Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Usage of surrogate endpoints in the design and analysis of clinical trials

Abberbock, Judah (2017) Usage of surrogate endpoints in the design and analysis of clinical trials. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

Submitted Version

Download (853kB) | Preview


There has been a shift in the conduct of early-stage breast cancer trials in recent years from long adjuvant trials with overall or disease-free survival as the efficacy endpoint to shorter neoadjuvant trials with pathological complete response (pCR), a binary marker, at time of surgery as the endpoint. The Food and Drug Administration (FDA) currently embraces this transition and deems evidence in pCR improvement sufficient for drug approval on condition that long-term data are collected to eventually show efficacy in survival. Incorporating data on pCR in the design and analysis of such a trial is therefore of public health interest. Here, we propose one method to assess the power and sample size of such a trial with using observed neoadjuvant data and another method to estimate certain causal treatment effects on survival conditional on pCR. In the first part, we propose an exponential mixture model for survival time with parameters for the response rates and an estimated benefit in survival from achieving response. Under a fixed sample size, we obtain the empirical power through simulations from the proposed mixture model. We also propose a more efficient method than the empirical approach by applying an estimated average hazard ratio to the Schoenfeld formula. The performance of our methods is assessed via simulation studies. Data from two neoadjuvant cancer clinical trials are used to illustrate these methods. Second, we propose a method under the principal stratification framework to estimate the causal effect of treatment on a binary outcome, conditional on a post-treatment binary response marker in randomized controlled clinical trials. Specifically, we estimate the treatment effect among those who would achieve response if given the treatment. We are able to identify this causal effect under two assumptions. First, we model the counterfactual probability of achieving response under treatment given baseline clinical markers and the outcome. Second, we assume a monotonicity condition: a patient who responds under control would respond under treatment as well. We compared the performance of proposed method with other standard approaches in simulation studies. Data from a neoadjuvant breast cancer clinical trial are used to demonstrate the proposed method.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Abberbock, Judahjua23@pitt.edujua23
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairTang,
Committee MemberAnderson,
Committee MemberDing,
Committee MemberRastogi,
Date: 25 September 2017
Date Type: Publication
Defense Date: 17 July 2017
Approval Date: 25 September 2017
Submission Date: 17 July 2017
Access Restriction: 2 year -- Restrict access to University of Pittsburgh for a period of 2 years.
Number of Pages: 82
Institution: University of Pittsburgh
Schools and Programs: School of Public Health > Biostatistics
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: binary response marker, survival, power, sample size, neoadjuvant, cancer clinical trials, principal stratification, causal inference
Date Deposited: 25 Sep 2017 15:04
Last Modified: 01 Sep 2019 05:15


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item