<img alt="" src="https://secure.lote1otto.com/219869.png" style="display:none;">
Skip to content

Significance of Bayesian Model-Based Approaches in Oncology Trials: An Interview with Dr. Satrajit Roychoudhury

Cytel conducted a webinar with Dr. Satrajit Roychoudhury, Senior Director, Statistical Research and Data Science Center, Pfizer. Dr. Roychoudhury talked about practical model-based approaches for phase I oncology trials. This webinar is a part of Cytel’s “Introduction to Complex Innovative Trial Designs” webinar series. You can watch the recording by clicking on the button below.

Access Replay

In this blog, we bring to you an insightful interview with Dr. Satrajit Roychoudhury where he talks to us about his interest in statistics, explains the concept of Bayesian model-based approaches and their importance in oncology trials.

Can you give us some background about your career? How did you get involved in Statistics, particularly in innovative statistical methodology?

Satrajit Roychoudhury (SR): I was first introduced to statistics during my high school. My teacher had a big influence on my continued interest in statistics. The book “The World Is Built on Probability” by L. V. Tarasov was my introduction to uncertainty quantification in real world applications, which motivated me to pursue a career in Statistics. For my undergraduate degree, MS and PhD, my focus was mainly on mathematical statistics and probability. Dr. Manish Bhattacharjee was a very helpful advisor who always encouraged independent and out of box thinking. My academic exposure to Bayesian statistics remains limited to an introductory course though since then I have used methods widely in application.

I received my first-hand experience in clinical trials at my first job as a clinical statistician in Schering-Plough (now Merck). During this time and later while working at Novartis, I learned different aspects of drug development and regulatory needs. It also helped me understand the areas where statistical innovation in clinical trial design and analysis can make a great difference for patients.

I am interested in model-based approaches and Bayesian statistics which provide a formal way of combining information from different sources for better decision-making in health care. My training in mathematical statistics enables me to work on the new methodologies, with a strong bent on applications. After working for 5 years as a clinical statistician, I moved to the Statistical Methods and Consulting group in Novartis. There, I worked with different trial teams to promote the need of innovative methods for better design, analysis and quantitative decision-making. In 2017, I moved to the Statistical Research and Innovation group in Pfizer and continued working in a similar role. I am greatly involved with different cross-industry and regulatory working groups including a lead role in DIA Bayesian Scientific Working Group. I was also the industry co-chair for ASA Biopharmaceutical Section Regulatory-Industry Statistics Workshop 2018.

I would also like to take this opportunity to thank my mentors and leaders. I am especially grateful to Dr. Beat Neuenschwander, Dr. Kannan Natarajan, Dr. Pulak Ghosh, Dr. Keaven M Anderson, Dr. Paul Gallo, Dr. Neal Thomas and, Dr. Demissie Alemayehu for their continuous support and mentorship.

Cytel: What are the methods that you are advancing? 

SR: In my webinar with Cytel, I talk about a practical model-based approach in Phase I Oncology trials. Oncology Phase I designs are challenging due to the limited sample size. The current medical practice uses sub-optimal algorithmic designs to determine the maximum tolerated dose and optimal dose for patients. Despite extensive statistical methodological research in this area, model-based approaches are rarely used in clinical trials due to the lack of practical aspects. The goal of this webinar is to share some practical experiences that can help successful implementation of a model-based approach in Phase I trials.

Apart from Phase I design, my research work includes the use of historical data and trial external information in design and analysis of clinical trials. These innovative approaches allow trials with smaller sample size or with unequal randomization (more subjects on treatment than control), which is ethically and practically appealing. In recent years, use of external information in design and analysis of clinical trials have generated extensive discussions in the literature and regulatory guidance. The use of external data in trials are currently used in earlier phases of drug development and occasionally in phase III trials. These techniques can also help to develop efficient designs for pediatric trials by borrowing information from adult patients.

Recently developed Immuno-Oncology drugs bring new hope to cancer patients. These drugs help a body's immune system to fight cancer. However, development of immunotherapy poses some unique challenges in the development program. A major challenge is the violation of constant treatment effect which is a major assumption of the traditional statistical methods applied in this area. This is referred to as a non-proportional hazard problem. Estimation of treatment effect in the case of non-proportional hazards is a major challenge for immuno-oncology clinical trials. In collaboration with a cross pharma working group, we have developed innovative hypothesis testing and estimation procedures for design and analysis of clinical trial when a non-proportional hazard is present. This research is widely discussed by scientists from academia, industry and Food and Drug Administration (FDA) in different scientific forums.

