Accounting for subject-level heterogeneity in sieve analysis of vaccine efficacy

by Jason Shao

Institution: University of Washington
Year: 2018
Keywords: Frailty models; Mixture models; Sieve analysis; Survival analysis; Vaccine trials; Biostatistics; Biostatistics
Posted: 02/01/2018
Record ID: 2212835
Full text PDF: http://hdl.handle.net/1773/40854


In randomized trials of preventative vaccines, sieve analysis tests whether vaccine efficacy differs by a characteristic of the disease endpoint. These methods often assume a leaky model, in which treatment proportionally reduces the hazard of each disease type homogeneously in all subjects. Significant biases can occur in estimation and testing when this assumption does not hold. To allow for unobserved heterogeneity in participant response to vaccination, we propose a frailty mixture model for sieve analysis which incorporates unobserved, subject- level, random effects into a competing risks survival analysis framework. We show that parameters in the model can be straightforwardly and reliably estimated using standard numeric optimization methods. In simulation studies, our approach performs favorably to existing methods in cases where the leaky vaccine assumption is not appropriate. Finally, we implement our method on existing clinical trial datasets and discuss the implications of our findings for the design and interpretation of vaccine efficacy studies.Advisors/Committee Members: Edlefsen, Paul T. (advisor).