2021 Webinars
Analysis of Multiple Outcome Measures with Applications to Disability Improvement in Multiple Sclerosis
Wenting Cheng; Biogen
January 19
In clinical trials, measurements or endpoints in various domains are assessed and usually combined to evaluate the totality of the treatment effect of a specific treatment. This strategy is common in neurological disease area where many patient performance assessments have been developed. The most common and simplest way to combine multiple measures is through constructing a composite endpoint when all outcome measures are binary or can be dichotomized, and an event occurrence is defined as if any of the component outcome measures achieves an event. Alternatively, the overall evaluation can be achieved analytically by analyzing each outcome measure separately, e.g., through multivariate regression. In literature, there is no systematic evaluation of these various approaches and performance comparisons. In this project, we first propose two general frameworks to combine multiple measures, a composite endpoint approach and a model-based approach. Statistical properties of the approaches are then evaluated using disability improvement in multiple sclerosis as an example. We finally illustrate our methodology through simulations and an application to a motivating clinical trial data.
Title: Variance estimation for the Kappa statistic in the presence of clustered data and heterogeneous observations
Mary Ryan; University of California, Irvine
January 19
We present methodology motivated by a controlled trial designed to validate SPOT GRADE, a novel surgical bleeding severity scale (Spotnitz et al., 2018). Briefly, the study was designed to quantify inter- and intra-surgeon agreement for characterizing the severity of surgical bleeds via a Kappa statistic. Multiple surgeons were presented with a randomized sequence of controlled bleeding videos and asked to apply the rating system to characterize each wound. Each video was shown multiple times to quantify intra-surgeon reliability, creating clustered data. In addition, videos within the same category may have had different classification probabilities due to changes in blood flow rates and wound sizes. In this work, we propose a new variance estimator for the Kappa statistic, for use in clustered data as well as heterogeneity among items within the same classification category. We then apply this methodology to data from the SPOT GRADE trial.
A Cardiologist and a Statistician Walk into a DMC
Janet Wittes and Marc Pfeffer
March 9
Whenever a physician and a statistician are involved in a clinical study, they should make sure that the statistician understands the medical issues and the physician understands the statistical approaches. The need for this mutual understanding is especially important in Data Monitoring Committees (DMCs) because the DMC must make important recommendations over the course of a few hours. In this conversation, we begin with a summary of medical and statistical lessons we learned from CAST, a trial that was formative to both of us even though neither of us had direct involvement in it. We then discuss a series of major cardiovascular and oncology trials in which the two of us served together – sometimes both as members of a DMC, sometimes when one of us (MP) chaired the DMC and when the other (JW) acted as the independent reporting statistician, and sometimes when MP was the Chair of the Steering Committee and JW reported to the DMC. The venues were varied – a castle in Bergen, Norway; the Hall of Mirrors in Versailles; the Chicago O’Hare Hilton, and, of course, most recently, Zoom. In all these cases, we acted both as teachers and as students, recognizing the importance of understanding the other’s discipline so that we would be able to formulate a sensible recommendation for an ongoing trial whose outcome could have a major impact on the public health.
Graphical approaches for the control of generalized error rates
Frank Bretz
May 20
When simultaneously testing multiple hypotheses, the usual approach in the context of confirmatory clinical trials is to control the familywise error rate (FWER), which bounds the probability of making at least one false rejection. In many trial settings, these hypotheses will additionally have a hierarchical structure that reflects the relative importance and links between different clinical objectives. The graphical approach of Bretz et al (2009) is a flexible and easily communicable way of controlling the FWER while respecting complex trial objectives and multiple structured hypotheses. However, the FWER can be a very stringent criterion that leads to procedures with low power, and may not be appropriate in exploratory trial settings. This motivates controlling generalized error rates, particularly when the number of hypotheses tested is no longer small. We consider the generalized familywise error rate (k-FWER), which is the probability of making k or more false rejections, as well as the tail probability of the false discovery proportion (FDP), which is the probability that the proportion of false rejections is greater than some threshold. We also consider asymptotic control of the false discovery rate, which is the expectation of the FDP. In this presentation, we show how to control these generalized error rates when using the graphical approach and its extensions. We demonstrate the utility of the resulting graphical procedures on clinical trial case studies.
Generalized pairwise comparisons for benefit/risk assessment in personalized medicine
Marc Buyse and Julien Péron
May 27
A novel statistical approach to the analysis of randomized clinical trials uses all pairwise comparisons between two patients, one in the treatment arm and one in the control arm. Each pair favors treatment (“win”), control (“loss”), or neither. The “net treatment benefit” is the difference between the proportion of wins minus the proportion of losses. Pairwise comparisons can incorporate several outcomes of interest and several thresholds of clinical relevance in the analysis, and as such, they can be used to personalize treatment choices and to assess the benefit/risk of randomized therapeutic interventions in a rigorous yet flexible manner (Buyse 2010).
The advantages and limitations of generalized pairwise comparisons will be illustrated using two typical examples:
- For a single time-to-event endpoint, the net survival benefit is a meaningful measure of treatment effect whether or not hazards are proportional. When a delayed treatment effect is anticipated, for example in immune-oncology trials, the net benefit is appealing because it stresses benefits that are clinically worthwhile on the time scale. The test based on the net survival benefit can also gain power as compared to the traditional logrank test if interest focuses on long-term survival differences (Péron 2016a).
