From the registration point of view, biomarkers need to be valid and reproducible. Regulatory authorities are required to ensure that standardized tests are available for biomarker testing. The average pathology laboratory should be able to deliver them in a rapid, cheap, and effective manner.
Equally important for registration, the biomarker needs to be included in any reimbursement schedule. A classic example of proof-of-concept assessment is the identification of the KRAS mutational status as a predictive marker in the context of anti-EGFR antibody therapy. The KRAS protein regulates downstream proteins in the EGFR signaling pathway that are associated with tumor survival, angiogenesis, proliferation, and metastasis.
This mutation is found in one third of CRC patients. According to the post-hoc analysis, however, when mutations status was considered, there was a distinct difference between the two treatment arms [2].
In KRAS-mutant patients, cetuximab did not show any benefit, whereas those with KRAS wildtype obtained significant overall survival advantage when treated with this antibody. Once biomarker development reaches the level of the diagnostic laboratory, very high standards are required.
Clinical validation should ideally be conducted in a prospective study, rather than in the retrospective setting, as biomarker analysis on the basis of existing data from randomized controlled trials has several drawbacks [4]. In many cases, the original study will not have been powered for a correlative science endpoint, and tissue has not necessarily been obtained in all of the randomized patients.
Outcomes obtained in a retrospective setting are generally considered hypothesis-generating and need to be confirmed in prospective studies. Gains in efficiency depend on the marker prevalence and the relative efficacy in biomarker-positive and biomarker-negative patients. Trials with an enrichment design only enroll patients who are likely to respond. For KRAS mutation testing, seven different methodologies were compared in [6]. Today, next-generation sequencing is increasingly becoming the standard technique as part of a much broader panel to examine mutations across several targetable pathways.
Today, all-RAS mutation testing is recommended by the guidelines [8]. Regulatory authorities are generally perceived as being difficult and bureaucratic, but they serve an important function in terms of protecting and promoting public health. They ensure quality, safety, and efficacy of treatment. Also, they provide adequate and appropriate information for both patients and physicians.
Naturally, their approach is conservative and careful, as mistakes on their part can have vast consequences. The instructions of the regulatory authorities should thus be followed closely. The approaches to regulation vary worldwide.
For instance, the Japanese regulatory authorities believe that clinical data from foreign patients are of limited applicability to Japanese patients, which is why they require stand-alone development programs.
However, there are many similarities. Regulatory authorities generally run comprehensive websites that provide useful information on biomarker qualification, such as numerous academic inputs into the optimal study design.
The regulatory authorities are still assessing the optimal ways to incorporate biomarkers into clinical trials. For the future, it can be expected that countries will increasingly cooperate in this respect.
At present, researchers have to provide protocols that detail all of the proceedings planned for the study, as well as the quality assurance measures. In most cases, the rationale for a trial will be supported by experimental findings provided by other study groups. Only a minority of researchers can provide their own laboratory data.
Adaptive designs Adaptive design strategies are a class of randomized Phase II designs by which a variety of marker signatures and drugs can be tested under one umbrella protocol. Table 1 Criteria for choice of design for initial marker validation trials. Criteria Design Enrichment Allcomers Adaptive Preliminary evidence Strongly suggest benefit in marker-defined subgroups. Figure 2. Allcomers design Hybrid designs In this design strategy, only a certain subgroup of patients based on their marker status are randomized between treatments, whereas patients in the other marker-defined subgroups are assigned the standard of care treatment s [ 12 ].
Marker by treatment interaction design In this design, all patients meeting the eligibility criteria are entered into the trial [ 17 ]. Sequential testing strategy designs Sequential testing designs are similar in principle to a RCT design [ 34 — 36 ]. Future perspective In this section, we speculate on the anticipated state of clinical trial designs for marker validation in the next 5—10 years.
Molecular predictors of response to epidermal growth factor receptor antagonists in non-small-cell lung cancer. Selecting patients for treatment with epidermal growth factor tyrosine kinase inhibitors.
Cancer Res. Wild-type KRAS is required for panitumumab efficacy in patients with metastatic colorectal cancer. Genomic and molecular profiling predicts response to temozolomide in melanoma. A genomic approach to identify molecular pathways associated with chemotherapy resistance. Cancer Ther. A genomic approach to colon cancer risk stratification yields biologic insights into therapeutic opportunities. Natl Acad. Lancet Oncol. Predictive biomarker validation in practice: lessons from real trials.
Equipoise lost: ethics, costs, and the regulation of cancer clinical research. Genomic advances and their impact on clinical trial design. Genome Med. Clinical trial designs for predictive biomarker validation: theoretical considerations and practical challenges. Allcomers versus enrichment design strategy in Phase II trials.
