Projects

Substance Abuse Data Analysis Project

This independent data analysis project aims to answer the fundamental question - who needs aid within the current context of the opioid epidemic that is sweeping the United States? Substance abuse is an important public health issue, and one of the first steps to addressing this increasingly prevalent issue is determining which populations are most at-risk and need more support. Through statistical modeling and inference, the final conclusion is that two communities stand out as requiring more support:

• people with co-occurring mental and substance abuse issues, and
• houseless people

To read the project in entirety, please reference my Github repository!

Mock Randomized Clinical Trial Study Protocol

For my clinical trials course at UW (BIOST 524), the culminating course project was to work in teams to create a study protocol for a mock randomized clinical trial. My group designed a trial studying restrictive versus liberal red blood cell transfusion strategies in pediatric acute respiratory distress syndrome. The clinicians in the group did the bulk of the clinical background, and myself along with another fellow biostatistics student did the bulk of the statistical methods work.

To read the project in entirety, please click here!

Comparing Type I Error Rate Control Methods

The goal of this independent undergraduate statistical research project is to compare ten different methods to control either the Family-Wise Error Rate, False Discovery Rate, or False Discovery Exceedance. I conducted a simulation study using the R package simsalapar to try to replicate the results described in a previously published paper written by Farcomeni et al. (2008). I used various error control packages in R to carry out the analyses, and I wrote my own R method for the Step-Down Lehmann-Romano adjustment method. I also presented the results at the McGill SUS Academic Week poster fair. It was impossible to pick one best adjustment method due to researchers varying in conservativeness, and hence guidance is given for which methods work best in different applied settings. (Adviser: Dr. Russell Steele)

For a more detailed explanation of the methods used, and to read the final paper, please reference my Github repository!