This is for a 12 week full time internship from June-Aug 2021
In the development of disease modifying therapies for degenerative diseases, efficacy of a new treatment is typically evaluated in terms of longitudinal changes in outcomes measured by clinical assessment scales. However, patients in a severe stage of the disease may die prior to being assessed on the primary outcome. For example, clinical trials for amyotrophic lateral sclerosis (ALS) typically use decline in ALS functional rating scale (ALSFRS) after 6-12 months as a key measure of efficacy, but due the rapid degenerative nature of the disease, some deaths during the study may be expected.
In this situation, there are several possible approaches to account for mortality in the analysis of efficacy. For example, traditional mixed modeling approaches (Fitzmaurice, et al. 2011) may consider death simply as a triggering event for missing data, with the statistical analysis implicitly or explicitly imputing future unobserved values under an assumed missing data mechanism. Other approaches, such as joint rank the test (Finkelstein and Schoenfeld, 1999; Ramchandani, et al.) or trimmed mean analysis (Permutt and Li, 2017), may consider death to be a clinically meaningful outcome that is worse than the worst possible outcome on the clinical scale, and the statistical analysis would then account for this composite ranking accordingly. Still other methods, such as those based on principal stratification (Frangakis and Rubin, 2002), may restrict inference to a subpopulation that excludes subjects based on death occurring as a potential outcome.
Guidance is needed to help statistical teams develop an appropriate analysis strategy that targets the relevant efficacy estimand in the specific clinical context, while optimally balancing statistical performance with plausibility of modeling assumptions. This project will evaluate various methods of accounting for mortality jointly with a longitudinal efficacy outcome, considering both statistical performance and how the methods fit into regulatory guidance on estimands. The project will focus on methods most likely to be of practical use in drug development.
The project will require the intern to read and synthesize key methodology papers and efficiently develop simulation code in R and explore results in a systematic way. Any new code should be documented and packaged in a user-friendly way, to encourage broader use in the department. A final technical write-up and presentation will be expected. The intern's responsibilities may include:
* Review selected literature on the joint rank test, trimmed mean, and principal stratification methods, including simulation studies evaluating or comparing these approaches as well as papers discussing causal estimand frameworks for these methods.
* Perform comprehensive simulations comparing performance characteristics of the trimmed mean approach and the joint rank test. Furthermore, compare each of these methods to mixed modeling approaches, building on work done by Healy and Schoenfeld (2012).
* Investigate available simulation tools for generating longitudinal data jointly with a survival outcome. Summarize existing tools, and develop and package new simulation code in R as needed to fill major gaps.
* Review a selection of estimation procedures for survivor average causal effects implement them in a simulation study. Compare results to those obtained from the joint rank test, trimmed mean method, or mixed modeling approaches.
To participate in the (COMPANY NAME) Internship Program, students must meet the following eligibility criteria:
* Legal authorization to work in the U.S.
* Grade point average of 3.2 or higher preferred
* At least 18 years of age prior to the scheduled start date
* Be currently enrolled in an accredited college or university
Enrolled in a PhD program.
All your information will be kept confidential according to EEO guidelines
|Posted on:||13 Jan 2021|
|Type of job:||Internship|
|Job duration:||3 months|