the biomarker for treatment selection is not well established and has weak credential, a more creative strategy of patient enrichment can be used. Sometimes, there exists a continuous biomarker with reasonable classification accuracy, but an optimal threshold of the putative biomarker that classifies patients has not yet been established prior to the conduct of the trial. In other times, when the trial is designed, a high-dimensional signature that predicts which patients will benefit most from the experimental therapy has not been developed. In these situations, adaptive strategies of patient enrichment can be incorporated to adaptively search for the optimal threshold of the putative biomarker or to build a predictive signature as the trial is in progress. As a matter of fact, adaptive patient enrichment can be used not only as a strategy for improving trial efficiency but also as a process leading to the ultimate goal of a specific trial. For example, one can use adaptive outcome-dependent randomization on accumulative information to increase (decrease) the probability that a patient will be randomized to the treatment to which the patient will be more (less) likely to respond. This process will lead to patients with different genetic profiles gradually enriched in different treatment arms, and the drug can be declared as the most effective treatment for patients with profiles similar to those enriched in the treatment arm.
It is worth emphasizing that our choice of patient enrichment strategy should depend on the study objectives of a specific trial. The biomarker-stratified design and its variants allow us to estimate the drug effect among biomarker-positive patients and biomarker-negative patients, the average treatment effect of all patients, the interaction between drug and biomarker, and the clinical benefit of utilizing the putative biomarker to select treatments. To address these study objectives, we often need to conduct hypothesis testing and statistical inference on different statistical quantities or parameters. In this case, a balanced patient enrichment strategy that could achieve a joint efficiency of the multiple study objectives is preferred. For example, when the evaluation of the clinical utility of a putative biomarker in treatment selection is of primary interest, investigators need to design trials to demonstrate that the benefit of a strategy utilizing biomarker-based treatment selection is larger than the benefit of a strategy that ignores the status of the biomarker in terms of clinical benefit. Equal allocation of positive and negative patients will achieve maximal efficiency in testing the interaction between drug and biomarker. However, if the average treatment effect of all patients is of primary interest, a simple random selection of all patients will achieve the highest efficiency. Different features of the trial design have to be balanced to achieve multiple study goals, which often require different optimal design strategies.
Overall, patient enrichment is a powerful tool that can be used to improve trial efficiency. To use this tool effectively, we need to consider the study objectives, the expected drug effects among different patient groups, and the credential of the putative biomarker.
|Submit a question or comment.|
Each month, one of our program investigators will introduce him/herself and will discuss their research, new research directions, or advancements made toward our goal of improving the clinical trial process. Readers are encouraged to send questions or comments. In addition, we will announce new software releases, publications, and upcoming events.
Date: February 15, 2016
Date: January 15, 2016
Date: December 16, 2015
Date: November 24, 2015
Date: October 15, 2015
Date: September 17, 2015
Date: August 24, 2015
Date: July 16, 2015
Date: June 4, 2015