Welcome

I am a Statistics PhD student at the London School of Economics, interested in using Bayesian methods, machine learning, and computational tools to solve complex problems in risk assessment, toxicology, and healthcare analytics. My advisors are Dr. Sara Geneletti, Dr. Kostas Kalogeropoulos, and Dr. Francesca Panero. My research focuses on developing robust statistical models to improve decision-making in health, toxicology, and risk.

Research Interests: Bayesian inference, dose-response modeling, model averaging, Gaussian processes, probabilistic risk assessment, survival analysis
Technical Skills: R, Python, and SQL, with experience in Bayesian modeling (PyMC3, Stan) and machine learning (TensorFlow, PyTorch)
Applications: Toxicity risk assessment, drug efficacy, healthcare analytics, operations research, and optimization problems
Personal Projects: Exploring predictive modeling in areas such as thoroughbred pricing, traumatic brain injury rehabilitation, and search optimization

Current Projects

Antimony Toxicology Risk Assessment: Focused on the effects of antimony trioxide on cancer risk in animal studies, this project employs Bayesian methods to understand dose-response relationships. Currently working to apply inferences from animal studies to human occupational and standard exposure limits.
Bayesian Regression for Benchmark Dose Estimation: This project utilizes Bayesian parametric modeling to estimate risk functions in dose-response assessments. By fitting nine biologically-plausible models with semi-informative priors, the project identifies the best-fitting model using the Bayes Information Criteria (BIC). Expected outcomes include improved model selection strategies and more conservative risk estimates using BMDL.
Bayesian Model Averaging: This project investigates Bayesian model averaging (BMA) as an approach to enhance risk function estimation by averaging across parametric models based on posterior probability weights. Methods include the use of uniform priors and bridge sampling. The expected outcome is a risk function with reduced uncertainty, allowing for a more reliable and robust assessment.
Gaussian Process Regression: This research applies Gaussian Process Regression with radial basis function (RBF) kernels to estimate dose-response relationships. The method ensures monotonicity and bound constraints for biological plausibility. Expected outcomes are highly flexible, accurate dose-response curves with low computational demands, making GPR a practical choice for diverse applications.
Effect of Prior Selection on Benchmark Dose Estimations: This project investigates how changes in prior distributions impact Benchmark Dose (BMD) estimations using both simulation data and real-world antimony data. It specifically examines the effects of using priors that differ from current U.S. and E.U. government recommendations. The study explores the robustness of BMD estimation against various prior configurations. Expected outcomes suggest that as long as the chosen priors are reasonable and not excessively constraining, their impact on BMD estimations remains minimal, providing flexibility in prior selection for dose-response modeling.

Education

PhD, Statistics: London School of Economics (Present)
MSc, Statistics (Research): London School of Economics
MMSc, Global Affairs, Economics and Business: Tsinghua University, Schwarzman College
BA, Public Health: Johns Hopkins University

Skills

Statistical Methods: Bayesian inference, MCMC, optimization, survival analysis, random effects models
Programming Languages: R, Python, SQL
Applications: Probabilistic risk assessment, toxicity, drug efficacy, healthcare analytics, risk management

Tools

Bayesian Modeling: PyMC3, Stan
Machine Learning: TensorFlow, PyTorch, Scikit-learn
Optimization: Gurobi, CPLEX, PuLP

Projects

Some of my past projects have included the following:

Global Financial Tuberculosis Management: This project examined the financial outlays associated with public health interventions for tuberculosis (TB) using data from the World Health Organization (WHO). We applied regression models for predicting cost overages in TB programs and utilized classification approaches to validate regional strategies. We identified abnormalities in country-level resource allocation for TB programs, and our classification analysis suggest that regional management techniques should be reconsidered.
Exploring Mental Health Predictors in EU Countries: This report investigated the factors that predict mental health in the European Union (EU). We found that education level, social exclusion, gender tension, self-reported para-emotional assessments, and country effects across EU states were particularly impactful. Our findings indicated that while including hierarchical structures for country-level variation is logical, it does not significantly reduce unexplained variance in the models. Furthermore, factors such as gender tension and self-reported para-emotional assessments were more predictive of mental wellbeing than more complex analyses.
Dynamic Adjustment of Healthcare Resources to Reduce Asthma-related ER Visits in California: This research attempted to predict asthma-related ER visits in California, emphasizing the roles of PM2.5 concentrations and demographics. Despite exploring various models, such as spatial and spatial-temporal approaches, the most insightful was an ecological regression model that included spatial components. The study revealed consistent trends but indicated the need for a broader range of predictors and more granular temporal data.
Evaluating Claims of Ethnicity-based Discrimination in the California Department of Developmental Services (DDS): This project analyzed expenditure data from the California DDS to investigate claims of ethnicity-based discrimination against Hispanic residents. Our analysis showed that differences in expenditures between White non-Hispanic and Hispanic residents were primarily due to age distribution rather than ethnicity.
Gender and Personality Differences in Benzodiazepine Use: This study explored gender differences in benzodiazepine use in relation to personality traits, behaviors, and other substance use. Using ordinal logistic regression, we found that neuroticism was positively associated with benzodiazepine use for both genders, while higher conscientiousness, especially among women, was linked to lower use levels. Interestingly, cannabis, but not alcohol, was associated with increased benzodiazepine use, indicating polysubstance use behavior. The analysis highlighted gender differences in how personality traits and behaviors influence benzodiazepine consumption patterns.

