CS3c: Panel: Advanced Modeling for Spatial and Longitudinal Health and Environmental Data

The opioid epidemic has been particularly severe in Ohio, prompting significant efforts to understand its spatial patterns, mainly using available data at the county level. However, relying solely on county-level analysis can overlook crucial information relevant to localized effects. To address this, we integrate spatially misaligned data observed at the county and ZIP code levels to explore the complex interaction of five opioid-related outcomes, providing a more detailed local understanding of the opioid epidemic. We demonstrate how to map ZIP-code level data to ZIP-code Tabulation Areas (ZCTAs) and relate the county-level and ZCTA-level outcomes to a spatially correlated latent factor. The latent factor is defined on the intersection of the misaligned areal units, which provides a more granular understanding of the opioid epidemic. Furthermore, this approach allows us to identify areas with varying levels of opioid burden and reveals local regions with relatively high burden that county-level analyses might miss. Finally, we highlight the need for careful consideration when relying solely on ZIP code level data for naloxone, as it may lead to misinterpretations, particularly in rural regions. 

Dengue is one of the most prevalent mosquito-borne diseases affecting humans. While it was typically concentrated mostly in areas of the world with tropical climates, with increasing temperature, it has started making its appearance also in the United States. One of the key parameters that epidemiologists use to characterize an infectious disease is the Force of Infection (FOI), the instantaneous rate at which susceptible individuals become infected. Force of Infection is typically represented as a function of age of the individuals and of time (e.g. calendar year). Approaches to estimate FOI from serological data have been presented and discussed in the literature since the 90's. In this talk, we discuss how to estimate FOI from prevalence data. Using prevalence data from multiple areal units in coastal Ecuador, we present a Bayesian hierarchical model to estimate parish-specific FOI's, which in turn allow us to investigate how parish-level characteristics related to urbanization influence dengue's FOI. 

Idiopathic CD4 lymphocytopenia (ICL) is a rare, enigmatic syndrome characterized by CD4 lymphopenia in the absence of any known cause of immunodeficiency. This talk addresses statistical considerations in modeling a natural history ICL study, which aimed to evaluate the clinical, genetic, immunologic, and prognostic characteristics of 91 patients with ICL over an 11- year period. I will discuss statistical challenges in modeling a rich and complex dataset including the handling of longitudinal measurements via linear mixed-models to evaluate T-cell count trajectories, and logistic regression to assess the prognostic value of T-cell counts, using tertiles derived from baseline measurements. Multiple outputation was applied to the logistic regression to handle the longitudinal nature of the data, allowing us to use methods designed for independent data in a repeated-measures context. Additionally, we compared the observed mortality and cancer prevalence in the ICL cohort to expected rates derived from the general
population in the Surveillance, Epidemiology, and End Results (SEER) Program. Our findings revealed that ICL continued to be associated with increased susceptibility to viral, encapsulated fungal, and mycobacterial diseases, as well as with a reduced response to novel antigens and an increased risk of cancer.
 

We present a statistical downscaling framework for generating fine-resolution climate projections by integrating high-resolution remote sensing data with coarse-resolution outputs from Earth system models. The proposed multivariate spatial statistical model leverages a basis function representation to enable scalable computation while accommodating potentially nonstationary spatial dependence. We apply the method to downscale sea surface temperature (SST) projections over the Great Barrier Reef region using CMIP6 model outputs. Results demonstrate substantial reductions in mean squared predictive error relative to state-of-the-art approaches, while also providing full predictive distributions that support comprehensive uncertainty quantification.