Dalla Lana School of Public Health

Presented by the DLSPH Biostatistics Division …

The Biostatistics Seminar Series presents:

“Analysis of Time-to-Event Outcomes Using Nonprobability and Probability Samples” by Dr. Zilin Wang, Wilfrid Laurier University

Abstract: In this talk, we introduced a flexible, doubly robust framework for estimating population survival functions using both the non-probability and probability samples. This framework is particularly relevant to genetic and population health studies and other nonprobability samples that rely on large volunteer cohorts, such as biobanks or disease registries. Due to the lack of selection mechanisms, estimation from nonprobability samples may be subject to selection bias. Probability sampling is designed to generate representative samples from the population. However, probability samples, for example, large-scale complex surveys, tend to have fewer variables than nonprobability samples due to their collection methods and costs. Using probability samples to adjust for selection bias, our framework analyzes time-to-event outcomes from nonprobability samples. The method combines inverse probability weighting and outcome regression models to provide a doubly robust estimator that remains design consistent if either model is correctly specified. We consider two approaches for survival estimation: one based on pseudo-observations to reconstruct full risk sets, and another based on direct modelling of censored data. The framework accommodates non-parametric, semi-parametric, and parametric survival models, including the Kaplan-Meier estimator and Cox regression. Variance estimation is conducted using a resampling procedure that captures uncertainty from both the sampling design and model fitting. (Joint work with Ipseeta Garg and David Soave)

For Dr. Wang’s biosketch, please see https://www.wlu.ca/academics/faculties/faculty-of-science/faculty-profiles/zilin-wang/index.html