Monday, Aug 4: 8:30 AM - 10:20 AM
4037
Contributed Papers
Music City Center
Room: CC-201B
Main Sponsor
Section on Statistics in Epidemiology
Presentations
Population health rankings can be a catalyst for the improvement of health by drawing attention to areas in need of relative improvement and summarizing complex information in a manner understood by almost everyone. We explore data-informed grouping (cluster analysis) as an easier-to-understand, empirical technique to account for rank imprecision that can be effectively communicated both numerically and visually. We consider k-means clustering as a simple approach to identify natural and meaningful groupings and gaps in health outcomes using a bias-corrected Wasserstein (earth mover's) distance to select the number of groups. Application to the 2022 County Health Rankings and Roadmaps identified 30 groupings (clusters) with sizes ranging from 9 to 184 counties. The method helped address many of the issues that arise from providing rank estimates alone. Public health practitioners can use this information to understand uncertainty in ranks, visualize distances between county ranks, have context around which counties are not meaningfully different from one another, and compare county performance to peer counties.
Keywords
Ranking
Clustering
Small area estimation
This study addresses the importance of analyzing multiple time-to-event outcomes in competing risk
settings, particularly for patients with rare disease. When multiple risk factors are present for the event
of interest, competing risk models yield more accurate insights than traditional survival models, as
overlooking these risks can result in biased conclusions. Data from the NCI
SEER-18 database (2000-2018) was utilized, focusing on patients diagnosed with soft tissue cancers,
including cardiac sarcoma. The study computed cumulative incidence and mortality risks while applying
causal inference techniques to understand the impact of treatments on patient survival.
Despite the limited sample size, the study findings support the model-based prediction of
treatment effects on rare malignancies. Surgical intervention and radiotherapy were linked to reduced
cause-specific mortality, while chemotherapy and other therapies did not show significant associations
with overall survival, highlighting the need for effective therapeutic strategies in managing cardiac
cancer.
Keywords
Cause-specific hazard
Causal inference
Machine learning
Cardiac cancer
SEER.
Structural Equation Modeling (SEM) offers several advantages over regression, such as analyzing multiple relationships, incorporating latent variables, assessing model fit, handling measurement error, and flexibility. SEM is widely used in clinical and epidemiology studies. Music therapy is used to improve cancer patients' well-being by reducing negative emotions, aiding stress management, and enhancing emotional expression. This study investigated music intervention during chemotherapy infusion using SEM. 750 cancer patients receiving outpatient chemotherapy participated in an open-label, multisite, permuted block randomized trial. Patients were assigned to the intervention (listening to a single genre for up to 60 minutes) or a control group (no music). Self-reported pain, positive and negative mood, and distress were modeled as latent variables. Adjusted for covariates, music significantly improved positive mood and reduced negative mood but had no significant effect on pain or distress. Pain, negative mood, and distress were correlated, while none correlated with positive mood. These findings highlight the psychological benefits of music therapy during chemotherapy infusion.
Keywords
Structural Equation Modeling
music therapy
chemotherapy
The relationship between body composition and bladder cancer outcomes is complex. Higher Body Mass Index seems to predispose to the development of bladder cancer, though the impact on survival is more convoluted. We sought to study this relationship in a cohort of older patients with bladder cancer.
We included patients from the Surveillance, Epidemiology, and End Results-Medicare Health Outcomes Survey database with bladder cancer diagnoses. We assessed demographic, bladder cancer characteristics and treatments, BMI, overall survival, and health-related quality of life (measured by the physical and mental component summary scores and activities of daily livings). The final cohort consisted of 8013 patients with a mean age of 77.6 ± 7 years, 85.6% white and 74.8% males.
Our findings suggest a dual role of BMI in older patients with bladder cancer: higher BMI provides a survival advantage, with improved survival for those who are overweight, obese, or severely obese when compared to those in the healthy range or underweight. Overweight patients also exhibited the highest physical and mental component summary scores, indicating better quality of life. In contrast, severely obese
Keywords
Bladder Cancer
Survival Analysis
Physical and Mental Component Summary Scores
Body Mass Index
Improving novel pathogen surveillance systems is of paramount importance, as detecting infections while at low prevalence can guide interventions to prevent epidemics and pandemics. The probability of surveillance system failure can be modeled using a framework that considers the probability of detection for a single case, π, and population size. However, accounting for population structure in disease incidence and π, and for subpopulation sizes, can reduce bias in quantifying surveillance system performance. We created models that incorporate these considerations: an "endemic model", which assumes emergence of a variant of a circulating pathogen, and an "outbreak model", which assumes an outbreak of a novel virus. Surveillance system failure probability estimates were higher using the outbreak model, and negatively correlated π and probability of local outbreak, which is likely in rural areas, resulted in higher failure probabilities. These results have ramifications for policy, as overestimation of surveillance system performance may lead to a reduction in the overall level, or efficacious distribution of, resources for epidemic and pandemic prevention.
Keywords
epidemiology
surveillance
population structure
In 2024, the Centers for Disease Control and Prevention (CDC) determined that 1.63 million middle and high school students used electronic cigarettes (e-cigarettes) in the United States. However, while e-cigarettes are the most frequently used tobacco product, the CDC has stated that there has been a reduction in e-cigarette usage amongst adolescents. But has there actually been a reduction in youth e-cigarette use? To answer these questions, the NYTS datasets for 2018-2023 were used to track the number of youth e-cigarette users over time. Additionally, the self-reported smoking status of students who vaped marijuana was shown for 2018-2023. Furthermore, the reasons why individuals first used e-cigarettes were compared to the reasons why students currently used e-cigarettes. While the reasons for first use were always because a friend used them or curiosity, the top reasons for current use were poor mental health and to get high. Although use of e-cigarette products has seemed to be on the decline, more adolescents who vape marijuana have started to identify as nonsmokers. From 2018 to 2023, the percentage of marijuana vapers that identify as nonsmokers increased by almost 10%. These results indicate that e-cigarette use may not be decreasing and that further interventions need to bring more awareness to the health impacts of e-cigarettes and emphasize student mental well-being.
Keywords
Vaping
adolescents
e-cigarette smoking
mental health
Statistical methods in causal inference often assume the positivity condition that every individual has some chance of receiving any treatment level, regardless of covariates. However, this assumption could be violated in observational studies with continuous treatments. In this talk, we introduce a novel integral estimator for dose-response curve without requiring the positivity condition. Our approach estimates the derivative of the treatment effect at each observed data point and integrates it to the treatment level of interest, addressing bias stemming from violations of the positivity condition. The validity of our approach relies on a weaker assumption, satisfied by additive confounding models in spatial confounding settings. We further propose a fast and reliable numerical recipe for computing our integral estimator in practice and derive its asymptotic properties. To enable valid inference on the dose-response curve and its derivative, we use the nonparametric bootstrap and establish its consistency. The performances of our proposed estimators are validated through an application assessing the impact of air pollution (PM2.5 exposure) on cardiovascular mortality rates.
Keywords
Causal inference
Dose-Response Curve
Positivity
Kernel Smoothing