- Most anticancer treatment have substantial toxicities that may counterbalance treatment benefits. Generalized pairwise comparisons can be used to assess the benefit-risk balance of new treatments. This will be illustrated using several randomized trials in patients with metastatic pancreatic cancer (Péron 2016b).
Understanding hypothetical strategies and defining the clinical question of interest (Bretz, Nie)
Frank Bretz (Novartis) and Lei Nie (FDA)
Panelists: Catherine Njue (Health Canada), Elena Polverejan (J&J) & Speakers
July 23
More and more clinical trials have started adopting the estimand framework in protocols after the release of the ICH E9(R1) “Addendum on Estimands and Sensitivity Analysis in Clinical Trials”. However, there is currently still a lack of clarity on the role of hypothetical strategies. The addendum recognizes that “when using the hypothetical strategy, some conditions are likely to be more acceptable for regulatory decision making than others”. However, it is still unclear as to what constitutes “acceptable hypothetical conditions” which could lead to an interpretable treatment effect. Therefore, it is of great interest to have a diverse group of experts take a deep dive into this topic and help shed light on considerations of hypothetical strategies relevant to the clinical question of interest. In this webinar, we will have speakers from regulatory agencies and industry to share their perspectives on this interesting topic and also participate in a panel-discussion.
A Bayesian phase I/II platform design for co-developing drug combination therapies for multiple indications
Ying Yuan
August 5
It is increasingly common to combine a new targeted or immunotherapy agent with the cancer-specific standard of care to treat different types of cancers. We propose a master-protocol-based, Bayesian phase I/II platform design to co-develop combination (BPCC) therapies in multiple cancers. Under the BPCC design, only a single master protocol is needed, and the drug is evaluated in different cancers in a concurrent or staggered fashion. For each cancer, we jointly model dose-toxicity and -efficacy relationships and employ Bayesian hierarchical models to borrow information across them for more efficient cancer-specific decision making. To account for the characteristic of targeted or immunotherapy agents that their efficacy may not monotonically increase with the dose, and often plateau at high doses, we use the utility to quantify the risk-benefit tradeoff of the treatment. At each interim, we update the toxicity and efficacy model, as well as the estimate of the utility, based on the observed data across cancers to inform the cancer-specific decision of dose escalation and de-escalation and identify the optimal biological dose for each cancer. Simulation study shows that the BPCC design has desirable operating characteristics, and that it provides an efficient approach to accelerate the development of combination therapies.
Putting Pharmacovigilance into Action
Frank W. Rockhold and Tjark Reblin
September 16
Every pharmaceutical company collects, analyzes, and reports safety data collected in trials and in the general use on their products to fulfill regulations and detect new signals to keep their product labels up to date. While those actions have an indirect benefit for the patient, to truly help them and their healthcare practitioners it is necessary to go beyond the raw data and translate the information into the benefit to risk context, informing risk management plans and specific mitigation actions that are derived from data. The goal of pharmacovigilance should be to minimize pain and suffering in patients who strive to benefit from the medicine prescribed by their physician. By properly characterizing benefits and risks of medicines in appropriate populations and semi-quantitatively or quantitatively estimating whether and to which extent the benefits outweigh the risks one can create tangible context for regulators, prescribers, and patients. This informs specific action plans for the health care provider to optimize the use of the treatment ensuring the right medicine gets to right patient. We will present our experience on some specific examples derived directly from our joint experience in safety and pharmacovigilance.
Joint PSI, EFSPI & ASA BIOP Webinar: Complex Innovative Designs (Scott, Collignon, Schmidli)
John Scott (FDA), Olivier Collignon (GSK), and Heinz Schmidli (Novartis)
October 21
PSI, the European Federation of Statisticians in the Pharmaceutical Industry (EFSPI) and the Biopharmaceutical Section of the American Statistical Association (ASA) are jointly organising a webinar on Complex Innovative Designs (CID) in practice. Speakers from regulatory authorities and industry will present on their experience, which will include the following aspects:
- Overview of the FDA Complex Innovative Trial Design pilot program and the applications received to date together with details on some of them
- Overview of the FDA guidance on interacting on Complex Innovative Trial Designs
- Detailed case study of a clinical trial in children which was evaluated within FDA’s CID pilot program, applying borrowing of information from external trials in adults
- Overview of statistical and regulatory considerations on master protocols, focusing on Phase III confirmatory trials.
Response Assessment in Cancer Clinical Trials (Gonen, Schwartz)
Mithat Gonen and Lawrence Schwartz
October 27
Overall survival or progression-free survival are considered gold standard primary endpoints in Phase III cancer trials but their use in earlier trials has been limited due to the long follow-up time required in most diseases. Instead response to treatment, defined as a certain amount of shrinkage in the sum of tumor sizes, is commonly used in Phase Ib and II settings. Despite this common use, definition of response is very challenging, in particular dealing with multiple lesions, determining the cutoff in shrinkage, the time of response assessment, dealing with new lesions and measurement error due to reader-to-reader variability. The correlation with clinical outcomes and survival is not always as strong as one would expect for a myriad of these reasons. We will present our joint work in this field with a history of 20 years, pointing out to improvements in the field as well as the remaining challenges.