Brain Region Activation Detection in an Audiovisual fMRI Study Functional magnetic resonance imaging (fMRI) is a technique used to generate sequential images of brain activity. It has been widely applied to study changes in brain function due to neurological conditions and psychological disorders. For fMRI studies, the blood oxygenation level dependent (BOLD) is commonly measured to map change in blood flow with a neural activity of interest. Although BOLD contrasts can give insight into brain activity through indirect estimates, it has some drawbacks. One of them is that cerebrospinal fluid movement and circulatory system can influence the measurements, thereby adding noise to the system. Using data from an audiovisual study, this project will investigate ways to identify activated regions associated with audiovisual stimuli.
Relating Hearing Rehabilitation for Age Related Hearing Loss to Cognitive Decline Age-related hearing loss (ARHL) is a result of cumulative effects of aging on the auditory system. ARHL is present in two-thirds of adults older than 70 years and is associated with frailty, falls, social isolation, late-onset depression and disability. There is growing evidence linking ARHL with cognitive decline and increased risk of dementia. Recent evidence suggests that hearing rehabilitation using hearing aids or cochlear implants may prevent ARHL-related cognitive impairment, including protection from cognitive decline, reduced dementia incidence with hearing aids, and greater cognitive dysfunction without hearing aids. We will study how measurements of neuropsychology are related to various measures of hearing loss and hearing function. This study will involve factor analysis to evaluate the many measures of neuropsychology. Those factor scores will be compared against real time ecological momentary assessment (EMA) measures using linear mixed regression models to evaluate the effectiveness of cochlear implants for ARHL related to cognitive impairment in real-world settings.
Machine Learning and Extrapolation: Dealing with the effects of COVID-19 on Infrequently Observed Outcomes Machine learning models have been applied to almost all aspects of the professional world, to the point that business intelligence services have become commodified and made available through standardized cloud platforms. Nevertheless, the unique features of seasonal or annualized outcomes present challenges for typical supervised learning techniques. For example, in the academic world, we are often interested in making predictions concerning student enrollment and success, but can only observe the outcome of interest on an infrequent annual or biannual basis. As a result, the data which is used to train supervised learning models is often different in important ways from what we can observe during a particular year; data on students entering college in Fall 2019 is different in crucial ways from similar data on students entering college during the COVID-19 pandemic. In this project, we’ll use real data on student success and recruitment at the University of Iowa to investigate algorithmic approaches to controlling this kind of extrapolation error, including the introduction of structural penalties in random-forest type models, and the incorporation of such measures as priors in Bayesian Additive Regression Tree (BART) models.
Radiomics for Clinical Outcome Prediction in Head and Neck Cancer Patients Radiomics is an emerging field of medical study in which large numbers of quantitative features are extracted from medical images of patients to aid in the characterization of disease and prediction of clinical outcomes. This project will focus on the application of radiomics to head and neck cancer patients enrolled in a multi-center clinical trial. Project participants will learn about the steps involved in radiomics analysis, including medical image visualization, tumor segmentation, feature extraction and quantification, and statistical analysis. They will apply statistical methods to study relationships between radiomic features and other clinical information available on the trial patients. In particular, regression and machine learning techniques will be employed to identify important features, develop multivariable statistical models to predict clinical outcomes, and evaluate the predictive performance of the models.
Understanding the “Who”, “Where”, and “When” of Congenital Heart Defects among a 10-year Cohort of Iowa Births Public health surveillance includes ongoing systematic collection, analysis, and interpretation of health data. Analysis of surveillance data focuses on the surveillance triad of person (who), place (where), and time (when). The Iowa Registry for Congenital and Inherited Disorders (IRCID) conducts statewide surveillance for major structural birth defects diagnosed among pregnancies of Iowa residents. Congenital heart defects (CHD)s are the most prevalent group of birth defects in the United States, affecting about 1% of births and are a leading cause of birth defect-associated infant mortality, morbidity, and healthcare costs. Previous studies suggest that the etiology of CHDs is multifactorial, with several genetic and environmental (broadly defined) antecedents. Risk associated with environmental antecedents can vary across time and geographic regions, due perhaps to differences in racial/ethnic distributions in these regions, but also perhaps to other unidentified risk factors related to environmental exposures, socioeconomic factors, neighborhood effects, etc. Lacking in Iowa are detailed studies of spatio-temporal risk for this defect in a well-described cohort. Using a 10-year cohort of CHD cases ascertained by the IRCID and statewide birth data for this same time period, this project will investigate the descriptive epidemiology and spatio-temporal risks for CHDs in Iowa using disease mapping spatial statistics and longitudinal data analysis methods. This analysis will provide important insights into geographic areas of increased risk, how risk has changed over time, and factors that may be related to changes in risk.
Discovering latent patterns in health behaviors and environment and their relationship with infant health outcomes in low-to-middle income countries Dozens of viral, bacterial, and protozoan enteric pathogens are transmitted by fecal contamination of the environment in low to middle-income countries (LMICs), collectively causing 2.5 billion episodes of diarrhea and 580,000 deaths in children under five years of age each year. Infection, whether diarrheal or asymptomatic, elevates a child’s susceptibility to co-infection, enteric dysfunction, and long-term cognitive and developmental stunting. Illness and death among children under 5 years of age accounts for 40% of the total global health impact of foodborne illness. Our prior research has shown that in low-income neighborhoods of Kenya the enteric pathome children ingest is taxonomically complex, and co-ingestion of multiple pathogens is common. This project’s focus is on the behaviors of infant caregivers and the environment in which infants spend time. We will utilize unsupervised machine learning algorithms to try to discover latent grouping structure in infants’ environments in low income neighborhoods in Kenya. We will then use statistical techniques to relate these discovered groups to enteric infections discovered by polymerase chain reaction (PCR) analyses of infant stool samples and self-reported infant diarrhea.
Do obesity, hypertension, and diabetes affect recovery from bone marrow transplantation? Bone marrow transplantation is an important treatment for many disorders, in particular blood cancers such as multiple myeloma and leukemia. In such cases, the recipient’s immune system is usually destroyed with radiation or chemotherapy before the transplantation. Recovery of the immune system after transplantation is therefore critical. The purpose of this study is to examine whether that recovery of immune function is affected by obesity, hypertension, and diabetes. Do white blood cell counts for patients with these conditions remain lower longer? Do they remain in the hospital longer? One reason these particular conditions are of interest is that all three are known to be associated with sleep apnea. Thus, this study may lend support to the idea that treating sleep apnea could improve the recovery of bone marrow transplant patients.
An Analysis of the Differential Impacts of COVID-19 Mortality in the United States As of June 1, 2021, the COVID-19 pandemic has resulted in nearly 600,000 deaths in the United States. It is well known that COVID-19 mortality has differentially impacted the populace based on age, race/ethnicity, and geographic region of residency. In this analysis, we use a COVID-19 case surveillance database to quantitatively characterize the relative risks of mortality among individuals diagnosed with COVID-19 based on geographic and demographic variables, as well as variables that reflect pre-existing medical conditions, ICU admission, and time of diagnosis. This database is maintained by the Centers for the Disease Control for public use, and is currently comprised of over 25 million patient records.