In March, students in the Department of Biostatistics Master of Science Capstone Program at the University of Washington presented their final capstone projects to project sponsors, program faculty, and peers.
The capstone is a culminating project, which gives students a chance to apply what they’ve learned in the program to a real-world problem in public health.
For 2024, teams engaged with healthcare organizations focusing on improving patient care, and a Washington state government organization engaged in optimizing healthcare planning and resource allocation.
”As educators, it’s always rewarding to see our students connect what they have learned in the classroom to real-life challenges involving biomedical or public health data,” said Kathleen Kerr, MS Capstone Program director and professor of biostatistics.
Lloyd Mancl is an adjunct research associate professor of biostatistics who works with the MS Capstone Program.
“One thing I love is watching their journey. I get them at the end of autumn quarter when they have these grandiose plans of what they’re going to do. Then reality hits and it’s amazing how they pivot and figure out how to solve these difficult problems. It’s really rewarding to watch them rise to the challenge, work hard towards an answer, and present what they learned,” said Mancl.
2024 Capstone Projects
Sponsor: University of Washington Department of Biomedical Informatics and Medical Education
Title: Piloting a Machine Learning Algorithm to Predict Mild Cognitive Impairment in Outpatient Clinical Visits
Team members: Yunge Li, Winnie Mei, Ruilin Zhou
Sponsor contact: Sean Mooney, PhD
Team statement: We collaborated with Dr. Sean Mooney and Peter Ju from the University of Washington Department of Biomedical Informatics and Medical Education in the University of Washington School of Medicine to develop machine learning models aimed at early prediction of mild cognitive impairment (MCI). MCI is the initial phase of cognitive decline, often preceding the onset of dementia and Alzheimer's disease. Detecting MCI at an early stage can significantly enhance patient outcomes. We used data from UW Medicine's electronic health records (EHR), including demographic details, medical histories, and prescription records. We applied various models including logistic regression, decision trees, gradient boosting, and support vector machines. Finally, we concluded the project by evaluating model performances to ascertain their efficacy. The best-performing model will serve as an early screening tool empowering clinicians to identify potential MCI patients.
Sponsor: Neighborcare Health
Title: Identifying disparities in outcomes and treatment of hypertension in community health center patients
Team Members: Junhyoun Sung and Weiqiao Shen
Sponsor contacts:
Rachel Wang Martínez, MHA, BSN, RN-BC
Alyssa Caucci, MHA
Our project with Neighborcare Health aimed to address the challenge of hypertension, a leading cause of mortality in the US, by analyzing the impact of patient and healthcare provider characteristics on blood pressure control. Through advanced statistical analyses, including the Chi-square test, Fisher's exact test, and multivariate regression, we identified disparities in blood pressure management across the diverse demographics and provider characteristics within the Seattle area's communities. Our findings enable Neighborcare Health to tailor healthcare strategies more effectively, enhancing medical care delivery and promoting health equity. This project not only bridges the gap in the organization's analytical capabilities but also contributes to the broader public health effort to combat hypertension and its complications, underscoring the importance of culturally competent care in diverse populations.
Sponsor: Country Doctor Community Health Centers
Title: An examination of well-child visits and vaccine coverage in a Seattle Federally Qualified Health Center
Team: Fernanda Montoya, Matthew Wong, Sujing Zhang
Sponsor Contact: Valerie J. Rock, MD, MPH
Team statement: Pediatric preventative health care addresses not only the development and well-being of a child, but also concerns the health of a population. Disruptions to preventative health care can have serious ramifications on both individual and population health. It is also known that these disruptions are not felt equally across a population, often with those of an underserved background most impacted. This project focuses on describing and identifying how both well-child visit attendance and vaccine coverage differs between subpopulations of interest. Our analyses utilized modified Poisson regression to estimate prevalence ratios describing the differences in both visit attendance and vaccine uptake for various subpopulations of patients. This work is a first step toward addressing disparities in the health care setting and identifying populations that may benefit from different approaches to well-child visits and vaccines.
Sponsor: Public Health – Seattle & King County
Title: Estimating parcel level population in the Puget Sound region
Team: Erika Fox, Kaizhi Lu, and Gavin Pierce
Sponsor Contact: Daniel Casey, PhD
Team statement: In collaboration with Public Health - Seattle & King County, we developed a novel model to produce parcel-level population estimates across King County, aiming to enhance resource allocation, disease surveillance, and healthcare planning. Leveraging American Community Survey data, our approach utilized multiple linear and Poisson regression models, refined through cross-validation and machine learning techniques, aiming to improve accuracy over existing prediction methods. This advanced modeling technique, validated against Census tract-level data, provides PHSKC with more precise population metrics, crucial for offering effective public health management to the diverse population of the greater King County area.