Papers

Here are some of my academic contributions:

Bayesian Dose-Response Modeling for Toxicology Risk Assessment with Application to Antimony Trioxide
Wrobleski, T.L. Master’s Dissertation, London School of Economics and Political Science, 2024.
Measuring Hospital Contributions to Community Health
Plott, C., Wrobleski, T.L., Sharfstein, J.M., and Thornton, R.L.J. Johns Hopkins Center for Health Equity. Bloomberg American Health Initiative, 2021.
The Urgency and Challenge of Opening K-12 Schools in the Fall of 2020
Sharfstein, J.M. and Morphew, C.C. (Wrobleski, T.L. – research assistance). JAMA, 324(2):133-134, 2020.
Enhancing Community Engagement by Schools and Programs of Public Health in the United States
Levin, M.B., Bowie, J.V., Ragsdale, S.K., Gawad, A.L., Cooper, L.A., and Sharfstein, J.M. (Brahmbhatt, H., and Wrobleski, T.L. – research assistance). Annual Review of Public Health, 42(1):405–21, 2020.
Kidney Donation in China: How Exchange Mechanisms Can Meet Increasing Demand (中国的肾脏捐赠: 肾脏交换机制如何满足增长的肾源需求)
Wrobleski, T.L. Master’s Capstone, Schwarzman College, Tsinghua University, 2019.
Preventing Road Traffic Injuries in Jamaica: Gap Analysis and Recommendations
Gielen, A.C., Pollack Porter, K., Wrobleski, T.L., and Tsai, S.H.L. The Johns Hopkins Center for Injury Research and Policy, 2018. Prepared for The National Road Safety Council of Jamaica.
Review on Methods of National Goal Setting for the Reduction of Non-Communicable Diseases in Low and Middle-Income Countries
Wrobleski, T.L. and Bukhman, G. Partners In Health – NCD Synergies, 2015.

Awards and Recognition

Economic and Social Research Council (ESRC) Doctoral Training Program Studentship: Awarded competitive funding to support PhD research in statistics at the London School of Economics.
Schwarzman College, Outstanding Capstone Award: Recognized for excellence in master’s thesis at Tsinghua University.
Bloomberg Fellow: Selected as a fellow in this prestigious program for leadership potential and academic excellence in global affairs.
Phi Beta Kappa: Inducted into the nation’s oldest and most prestigious academic honor society, recognizing exceptional academic achievement.
University Honors: Graduated with distinction from Johns Hopkins University, acknowledging outstanding academic performance and commitment to excellence.
Gilman Scholarship: Awarded for academic achievement and commitment to international education, supporting studies and research in Ghana.

Personal Projects

I also enjoy building some personal projects to explore ideas and technologies. Here are a few current examples:

[Battleship-Inspired Search Optimization]: Predictive modeling and search algorithm project that leverages prior knowledge from multiple “experts” (e.g., past history, demographics, and strategy) to optimize the search for a hidden target. Built with Python, this project uses Bayesian inference to combine expert input into a dynamic probability map and implements search algorithms - such as Greedy Search, Monte Carlo Tree Search (MCTS), Reinforcement Learning, and Information Gain - to balance exploration of new areas and exploitation of high-probability regions.
[Translation & Transcription Tool]: A web app that transcribes and translates text between English and Chinese to support human interpreters. Considers context. Demo available soon.
[Thoroughbred Pricing]: Predictive model to estimate the value of thoroughbred racehorses based on factors such as pedigree, physical attributes, racing performance, and market trends. Exploring drivers of racehorses as assets to consider the earning potential and costs of the horses across life stages.
[Traumatic Brain Injury Rehabilitation]: Predictive model to assess recovery trajectories for patients with traumatic brain injuries (TBI). By integrating patient-specific variables, treatment protocols, and therapy progress data, we seek to support decision-making processes for healthcare providers in occupational and physical therapy settings. Still exploring methods such as Random Forests, Gradient Boosting Machines, and Recurrent Neural Networks to model interactions and time-series nature of the data.

About Me

Outside of research, I enjoy running, tennis, hiking, reading, and playing piano.

You can contact me from my university page.