Skip to content

Loading Events

« All Events

  • This event has passed.

Webinar: Biostatistics internship presentations

October 2, 12:00 pm to 1:00 pm

Biostatistics PhD students Chenqi “Stacey” Fu and Xi “Ada” Wang will present work done during their summer internships, which were at Vertex and Merck respectively.

Fu will present “Bayesian Shrinkage Subgroup Analysis: A Simulation Study.”

Subgroup analysis plays an essential role in the interpretation of the clinical trial findings. Consistency of treatment effect across subgroups indicate that the average treatment effect is in general applicable regardless of heterogeneity in the study population. In recent years, there are increasing concerns with classical subgroup analysis approaches. Potential issues include random high and random low estimates and large confidence intervals. To raise awareness from industry, FDA statisticians summarized different considerations when dealing with subgroup analysis in clinical trials. To catch the trend, in this summer, as an intern at Vertex, Fu conducted simulation studies to understand some important aspects of Bayesian shrinkage method for subgroup analysis and identified the critical issues that may encounter in future applications.

Wang will present “Use of Non-Concurrent Control in Umbrella Platform Studies.”

When conducting a randomized clinical trial, it is possible that the control arm data already exist in other clinical trials. For umbrella platform studies with a shared control arm and allowing experimental arms to enter at different times, in addition to historical control arm data from other clinical trials, there may be non-concurrent control data available within the same study but are not part of the concurrent control randomized with the experimental arm of interest. Borrowing information from non-randomized control, especially in the exploratory clinical trial setting (i.e. phase II), has the potential to gain efficiency and accelerate clinical development. When the exchangeability assumption holds (i.e. historical control/non-concurrent control data are sufficiently similar to the concurrent control), there can be a more accurate point estimate, increased power, reduced Type I error and increased efficiency. In the existing methods, propensity score (PS) approaches increase the similarity between historical and current trial by balancing the covariate distribution, and Bayesian approaches dynamically adjust the amount of borrow depending on the agreement between historical data and current data. Hybrid methods of PS adjustment plus Bayesian information borrowing have been proposed recently. Is the additional Bayesian borrowing necessary after the PS adjusting? Wang’s project performed simulation studies under different scenarios using binary endpoint as the outcome of interest, to compare the existing and newly proposed methods and offer practical recommendations. Finally, Wang’s presentation illustrates these approaches through a real data example.

Watch the seminars via Zoom

Participants may also join the seminar by calling 929-205-6099, meeting ID 322 043 373. If prompted, enter passcode 167660.


October 2
12:00 pm to 1:00 pm
Event Category:


Department of Public Health Sciences

Add events

Anyone affiliated with Penn State Health and Penn State College of Medicine can request access to add events to this calendar here.

Infonet calendar

Faculty, staff and students with Penn State Health network access can view selected internal events on the Infonet calendar here.