.png)
Research & Impact
I have completed impactful projects in my statistics career, including a predictive model for patient outcomes and a data visualization dashboard for healthcare analytics. These showcase my analytical skills and ability to derive insights from complex data. I am well-equipped to contribute to data science, healthcare analytics, and consulting.
01
Predicting SLE Flare Risk After Steroid Withdrawal
Domain: Clinical Research | Rheumatology | Predictive Modeling
Timeline: Fall 2024 – Spring 2025
Tools: R, GLMM, Nonlinear Splines, Longitudinal Analysis
The Challenge
Glucocorticoids are a cornerstone of lupus (SLE) management, but withdrawing treatment carries a risk of disease flare. The relationship between patient factors—like age, BMI, and mental health—and post-withdrawal flare risk is complex and poorly understood, limiting personalized care strategies.
The Approach & Insight
Using longitudinal data from the FORWARD Lupus Registry (1,085 patients), I developed and compared predictive models. A Generalized Linear Mixed Model (GLMM) with nonlinear splines significantly outperformed a standard linear model, uncovering critical, non-obvious risk patterns. The analysis revealed a U-shaped relationship with age (highest risk for patients under 50 and over 70) and an inverse relationship with BMI, where lower weight was associated with greater flare risk.
The Impact
The nonlinear model achieved an AUC of 0.913, providing a superior tool for identifying high-risk patients. The findings advocate for targeted monitoring of younger patients, those with lower BMI, and individuals with moderate depression symptoms. This work directly supports a shift toward more personalized treatment plans in autoimmune care.
02
A Predictive Model for Senior Institutionalization Risk
Domain: Public Health | Geriatrics | Machine Learning
Timeline: Spring 2025
Tools: R, GLMM, XGBoost, Cluster-based Undersampling, NHATS Data
The Challenge
Institutional care for seniors is a major personal and financial burden. Early identification of at-risk individuals through scalable means is a significant public health challenge, especially using rare-event longitudinal data where only ~5% of the population transitions.
The Approach & Insight
Analyzing 9,844 seniors from the National Health and Aging Trends Study (NHATS), I engineered a two-stage framework to address severe class imbalance. First, GLMMs modeled individual trajectories of social activity decline. Second, these trajectories were fed into XGBoost to predict institutionalization. The key finding: a sharp decline in social participation more than doubled the risk (OR=2.3) of transitioning to institutional care.
The Impact
The final model achieved an 81% Precision-Recall AUC, a 40% improvement over traditional logistic regression. From this, I derived a simple 3-question screening tool capable of identifying 80% of high-risk individuals. Projected to scale, this tool could help prevent unnecessary institutionalization, saving an estimated $3.8M annually per 10,000 seniors through early, low-cost interventions.