Subgroup Analysis Identification: The Hardest Problem There Is
Stephen Ruberg (Analytix Thinking, LLC)
February 4
There have been many cautionary publications warning about findings from subgroup analyses, and rightfully so. Quite often positive subgroup claims are based on exploratory analysis or post hoc assessments after data have been unblinded, and quite often those findings are not reproduced in subsequent research. Consequently, the assessment of subgroups in clinical trials is viewed skeptically and even dismissed. And yet, we know there is inherent heterogeneity of treatment effects, and the push for more tailored therapeutics and personalized medicine demands that we look for subgroups of patient who may have a differential treatment effect – either positive or negative. This talk will explore the balance between the skepticism associated with subgroup analysis and the optimism with subgroup identification. Some perspectives on quantifying confidence in subgroup findings will be presented as well as some different ideas on the notion of defining subgroups. The goal is to make our thinking patient centric and to hopefully improve our approaches for getting the right medicines to the right patients.
Mini-symposium on Real World Data/Real World Evidence
Weili He (AbbVie), Martin Ho (FDA/CBER) & Diqiong Xie (Seattle Genetics)
March 5th
Real-world data (RWD) generated from clinical practice and utilization of digital health technologies – outside of clinical trials – are regarded as a pragmatic data source with high potential to generate real-world evidence (RWE). There have been increasing interest in using RWD and translating RWD to RWE to support clinical development and life cycle management of medical products. Intriguing statistical challenges include assessment of fit-for-purpose RWD, leadership in improving internal processes and infrastructure, consideration of RW study design that minimizes biases and confounding, and utilization of advanced analytics for analysis of RW studies. In this webinar, speakers from the industry and the FDA will share their perspectives on this topic. The first speaker will share some latest findings of the ASA Biopharmaceutical Section RWE Working Group, which was chartered in 2017 to facilitate statisticians’ leadership in this area. The second speaker will discuss various RWE examples at AbbVie, along with challenges in RWE research. The talk will focus on addressing challenges using quantitative approaches in translating RWD to robust RWE. The last presentation will talk about the current practice using real-world data in FDA, which includes the types of data we have used. A few examples will be presented from both pre-marketing and post-marketing settings.
Hazards of Hazard Ratios in Survival Analysis
L.J. Wei (Harvard University)
May 17
In a longitudinal clinical study to compare two groups, the primary end point is often the time to a specific event (for example, disease progression, death). The hazard ratio estimate is routinely used to empirically quantify the between-group difference under the assumption that the ratio of the two
hazard functions being approximately constant over time. Even when this assumption is plausible, such a ratio estimate may not give us a clinically meaningful summary of the group contrast due to lack of a reference value of hazard function from the control arm. Moreover, the clinical meaning of such a ratio estimate is difficult, if not impossible, to interpret when the underlying proportional hazards assumption is violated (namely, the hazard ratio is not constant over time). For this case, the hazard ratio-based tests may not have power to detect the group difference. In this talk, we summarize several critical concerns regarding this conventional practice and discuss alternatives (e.g., via the t-year mean survival time) for quantifying the differences between groups with respect to a time-to-event end point. The data from several recent cancer and cardiovascular clinical trials, which reflect a variety of scenarios, are used throughout to illustrate our discussions. In this talk, we are mostly interested in estimation the treatment effect beyond the hypothesis testing paradigm. An estimation procedure can also be used as a test statistic. On the other hand, most tests in survival analysis, such as the weighted logrank tests, do not have appropriate estimation counterparts. We will also discuss other relevant issues in clinical studies, for example, estimating the duration of response, quantifying long term survival et al.
Survival Analysis Using a 5-STAR Approach in Randomized Clinical Trials
Devan Mehrotra (Merck & Company)
May 26
Randomized clinical trials are often designed to assess whether a test treatment prolongs survival relative to a control treatment. Increased patient heterogeneity, while desirable for generalizability of results, can weaken the ability of common statistical approaches to detect treatment differences, potentially hampering the regulatory approval of safe and efficacious therapies. A novel solution to this problem is proposed. A list of baseline covariates that have the potential to be prognostic for survival under either treatment is pre-specified in the analysis plan. At the analysis stage, using observed survival times but blinded to patient-level treatment assignment, ‘noise’ covariates are removed with elastic net Cox regression. The shortened covariate list is subsequently used by a conditional inference tree algorithm to segment the heterogeneous trial population into subpopulations of prognostically homogeneous patients (risk strata). After patient-level treatment unblinding, a treatment comparison is done within each formed risk stratum and stratum-level results are combined for overall statistical inference. The impressive power-boosting performance of our proposed 5-step stratified testing and amalgamation routine (5-STAR), relative to that of the logrank test and other common approaches that do not leverage inherently structured patient heterogeneity, is illustrated using a hypothetical and two real datasets along with simulation results. In addition, the importance of reporting stratum-level comparative treatment effects (time ratios from accelerated failure time model fits with model averaging and, if needed, hazard ratios from proportional hazard model fits) is highlighted as a potential enabler of personalized medicine. An R package is offered for implementation.
