2023 Summer Internship - Biostatistics - Immunology

By clicking the “Apply” button, I understand that my employment application process with Takeda will commence and that the information I provide in my application will be processed in line with Takeda’s Privacy Notice and Terms of Use. I further attest that all information I submit in my employment application is true to the best of my knowledge.

At Takeda, we are committed to lifelong learning.

To that end, Takeda's summer internship program blends real world experience with an extensive overview of the pharmaceutical industry. Knowledgeable mentors will provide guidance as you gain professional hands-on experience to start your career or further develop in your expertise.

The summer internship program is a full-time commitment of 12 weeks in length and offers a unique perspective into a world-class pharmaceutical company. Our internship program also provides you the opportunity to network with people at Takeda through various planned events and activities.

Topic: Borrowing historical longitudinal control data using Bayesian hierarchical mixed effect modeling

FDA’s Adaptive Designs for Clinical Trials of Drugs and Biologics Guidance demonstrates explicit borrowing of historical control via informative prior distribution to improve the efficiency of a trial [1]. In this evolving era, pharmaceutical industry and regulatory agency are embracing the use of historical control as a supplement to a concurrent control especially when a concurrent control is either unethical or difficult to enroll. A number of Bayesian adaptive designs borrowing historical control data have been proposed and implemented in early phase clinical trials [2].

Motivation:
The common strategy of borrowing historical control data mainly directs the attention to one time point data, for instance, only the data of final treatment visit is concerned. However, longitudinal repeated measures are often collected in clinical trials to examine the effect of treatment on the disease process over time. Incorporation of longitudinal data in the analysis will help obtain a more precise estimate of clinical outcome, monitor the time profile of treatment efficacy or safety, and mitigate the potential bias caused by missing data [3, 4]. The mixed model for repeated measures (MMRM) has been widely used as the primary efficacy analysis or sensitivity analysis for continuous endpoints in clinical trials. There are growing appeals for incorporating longitudinal data in the paradigm of historical borrowing designs.

Little research has investigated borrowing longitudinal historical study-level control data. In reality, subject-level historical data are usually not available. However, in most cases, the patient-level historical data are usually difficult to be attained due to confidential commercial information with conflict interest, restriction of personal privacy data sharing, and/or regulatory or ethical concerns. We propose a new method to exploit the historical longitudinal control data from the previous scientific publications provided with the summary statistics over time. The Bayesian hierarchical mixed-effects model will take the within-subject correlation into account by constructing a correlated meta-analytic-predicative (MAP) prior. Neuenschwander et al. [5] proposed the meta-analytic-predictive (MAP) approach to integrate the historical control information as an informative prior, which is then combined with the concurrent trial control data in the framework of Bayesian analysis. The model also poses the nice feature of validly handling the missing data under the missing at random (MAR)mechanism. The simulation study aims to demonstrate the operating characteristic of proposed method in various scenarios of discrepancy of historical control vs concurrent control and in different missing data patterns.

Objectives:
During the 12-week period, the intern will be guided and is expected to finish the action items below:
1. Literature review (2 weeks):
• Review FDA adaptive design guidance, Bayesian historical borrowing methods, Bayesian hierarchical model, SAS PROC MCMC user guide, R JAGS and BUGS tutorials, FDA/EMA missing data guidance etc.
• Consult Fang Chen, SAS Institute Inc. PROC MCMC SME.
2. Develop SAS PROC MCMC programs and R programs for Bayesian hierarchical modeling. (2 weeks)
3. Design and conduct the simulation study. (5 weeks)
4. Demonstrate the application of Bayesian hierarchical models in a case study if applicable

EEO Statement

Takeda is proud in its commitment to creating a diverse workforce and providing equal employment opportunities to all employees and applicants for employment without regard to race, color, religion, sex, sexual orientation, gender identity, gender expression, parental status, national origin, age, disability, citizenship status, genetic information or characteristics, marital status, status as a Vietnam era veteran, special disabled veteran, or other protected veteran in accordance with applicable federal, state and local laws, and any other characteristic protected by law.