Novel Assessments of the Health Impacts of Tropical Cyclones

In the United States, tropical cyclones, such as hurricanes and tropical storms, have a devastating impact on society. However, beyond some limited studies, there remains a critical research gap in understanding the full extent of the impact of tropical cyclones on health. The objective of this K99/R00 application is to fill this research gap with several novel assessments of the health impacts of tropical cyclones. To be able to fulfil this objective, this K99/R00 application is interdisciplinary, involving the collaboration of experts in environmental epidemiology, exposure assessment, Bayesian statistics, machine learning, computer vision, and social epidemiology. The K99 phase is designed to augment the candidate’s prior research experience through coursework, mentorship, and directed readings, with specific training in (1) disaster epidemiology and exposure assessment; (2) advanced Bayesian statistics methodology; (3) machine learning and computer vision for public health; and (4) social epidemiology in a disaster and public health context. The skills gained during this award are critical to the candidate’s long-term goal to become a leading and methodologically strong environmental epidemiologist who conducts rigorous large-scale research that contributes to society’s understanding of tropical cyclones and other environmental hazards to help inform policies in the United States and worldwide. The proposed project will draw on rich data sources on hospitalization (Medicare and Medicaid cohorts); death (National Center for Health Statistics); tropical cyclone exposure; and satellite- and ground-based imagery, all of which span several recent decades and cover all of the United States exposed to tropical cyclones. Aim 1 (K99 phase) will improve and harmonize estimation of excess hospitalizations and deaths after each named hurricane by (a) applying an ensemble of Bayesian models to hospitalization and mortality data to estimate weekly hospitalization and deaths rates that would have been expected had hurricane exposure not occurred; then (b) comparing the actual historical hospitalization and death rates to calculate excess hospitalizations and deaths. Aim 2 (R00 phase) will (a) determine the impact of repeated tropical cyclone exposure on chronic health outcomes by analyzing the association between tropical cyclone exposure and monthly hospitalizations or deaths by applying Bayesian spatio-temporal hazard models; then (b) accurately forecast health impacts by using results. Aim 3 (R00 phase) will characterize how physical neighborhood features explain differences in health impacts of tropical cyclones by (a) utilizing machine learning and computer vision techniques to identify various physical neighborhood features in tropical cyclone-exposed areas using satellite and street-level imagery; then (b) converting features into metrics in health models to investigate if and how health impacts of tropical cyclones vary by those metrics.