Edward L. Korn

Meta-Analysis of Phase II Cooperative Group Trials in Metastatic Stage IV Melanoma to Determine Progression-Free and Overall Survival Benchmarks for Future Phase II Trials

PURPOSE Objective tumor response rates observed in phase II trials for metastatic melanoma have historically not provided a reliable indicator of meaningful survival benefits. To facilitate using overall survival (OS) or progression-free survival (PFS) as an endpoint for future phase II trials, we evaluated historical data from cooperative group phase II trials to attempt to develop benchmarks for OS and PFS as reference points for future phase II trials. PATIENTS AND METHODS Individual-level and trial-level data were obtained for patients enrolled onto 42 phase II trials (70 trial arms) that completed accrual in the years 1975 through 2005 and conducted by Southwest Oncology Group, Eastern Cooperative Oncology Group, Cancer and Leukemia Group B, North Central Cancer Treatment Group, and the Clinical Trials Group of the National Cancer Institute of Canada. Univariate and multivariate analyses were performed to identify prognostic variables, and between-trial(-arm) variability in 1-year OS rates and 6-month PFS rates were examined. RESULTS Statistically significant individual-level and trial-level prognostic factors found in a multivariate survival analysis for OS were performance status, presence of visceral disease, sex, and whether the trial excluded patients with brain metastases. Performance status, sex, and age were statistically significant prognostic factors for PFS. Controlling for these prognostic variables essentially eliminated between-trial variability in 1-year OS rates but not in 6-month PFS rates. CONCLUSION Benchmarks are provided for 1-year OS or OS curves that make use of the distribution of prognostic factors of the patients in the phase II trial. A similar benchmark for 6-month PFS is provided, but its use is more problematic because of residual between-trial variation in this endpoint.

Clinical Trial Designs for Cytostatic Agents: Are New Approaches Needed?

Preclinical data suggest that some new anticancer agents directed at novel targets demonstrate tumor growth inhibition but not tumor shrinkage. Such cytostatic agents may offer clinical benefits for patients in the absence of tumor shrinkage. In addition, lower doses of some of these agents may be just as effective as higher doses, implying that toxicity may not be an ideal end point for dose finding. Because of these factors, the sequence and design of traditional phase I, II, and III trials used for cytotoxic agents (which typically shrink tumors and in a dose-dependent manner) may not be appropriate for cytostatic agents. This article discusses options for modifying trial designs to accommodate cytostatic agents. Examples are given where these options have been tried or are currently being tried. Recommendations given for choosing among the trial designs depend on what is known preclinically about the agents (eg, does one have a validated and reproducible biologic end point that can be used to guide a dose escalation?), what is known about the patient population being studied (eg, does one have a well-documented historical progression-free survival rate at 1 year for comparison with the experience of the new agent?), and the numbers of agents and patients available for participation in trials. Planned and ongoing trials will test the utility of some of these new approaches.

Randomized Clinical Trials With Biomarkers: Design Issues

Clinical biomarker tests that aid in making treatment decisions will play an important role in achieving personalized medicine for cancer patients. Definitive evaluation of the clinical utility of these biomarkers requires conducting large randomized clinical trials (RCTs). Efficient RCT design is therefore crucial for timely introduction of these medical advances into clinical practice, and a variety of designs have been proposed for this purpose. To guide design and interpretation of RCTs evaluating biomarkers, we present an in-depth comparison of advantages and disadvantages of the commonly used designs. Key aspects of the discussion include efficiency comparisons and special interim monitoring issues that arise because of the complexity of these RCTs. Important ongoing and completed trials are used as examples. We conclude that, in most settings, randomized biomarker-stratified designs (ie, designs that use the biomarker to guide analysis but not treatment assignment) should be used to obtain a rigorous assessment of biomarker clinical utility.

Controlling the number of false discoveries: application to high-dimensional genomic data

Abstract Researchers conducting gene expression microarray experiments often are interested in identifying genes that are differentially expressed between two groups of specimens. A straightforward approach to the identification of such “differentially expressed” genes is to perform a univariate analysis of group mean differences for each gene, and then identify those genes that are most statistically significant. However, with the large number of genes typically represented on a microarray, using nominal significance levels (unadjusted for the multiple comparisons) will lead to the identification of many genes that truly are not differentially expressed, “false discoveries.” A reasonable strategy in many situations is to allow a small number of false discoveries, or a small proportion of the identified genes to be false discoveries. Although previous work has considered control for the expected proportion of false discoveries (commonly known as the false discovery rate), we show that these methods may be inadequate. We propose two stepwise permutation-based procedures to control with specified confidence the actual number of false discoveries and approximately the actual proportion of false discoveries. Limited simulation studies demonstrate substantial gain in sensitivity to detect truly differentially expressed genes even when allowing as few as one or two false discoveries. We apply these new methods to analyze a microarray data set consisting of measurements on approximately 9000 genes in paired tumor specimens, collected both before and after chemotherapy on 20 breast cancer patients. The methods described are broadly applicable to the problem of identifying which variables of any large set of measured variables differ between pre-specified groups.

