Perelman School of Medicine Events Calendar

This comprehensive institute is designed to develop skills that are useful in qualitative and mixed methods research, including:

1. Structuring and conducting an interview
2. Selecting an appropriate sample
3. Organizing and facilitating a focus group
4. Managing, coding and analyzing data
5. Publishing your work
6. Use and application of NVivo software. Advanced NVivo training will also be available.

Rates:

General Registration (3 days): $850
Student rate (3 days): $450
Group rate (3 days, 3 people or more - contact lahall@exchange.upenn.edu to register): $450
Lectures-only (2 days): $550

Abstract: In many circumstances, medical studies evaluating the efficacy of an intervention need either long follow-up periods or expensive or invasive procedures to obtain the primary outcome. This motivates the considerable attention to surrogate evaluation in recent years, which aims to use alternative measures ("surrogate marker(s)") in lieu of the primary outcome to evaluate the efficacy of an intervention. However, conventional surrogate evaluation methods fail to provide a causal interpretation, as surrogate markers that are conditional on in regression, are post-randomization variables. Principal surrogacy, defined based on the concept of principal stratification, overcomes such shortcomings. The current literature of principal surrogacy focuses on normally distributed continuous primary outcomes or binary outcomes. We propose a shared gamma frailty proportional hazard causal model to study principal surrogacy for time-to-event primary outcomes. The proposed model is constructed under the potential outcome framework with principal stratification approach, and a gamma frailty model is used to correlates the potential outcomes of an individual under different treatment arms. With the proposed model, we define the principal hazard ratio, expected associative effect and expected dissociative effect to evaluate principal surrogacy. Because of the complicated missing data structure, we adopt a Bayesian estimation method using Markov chain Monte Carlo algorithm. We use simulations to study the repeated sampling property of the proposed model. We illustrate the proposed model and estimation method with a randomized clinical trial of colorectal cancer.