Each year, Brown’s Undergraduate Statistics Concentrators present Capstone projects to the Department of Biostatistics faculty, students and visitors. This year, 4 graduating statistics undergraduate concentrators presented their Capstone work on April 29th. This opportunity for graduating students to present and discuss their Capstone projects which meets a graduation requirement and also qualifies each candidate for the annual Undergraduate Statistics Concentration Capstone Poster Award of Excellence which is announced at our annual Biostatistics Commencement Celebration.
Author: Tova Ibbotson
Poster Title: An Introduction to the Sequential Cox Approach for Estimating Survival under Time-varying Covariates
Abstract:Target trial emulation is a common approach to analyzing observational studies. Misspecification of the target trial in analysis can result in imprecise effects estimates, and in some cases falsely conclude a statistically significant effect. Clinical researchers without an advanced statistical background may not correctly emulate the target trial when time-varying covariates are present. This work introduces the sequential Cox approach as an intuitive conceptualization of target trial emulation in the presence of time-varying covariates. Treating the observational data as a sequence of nested “trials” allows for the pooling of the “trials” into a single Cox proportional hazards model. The pooled model is able to consider time-varying treatment and adjust for time-dependent confounding. Thus, the approach may enable clinical researchers with limited statistical resources to appropriately include time-varying covariates in their studies. This work also simulates data to demonstrate the effect of disregarding time-varying covariates in survival analysis. The sequential Cox approach offers an intuitive outline for clinical researchers to emulate intention-to-treat analyses with observational data.
After graduation, Tova will be entering Brown’s Biostatistics Master’s Graduate Program
Author: Samuel Rhee
Poster Title: Simulating trials to resample unbalanced data
Abstract: Unbalanced datasets are those with a target class with a significantly higher sample size than others. Predictive models generated from such datasets can be unreliable and lack precision in respect to the minority classes. To create more accurate models, one can either undersample or oversample. While oversampling generally provides more robust data, it can be costly or time-intensive and may not be feasible within a study’s structure. To work around this limitation, this project sought to replicate an oversampling by randomly generating new trials as new data points. I then compared predictive models from the oversampled and original datasets and used AUC analysis to compare predictive strength.
Author: Joseph Vayalumkal
Poster Title: Physical Activity Intervention, Latent Classes of Heart Rate Responses, and Biological Health Outcomes
Author: Jessica Wang
Poster Title: Trend Analysis of the Financial and Quality Performance of Accountable Care Organizations 2014 through 2020
Abstract: Since receiving federal recognition in the Affordable Care Act (ACA) of 2010, how well have Accountable Care Organizations (ACOs) achieved the “Triple Aim” - better health, improved patient experiences, and lower costs? In this work, we use publicly available data from CMS to investigate financial and quality performance of ACOs from 2014 (near inception) to 2020. Using a hierarchical multilevel linear model nested on ACO age in months, we study the relationship between different characteristics of ACOs, such as risk model and level of Electronic Health Record (EHR) integration into provider care, and the two primary outcomes of interest, quality scores and earned shared savings per beneficiary. We found that older ACOs are positively correlated with slightly higher earned shared savings per beneficiary, but slightly lower quality scores. Despite risk model data only being available for 2019-2020, we also found a significant difference in the means between one-sided and two-sided risk models in both years, with one-sided risk models having greater shared savings in 2019 but two-sided risk models having greater shared savings in 2020. Finally, we found insignificant evidence to support the idea that greater EHR integration is associated with higher earned shared savings and quality scores, indicating that future studies should consider looking at new variables that might contribute to ACO success. Other study variables had insignificant or otherwise negligible effects on the two outcome variables.
After graduation, Jessica will be working in New York City at NERA Economic Consulting.