The Cytel blog keeps you up to speed with the latest developments in biostatistics and clinical biometrics.
A common challenge of working with small sample sizes is determining proper bias correction methods when evaluating a given set of data. Oftentimes, statisticians depend on large sample sizes to naturally correct for any bias. Small sample sizes, by contrast, require innovations like Firth’s famous bias correction method.
Recently, Cytel statisticians Ashwini Joshi and Sumit Singh gave a talk entitled “How to Reduce Bias in the Estimates of Count Data Regression.” Using a number of case studies and simple to use LogXact PROC software, they demonstrate the ease with which bias correction can be implemented for small sample clinical data.
When planning a conventional trial, one can anticipate the drug supply necessary for the trial by determining how the number of patients reflected in the sample size will distribute across the trial sites. Implementing an adaptive trial, by contrast, raises many challenges for predicting the necessary drug supply. It can require planning for different sample sizes depending on the outcome of an interim look; or preparing different dosages if certain arms of a multi-arm trial are to drop after the interim look. In the case of a biomarker-driven adaptive design, determining adequate drug supply may require the ability to predict which doses are necessary for different subpopulations at particular trial sites.
Professor LJ Wei holds that rules are for lawyers, not (necessarily) clinicians. When designing modern clinical trials, the impetus is often to use “efficient and reliable procedures, to obtain clinically interpretable results with respect to risk-benefit analysis…” Yet these efficient and reliable procedures are often just conventions and rules that provide information that is incomplete or difficult to make clinically interpretable.
In a presentation to the East User Group Meeting, Professor Wei identifies 11 problematic areas that currently challenge trial designers. After giving an overview of the challenges that arise in each, Professor Wei provides a few simple solutions about how to overcome them. All the solutions, however, require moving beyond the comfort zone of conventional procedures.
In the slides attached Wei discusses:
A key stage of exploratory drug development is implementing a proof-of-concept study to demonstrate the safety of a drug. Given the importance of accurate dose-finding for Phase 3 success, methodological improvements to proof-of-concept studies in Phase 2 can translate into greater likelihood of getting a drug to market.
Phase 1 oncology trials typically use either rule-based methods or model-based methods to determine the most acceptable level of dose toxicity with which to move forward in Phase 2. This level of toxicity, called the maximum tolerated dose (or the MTD), is the dose which best balances the medical benefits of a higher dose with the risk of toxicity which comes from subjecting a patient to that same dose. Both rule-based methods and model-based methods determine the MTD by relying on small cohorts of patients who test a set of doses against their dose limiting toxicity.
A recent Cytel Seminar on Adaptive Statistical Designs featured a talk by Michael Elashoff (Patient Profiles) on Multivariate Approaches for Risk-Based Monitoring. Elashoff, a former statistical reviewer at the Food and Drug Administration, recommended combining cluster and rules based methods for statistical monitoring. Such adaptive monitoring approaches can substantially reduce the time and expense of data monitoring while ensuring consistently high data quality.
The rise of biomarker based treatments in oncology has meant a reconceptualization of what constitutes a particular disease. According to the American Society for Clinical Oncology, “We can no longer think of cancer as one disease. Even something like lung cancer could be hundreds of different cancers, each defined by specific molecular characteristics requiring different treatment approaches.”  This means that many oncology trials are slowly moving from large-scale studies of generic populations, towards a system where targeted therapies are offered to smaller sets of patients who all possess certain genetic characteristics.
Nina Selaru of Pfizer Oncology, recently gave a talk at a Cytel Seminar in San Diego in which she described a trial for Xalkori, a therapy for non-small cell lung cancer (NSCLC). Pfizer conducted two Phase 3 trials for Xalkori, one for patients who possessed anaplastic lymphoma kinase (ALK-positive patients) and another for other ‘unselected’ patients. The ALK-positive patients were found to respond very well to treatment. Unfortunately, the ALK-positive patients also displayed certain characteristics not present in the other patients: they were younger, non-smokers who displayed signs of adenocarcinoma. There was concern that these characteristics were driving the efficacy of Xalkori.
A new JAMA study on discontinued randomized trials in Switzerland, Germany and Canada, reports that poor recruitment accounts for 101 out of 253 trials that were eventually discontinued (or about 10% of the 1017 trials which participated in the study). When restricted to industry-sponsored trials with non-healthy volunteers, poor recruitment accounted for the discontinuation of 40 trials out of 119 that were discontinued. Across the board, poor recruitment was the foremost cause of trial discontinuity.