Presenters: Glenn Palmer and Christine Shen
Moderator: Aidan Gleich
Speaker: Glenn Palmer
Title: Quantifying heterogeneity in sleep apnea severity using hierarchical Bayesian modeling
Abstract: Obstructive sleep apnea (OSA) is a nighttime breathing disorder involving intervals of shallow or no breathing, that when left untreated is associated with a number of poor health outcomes including cardiovascular disease, metabolic disorders, and depression. The process of diagnosing OSA involves an overnight sleep study that collects a huge amount of physiological data; but after this data collection, severity is summarized in terms of a single overly-simplistic index. In this work, we seek to develop a Bayesian hierarchical model that better captures the heterogeneity in sleep apnea expression and severity by modeling patients' sleep dynamics jointly with their OSA event occurrences. Our inference focuses on a vector of patient-level random effects that captures variability in both the rate of OSA events and the level of disruptiveness of those events to subsequent sleep-stage dynamics. In this talk, I will present our inference goals, describe the model setup, and then show some initial data analysis results for the Apnea Positive Pressure Long-term Efficacy Study (APPLES).
Speaker: Christine Shen
Title: A Bayesian competing risk model for readmission risk among elderly patients with upper extremity fractures
Abstract: Upper extremity fractures in elderly patients pose substantial clinical and economic challenges in the U.S. Beyond the immediate impact on health and functional status of the patients, these injuries often precede further falls and fractures, leading to increased healthcare costs and burden. Identifying geographic and demographic factors associated with hospital readmissions among these patients is essential for informing preventive care strategies. Using Electronic Health Records from Duke, we analyzed readmission risk among elderly North Carolina residents with upper extremity fractures. We developed a proportional hazards Bayesian survival model which accounts for death as a competing risk event. We explored the associations between readmission risk and Social Determinants of Health (SDOH), including the Area Deprivation Index, Social Vulnerability Index, and poverty levels, adjusting for patient age, gender, race, and comorbidity scores etc. Our findings reveal that while SDOH factors were not significantly associated with readmission risk, certain demographic factors, such as smoking status and relationship status, are predictive of readmission risk. These insights highlight potential areas for targeted interventions for at-risk patient groups to reduce readmission rates.