My research work also includes a wide range of applications of Bayesian model-based approach and survival analysis, that helps to facilitate better quantitative decision-making and advancement of novel therapy.  Some highlights include efficient design of platform trials, evidence synthesis, and use of multi-state modeling. All my research is motivated by the real-life problems arising in a drug development process and has the objective of  helping practitioners in clinical research.

Cytel: How did you get excited about Bayesian model-based approaches?

SR: There are plenty of great philosophical reasons to focus on Bayesian statistics but for me, my interest grows in this area with practical needs in drug development. As a statistician, I see the need of advanced quantitative techniques to combine all available information for better decision making. It is crucial especially at early stage of development when information is sparse and fragmented. Bayesian model-based approach is a natural choice for me to solve this problem. Bayesian statistics allow us to make probabilistic inference on the parameter of interest which is missing in a traditional frequentist approach. Apart from the philosophical issues, Bayesian analysis provides a practical and intuitive tool for interpretation of study results and risk evaluation of clinical hypotheses. This is particularly useful for non-statisticians. The “prior distributions” is a natural way to incorporate all information available at the beginning of a clinical experiment as I mentioned earlier. Use of Bayesian statistics enhances the model-based approach further with modern advanced computational techniques and intuitive inference of the model parameters. Moreover, Bayesian methods provide a natural way of future trial prediction via “predictive distribution” which is useful to sponsors for further planning.

Cytel: Have there been others who needed persuading and how did you persuade them?

SR: During my career as a statistical methodologist, I faced several situations where a clinical team had to be persuaded to use a new innovative method. The main reason is that many people find implementing new and innovative things to be risky. My approach for such teams is to have an open dialogue to explain the risk associated with the traditional approaches and the need for new approaches. I also see the direct benefit of presenting case studies, good and interactive software, and simulations of trial and decision-making process. Discussions around regulatory impact also help a lot. It is important that both statistician and non-statisticians are onboard with these innovative ideas. I have both success and failure stories depending upon the situation, which shows that we still have a lot of work to do.

Cytel: How should these methods be used given the industry’s current challenges?

SR: In the current environment, the new drugs are targeting specific patient population. Therefore, recruiting an adequate number of patients for each trial is a great challenge. Bayesian model-based approaches provide a formal mathematical method for combining external information with current information at the design stage, during the conduct of the trial, and at the time of analysis. This allows efficient design, lower trial cost, better quantitative decision-making and faster drug development. We are in the right direction now, but it still needs work.

Cytel: Where can a beginner learn more?

SR: There is a wide range of materials including books, online courses and webinars that are easily available now. Experts from academia, industry and regulatory agencies are involved in developing these materials. Young statistician can greatly learn from them. Also, organizations like American Statistical Associations offer mentorship programs where the beginners can get good mentors who can guide and help them to avail the right resources. Such mentorship programs are often very effective.


 

Webinar details:

Practical Model-based Approaches for Phase I Oncology Trials

Phase I Oncology trials used dose-escalation to determine the maximum tolerated dose for further study. Modern model-based techniques help to acquire knowledge about the chemistry of the new drug and select appropriate dose for later phases. It offers clinicians more flexibility than more traditional designs like the 3+3. Through this introduction to advanced dose escalation methods, you will first learn about the traditional 3+3 method for dose-escalation and then gain an understanding of how Bayesian logistic regression model (BLRM) techniques can improve the traditional method.

Access Replay

 

About Dr. Satrajit Roychoudhury

Satrajit_Head_shotDr. Satrajit Roychoudhury is a Senior Director and a member of Statistical Research and Innovation group in Pfizer Inc. Prior to joining, he was a member of Statistical Methodology and consulting group in Novartis. He started his career as a research statistician in Schering Plough Research Institute (now Merck Co.). He has 12+ years of extensive experience in working with different phases of clinical trial. His primary expertise includes implementation of innovative statistical methodology in clinical trial. He has co-authored several publications/book chapters in this area and provided statistical training in major conferences. His area of research includes survival analysis, use of model-based approaches and Bayesian methods in clinical trials. Satrajit was a recipient of a Young Statistical Scientist Award from the International Indian Statistical Association in 2019.

 

contact iconSubscribe back to top