Lu Tang, PhD

"Pattern-Set Mixture Models by Penalized Generalized Estimating Equations"

Abstract: Handling of missingness in longitudinal data is undertaken in many medical studies. A popular strategy is the pattern-mixture modeling approach proposed by Little (1993, 1994), which proceeds the analysis via stratification of missing data patterns. Two issues of the stratification-based strategy are (i) whether individual missing data patterns are sufficiently distinctive to warrant stratification, and (ii) sample size attrition resulted from the stratification leads to loss of statistical power. To overcome these issues, we propose a penalized generalized estimating equations approach to identifying homogeneous pattern-sets in the longitudinal data analysis with nonignorable missing data, where we allow a data-driven stratification. Specifically, we develop a fusion approach that identifies homogeneous parameters from strata-specific models, so that we achieve both adequate stratification and larger sample sizes in individual strata. The proposed approach also allows to assess whether or not, if so how many, strata are needed to build the pattern-mixture model. The method is evaluated by numerical studies and applied to data from a psychiatric study.

Lu Tang joined the Department of Biostatistics at University of Pittsburgh as an assistant professor in August, 2018, after receiving his PhD in biostatistics from the University of Michigan. He received his bachelor's degree in mathematics and master's degree in statistics from the University of Virginia in 2012 and 2013, respectively. He is broadly interested in statistical machine learning and methods for high dimensional data, with the goals of inference, prediction and grouping. His current research focuses on fusion learning and distributed computing for detection of heterogeneous subpopulations or differential treatment effects in large scale data analyses. He also develops methods and tools for analyzing high-dimensional metabolomics data, accelerometer data and epigenetics data.