Thorac Oncol. Association of tumour site and sex with survival benefit from adjuvant chemotherapy in colorectal cancer. Maitournam A, Simon R. On the efficiency of targeted clinical trials. Clinical trial designs for predictive marker validation in cancer treatment trials. I-SPY 2: an adaptive breast cancer trial design in the setting of neoadjuvant chemotherapy.
Bayesian adaptive design for targeted therapy development in lung cancer — a step towards personalized medicine.
Clinical Trials. Cancer Disc. Cetuximab for the treatment of colorectal cancer. K-RAS mutations and benefit from cetuximab in advanced colorectal cancer. Cetuximab and chemotherapy as initial treatment for metastatic colorectal cancer.
Trastuzumab plus adjuvant chemotherapy for operable HER2-positive breast cancer. HER2 status and benefit from adjuvant trastuzumab in breast cancer. Hayes DF. Steady progress against HER2-positive breast cancer. Phase III trial of chemotherapy plus radiotherapy compared with radiotherapy alone for pure and mixed anaplastic oligodendroglioma: Intergroup Radiation Oncology Group Trial A t 1;19 q10;p10 mediates the combined deletions of 1p and 19q and predicts a better prognosis of patients with oligodendroglioma.
Cairncross G, Jenkins R. Cancer J. Sparano JA, Paik S. Development of the gene assay and its application in clinical practice and clinical trials.
Simon R, Wang SJ. Use of genomic signatures in therapeutics development. Pharmacogenomics J. Bauer P.
Multiple testing in clinical trials. A method for testing a prespecified subgroup in clinical trials. Biomarker-adaptive threshold design: a procedure for evaluating treatment with possible biomarker-defined subset effect.
Natl Cancer Inst. Assessment of reproducibility of a classifier within a study is referred to as an internal validation, and across studies is as an external validation.
In theory, some classifiers may have good predictability but poor reproducibility, or vice versa. However, predictability appears to be a necessary condition for a classifier to have good reproducibility. Classifiers with high predictability and reproducibility are obviously desirable. A predictive classifier should perform well in both predictability and reproducibility. To completely validate a predictive classifier, more than one retrospective and prospective trial may be needed.
Predictive classifier should be internally and externally validated prior to clinical validation. For survival outcomes, the predictive scores are estimated before assigning a threshold cutoff to define biomarker-positive and biomarker-negative patients.
There are several measures and methods for the evaluation of the estimated predictive scores These measures primarily evaluate agreement between the predictive scores and the observed survival times.
They include the concordance index 66 - 68 , Brier scores 69 , log-rank P value and several others 70 - A prediction model should have a high concordance score before determining the threshold cutoff to classify positive and negative biomarker subgroups.
Adaptive clinical trial design is used not only to validate a predictive biomarker classifier, but also to optimize treatment selection. Performance of a classifier depends on the prevalence proportion, sample size, and the biomarker set identified. It is worth mentioning that for prediction problems, the omission of biomarkers false negatives would have more serious impact on accuracy than the inclusion of non-biomarkers false positives in the classifier.
The FDR approach uses a less stringent criterion to select potential predictive biomarkers, it should be a appropriate error measure for the interaction test to identify predictive biomarkers. The ASD was proposed as a supplementary test when the test for overall treatment effect is not significant.
Scher et al. Reducing the significance level requires an increase of the trial sample size. Furthermore, the power of the subgroup test depends greatly on the prevalence proportion and the effect size in the specific subgroup of interest. In general, a larger sample size is often needed in an ASD design if detection of the subgroup effect is of primary interest and the effect size in the subgroup is not large. An in silico experiment was conducted to illustrate development of a biomarker adaptive design in a two-arm control and treatment study.
The number of patients per arm was ; the proportion of the biomarker positive subgroup was 0. The probability of response for all patients in the control arm was 0. The total number of genomic variables was 5,, among which there were 10 predictive biomarkers. The genomic variables were generated from an independent normal distribution with mean 0 for the non-predictive biomarkers and with mean 1.
The standard deviation was 0. A typical observed dataset was analyzed for illustration. The total number of biomarker positive patients is 40 and the total number of biomarker negative patients is From the control arm, the observed positive responses were 3 and 30 for the positive and negative subgroups, respectively; from the treatment arm, the observed positive responses were 15 and 36 for the positive and negative subgroups, respectively.
The level of significance in the interaction test was set at 0. For the fold cross validation, the average number of significant genes identified was For the 2-fold cross validation, the average number of significant genes identified were 3. The analyses focused on comparison of the two classifiers to identify biomarker positive patients. Two main considerations in the evaluation are: I the sensitivity and specificity of the classifiers and II the power of the subgroup test.