Incorporate External Control Data in New Clinical Trial Design and Analysis
Lanju Zhang (AbbVie)
June 11
Many clinical trials are designed with plenty of external control data available. To reduce the ever-increasing cost and timeline of the clinical drug development, external control data are incorporated in clinical trial design and analyses. Two methods can be applied. The first method is to incorporate summary level information into the new trial through a Bayesian framework. Bayesian methods include power prior approach, commensurate prior approach and mixture prior approach. In this presentation, we will discuss a newly proposed statistical approach which combines elements of all the three approaches mentioned above but admits a closed-form formula for easy implementation. The second method is to incorporate subject level external data through propensity score methods. In this presentation, we will share our experiences of using propensity score matching to create a synthetic control arm. We will demonstrate these approaches with case studies and share our interaction experiences with regulatory agencies.
The Leadership Laboratory: Using Observational Study to Develop Leadership Skills for Statisticians
Gary Sullivan (Espirer Consulting)
June 18
Where do you start if you want to improve your leadership? Take a leadership course? Read a leadership book? Find an experienced mentor? All of those are good options but consider one more: Leadership practice - both good and bad - is on display every day right before your eyes. Each of you is living in a leadership laboratory where observational study can provide great leadership learning. You can observe business partners and statistical colleagues applying and practicing communication techniques, negotiation tactics, and influence strategies in your meetings, interactions, and collaborative projects. In this presentation, I will discuss some very simple practices which provide powerful lessons on influence, and I will point out the most important leadership skills for statisticians to study. I will also demonstrate through personal experiences and examples how to identify leadership concepts, assess their application, and use the information to build and improve your skills.
Biopharmaceutical Section Offers Advice on Strategic Planning for ASA Fellow Nomination (video)
Ivan Chan (AbbVie), Paul Gallo (Novartis), Christy Chuang-Stein (Chuang-Stein Consulting), Stephen Ruberg (Analytix Thinking, LLC)
September 17
This webinar will provide nominators and nominees for ASA Fellowship with a background on the current process and advice on optimizing nomination strategies. The webinar will feature several speakers who had been successfully nominating candidates throughout their careers. The talks will be followed by a panel discussion including past and current members of the ASA Fellows Committee and the Fellows Committee of the ASA Biopharma Section, and contributors to a recent publication in Amstat News.
We hope that the advice and discussion will be useful for a broad audience of Pharma statisticians who are considering the nomination process.
Detangle Modern Dose-Finding Designs: A Tutorial (video)
Ji Yuan (University of Chicago)
October 1
The landscape of dose-finding designs for phase I clinical trials is rapidly shifting in the recent years, noticeably marked by the emergence of designs utilizing toxicity probability intervals. While many de- signs appear to be similar on the surface, the underlying statistical frameworks can be quite different. In addition, it becomes challenging to assess the operating characteristics of these designs via simulations as different designs might be superior in different scenarios, based on different criteria, etc. Recent review papers tend to focus on some aspects of design comparison which may not reveal the global picture of pros and cons of each design. We provide a tutorial and a new criterion that summarize the multi-dimensional operating characteristics into a single value, called the "J-Score". J-Score takes into account user preference in drug development strategy and safety protection. For example, for some drugs, dose-finding strategy might prefer fast escalation as the safety of the drug has been reliably established. The users can require J-Score to reflect the preference of aggressive dose-finding when assessing various designs. We conduct a comprehensive simulation study to illustrate the performance of main-stream dose-finding designs based on the J-Score. We show the top-designs based on different preferences.
Biomarker Analysis in Clinical Trials Using R (video)
Nusrat Rabbee (Eisai)
October 21
This course is based on “Biomarker Analysis in Drug Clinical Trials with R” - a book, which offers practical guidance to statisticians in the pharmaceutical industry on how to incorporate biomarker data analysis in clinical trial studies. We will discuss the appropriate statistical methods for evaluating biomarkers in different stages of clinical development: as pharmacodynamic, predictive, and surrogate biomarkers for delivering increased value in the drug development process. The topic of combining multiple biomarkers to predict drug response using machine learning is covered. We will discuss uses and abuses of biomarkers in drug development.
We will cover the models which will help students, researchers, and biostatisticians get started in tackling the hard problems of designing and analyzing clinical trials with biomarkers.
The course will highlight:
- Analysis of pharmacodynamic biomarkers for lending evidence target modulation;
- Design and analysis of trials with a predictive biomarker;
- Framework for analyzing surrogate biomarkers;
- Methods for combining multiple biomarkers to predict treatment response;
- Writing Biomarker statistical analysis plan;
- Uses and abuses of biomarkers.
ESPSI-PSI-BIOP Joint Webinar on Estimands (video)
John Scott (FDA/CBER), Andreas Brandt (BfArM), Evgeny Degtyarev (Novartis) & Vladimir Dragalin (Janssen)
November 5
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 Estimands in Practice. Speakers from regulatory authorities (FDA and EMA) and industry will present on their experience on the recent ICH E9 (R1) guidance, including the following aspects:
- Experience with proposals submitted to FDA and EMA on implementation of estimands;
- How the estimands framework facilitates interaction with clinicians in different therapeutic areas;
- Common problems where the Estimands framework can help advance research;
- Where further discussions and research is required, and particularly where industry and regulators can collaborate;
- Issues related to alignment between different estimators to a given estimand;
- Special considerations of estimand framework in COVID-19 vaccine trials.