Collaborative HIV/AIDS and Global Health Research
I have been fortunate to work with researchers throughout the world on many interesting studies. Since 2006, I have been the lead statistician for the Caribbean, Central and South American network (CCASAnet) of the International epidemiology Databases to Evaluate AIDS (IeDEA). I am currently the Associate Director of the Data Science Core (DSC) of the Tennessee Center for AIDS Research (TN-CFAR); I served as Director of the TN-CFAR DSC from 2015-2024 and before that I was the lead statistician for the Vanderbilt-Meharry CFAR. I have also supervised biostatistical support for the Vanderbilt Institute for Global Health (VIGH) since 2006. Since 2022, I have led the Vanderbilt-Nigeria Biostatistics Training Program, which aims to develop biostatistics leaders in northern Nigeria. In these roles, I have collaborated with many researchers on a wide variety of topics. Our team takes pride in using appropriate and modern statistical methods, in making findings interpretable, and sharing analysis code so that results are quasi-reproducible (link). These collaborations have also led to many fascinating statistical problems that have motivated my statistical methods research.
More Information
Ordinal / Rank-based Analysis Methods
Motivated by a simple question we encountered in our collaborative research, my colleague, Chun Li and I developed a new statistical method to test for association between two ordered categorical variables while adjusting for covariates. In the process, we developed a new residual for ordinal outcomes, which we have since discovered to be useful for many other outcome types. This work has opened new directions for the analysis of continuous data using ordinal models, which are rank-based and make fewer assumptions than traditional approaches. This research has been funded by an R01 from the National Institutes of Health since 2011.
More Information
Methods to Address Error-Prone Data
Audits in CCASAnet suggested that data from many of our sites had high error rates. We proposed to use information learned from the audits to correct otherwise biased estimates. This required developing new statistical methods that extend the measurement error literature to handle situations where both covariates and outcomes are measured with error, and the magnitude of these errors are correlated. We have developed and applied a variety of methods (e.g., generalized raking, multiple imputation, and sieve maximum likelihood estimation) for a variety of outcomes (e.g., time-to-event, binary, continuous). We have also studied optimal designs for selecting an audit (i.e., validation) subsample. This research is directly applicable to validating electronic health records data. Starting in 2017, we received funding from PCORI and the NIH to do this research; Pam Shaw (co-principal investigator) and I were given a MERIT award (R37) from the NIH for this research.
More Information
Causal Inference Methods
My early methods research was focused on developing statistical methods to compare clinical trial outcomes only measured in subsets chosen after randomization. These methods were particularly motivated by HIV vaccine trials, where many outcomes (e.g., set point viral load, time from infection diagnosis to AIDS) only exist in participants who become infected. These methods have also been applied to other settings (e.g., truncation by death and cancer studies). My work in this area focused on sensitivity analyses within the principal stratification framework – explicitly specifying assumptions to identify causal estimands and then relaxing assumptions through the use of sensitivity parameters elicited by subject-matter experts. As I have dealt more with data from large observational cohorts, my causal inference research has dealt more with time-varying confounding (e.g., dynamic marginal structural models), difference-in-differences methods, and the intersection between causal and ordinal methods.
More Information
Other Topics
Interesting collaborative work has brought up issues that have led to a paper or two on specific topics. A few of these papers are here.