These four cases cover the best choices for the two parameters. Table 1 shows the performance of the two classifiers. The fold cross validation approach shows much better performance than the 2-fold cross validation approach. In the MLV method, a small value of ln R or G represents a mild tuning parameter to select biomarker positive patients resulting in higher sensitivity and lower specificity.
MLV[1,2] correctly identified 40 biomarker positive patients high sensitivity , but it also incorrectly identified 51 biomarker negative patients as biomarker positives low specificity. Table 2 shows the P values of the subgroup test on the biomarker positive patients identified by the two classifiers. The 20 patients included all 3 positive outcomes in the control arm and all 15 positive outcomes in the treatment arm.
The P value of the subgroup test was 0. MLV[1,2] identified 91 biomarker positive patients, 63 from the control arm and 28 from the treatment arm. Among of the 63 patients in the control arm, 25 showed positive outcomes, but only 3 were biomarker positive patients.
Among the 28 patients in the treatment arm, 18 showed positive outcomes of which 15 were biomarker positive patients. The subgroup test for biomarker negative patients was not significant. The performance of a biomarker adaptive design depends on n, p, u 00 , u 01 , u 10 , and u Further simulation for various combinations of these parameters will be helpful in the development of biomarker adaptive design for that analysis of subgroup effects.
Development of predictive biomarkers remains challenging in clinical trials. Many factors contribute to such slow progress; a main reason is that genomic data analysis is considered as an exploratory objective for hypothesis generating due to the complexity of the high-dimensional nature and generally without well-defined clinical hypotheses.
The ASD attempts to address this issue by integrating the signature development into the primary objective for phase III randomized trials. This strategy sheds new light on plausible use of genomic biomarkers as a trial objective. However, vigorous statistical methodology is needed to achieve this goal. In this review, we present the key steps in the development of predictive biomarkers to assess treatment effect and discuss statistical issues encountered in the development of a biomarker adaptive design for clinical trials.
This involves a statistical test for the interaction effect model to identify potentially useful biomarkers to classify patients into subgroups. It is worth noting again that the biomarker identification and patient subgroup classification were developed sequentially in a single trial. In a clinical trial dataset involving high dimensional genomic variables, such as gene expression data, whole genome scanning data, next generation sequencing data, the number of subjects studied is often far less than the number of genomic variables due to financial feasibility.
There can be multiple biomarker classifiers that are similarly plausible with comparable performances. There are several challenges in both biomarker identification and classifier development. One major challenge is multiplicity in biomarker classifier development. Here, multiplicity does not refer to hypothesis testing, but to the selection among multiple plausible predictive models. However, some may argue that the multiplicity in biomarker classifier development is irrelevant as long as a responsive patient subset can be identified.
Due to the empirical nature of classifier development, the predictive models are governed by the pre-specified tuning parameter sets if, for example, an ASD classifier is considered. In the absence of true effect sizes, it is worth noting that there may be no clearly predictable relationship between the choices of the tuning parameters and the ability to demonstrate treatment effect in the biomarker positive subset when the unknown true treatment effect size is not large in the biomarker positive patients, see scenario V in This context of use may be expanded to additional patient populations if we receive follow-up submissions that demonstrate the usefulness of galactomannan in these groups.
Patients with IA were previously diagnosed based on microbiology cultures alone or together with histopathologic, microscopic, or radiologic test results.
All of these tests, alone or in combination, have sensitivity or specificity issues, and may take a long time to produce results. Due to the high morbidity and mortality of this disease and the challenges of earlier diagnostic methods, it was often too late to enroll patients with IA in a clinical trial. Measurement of galactomannan is more sensitive than other methods currently in use, and as noted in our qualification draft guidance, it can replace all other diagnostic criteria used to select the specified patient subpopulation for these clinical trials.
It is hoped that this qualification will help streamline the identification of patients with probable IA for enrollment in clinical trials and improve the development process for IA drugs.
Many stakeholders share the hope that increased availability of qualified biomarkers will facilitate innovation in drug development and view biomarker development and qualification as a collaborative undertaking necessitating input from experts across the research and development landscape. We are reaching out to a wide array of individuals with scientific, clinical and toxicological expertise to identify biomarkers that are needed to improve drug development.
In , we convened a workshop with Howard Hughes Medical Institute to discuss potential use cases for biomarkers. This year, a meeting hosted by the Engelberg Center for Health Care Reform and convened by the Brookings Institution and FDA was held to help advance the use of biomarkers in drug and biologic product development. The BQP will continue to work with our colleagues in academia, industry and other government and regulatory agencies to develop evidentiary standards and facilitate biomarker development and qualification.
0コメント