Examples of Differing Weighted and Unweighted Estimates from a Sample Survey

Abstract Unweighted estimators using data collected in a sample survey can be badly biased, whereas weighted estimators are approximately unbiased for population parameters. We present four examples using data from the 1988 National Maternal and Infant Health Survey to demonstrate that weighted and unweighted estimators can be quite different, and to show the underlying causes of such differences.

Chromosome Transfer Induced Aneuploidy Results in Complex Dysregulation of the Cellular Transcriptome in Immortalized and Cancer Cells

Chromosomal aneuploidies are observed in essentially all sporadic carcinomas. These aneuploidies result in tumor-specific patterns of genomic imbalances that are acquired early during tumorigenesis, continuously selected for and faithfully maintained in cancer cells. Although the paradigm of translocation induced oncogene activation in hematologic malignancies is firmly established, it is not known how genomic imbalances affect chromosome-specific gene expression patterns in particular and how chromosomal aneuploidy dysregulates the genetic equilibrium of cells in general. To model specific chromosomal aneuploidies in cancer cells and dissect the immediate consequences of genomic imbalances on the transcriptome, we generated artificial trisomies in a karyotypically stable diploid yet mismatch repair-deficient, colorectal cancer cell line and in telomerase immortalized, cytogenetically normal human breast epithelial cells using microcell-mediated chromosome transfer. The global consequences on gene expression levels were analyzed using cDNA arrays. Our results show that regardless of chromosome or cell type, chromosomal trisomies result in a significant increase in the average transcriptional activity of the trisomic chromosome. This increase affects the expression of numerous genes on other chromosomes as well. We therefore postulate that the genomic imbalances observed in cancer cells exert their effect through a complex pattern of transcriptional dysregulation.

Simultaneous testing of regression coefficients with complex survey data: use of Bonferroni t statistics

Abstract The Wald statistic is frequently used to test hypotheses about beta coefficients from multiple linear regression analyses of complex survey data. This statistic requires a consistent estimate of the variance–covariance matrix of the regression coefficients. In a survey such as the Second National Health and Nutrition Examination Survey, the sample design limits the researcher to at most 32 degrees of freedom for estimating the variances and covariances with either the balanced half-sample replication or the Taylor series linearization procedure. This article considers the properties of the Wald statistic when the number of beta coefficients approaches the degrees of freedom available from the variance estimation. In this situation, Bonferroni-adjusted t statistics are an attractive alternative.

Blinded Independent Central Review of Progression-Free Survival in Phase III Clinical Trials: Important Design Element or Unnecessary Expense?

Progression-free survival is an important end point in advanced disease settings. Blinded independent central review (BICR) of progression in randomized clinical trials has been advocated to control bias that might result from errors in progression assessments. However, although BICR lessens some potential biases, it does not remove all biases from evaluations of treatment effectiveness. In fact, as typically conducted, BICRs may introduce bias because of informative censoring, which results from having to censor unconfirmed locally determined progressions. In this article, we discuss the rationale for BICR and different ways of implementing independent review. We discuss the limitations of these approaches and review published trials that report implementing BICR. We demonstrate the existence of informative censoring using data from a randomized phase II trial. We conclude that double-blinded trials with consistent application of measurement criteria are the best means of ensuring unbiased trial results. When such designs are not practical, BICR is not recommended as a general strategy for reducing bias. However, BICR may be useful as an auditing tool to assess the reliability of marginally positive results.

Outcome-Adaptive Randomization: Is It Useful?

Outcome-adaptive randomization is one of the possible elements of an adaptive trial design in which the ratio of patients randomly assigned to the experimental treatment arm versus the control treatment arm changes from 1:1 over time to randomly assigning a higher proportion of patients to the arm that is doing better. Outcome-adaptive randomization has intuitive appeal in that, on average, a higher proportion of patients will be treated on the better treatment arm (if there is one). In both the randomized phase II and phase III settings with a short-term binary outcome, we compare outcome-adaptive randomization with designs that use 1:1 and 2:1 fixed-ratio randomizations (in the latter, twice as many patients are randomly assigned to the experimental treatment arm). The comparisons are done in terms of required sample sizes, the numbers and proportions of patients having an inferior outcome, and we restrict attention to the situation in which one treatment arm is a control treatment (rather than the less common situation of two experimental treatments without a control treatment). With no differential patient accrual rates because of the trial design, we find no benefits to outcome-adaptive randomization over 1:1 randomization, and we recommend the latter. If it is thought that the patient accrual rates will be substantially higher because of the possibility of a higher proportion of patients being randomly assigned to the experimental treatment (because the trial will be more attractive to patients and clinicians), we recommend using a fixed 2:1 randomization instead of an outcome-adaptive randomization.

Choice of column scores for testing independence in ordered 2 X K contingency tables.

The numerous statistical methods for testing no association between a binary response (rows) and K ordered categories (columns) group naturally into two classes: those that require preassigned numerical column scores and those that do not. An example of the former would be a logistic regression analysis, and of the latter would be a Wilcoxon rank-sum test. In this paper we demonstrate that the perceived advantage of not preassigning scores is illusory. We do this by presenting an example from our consulting experience in which the midrank scores used by the rank tests that do not require preassigned scores are clearly inappropriate. Our recommendations are to assign reasonable column scores whenever possible, and to consider equally spaced scores when the choice is not apparent. Midranks as scores should always be examined for their appropriateness before a rank test is applied.