Bayesian Models of Environmental Prevention of Lyme Disease Lyme disease is an important vector-borne illness which can result in severe symptoms which may persist for years. In recent years, the prevalence and geographic range of Ixodes ticks which transmit the infection have grown and expanded due to climate change and changes to habitat. This project will build Bayesian longitudinal models to analyze data from an ongoing trial to tackle Lyme risk at its source: in the environment. Using a novel reservoir targeted vaccination method, our project aims to interrupt the maintenance of the pathogen in the environment, and thereby to reduce risk to incidental hosts like humans and domestic dogs. This work will involve modelling outcomes for multiple species simultaneously in an ecological context, and will provide a basis for planning interventions with direct human health implications.
Comparing exposure pathways of enteric pathogens to infants living in low- to middle-income countries Over half a million children die each year from diarrheal diseases. This burden disproportionately affects those living in low- to middle-income countries. Children in these areas experience a wide range of exposures to pathogens (e.g., milk, food, caregiver hands, soil, animal feces), and it is unclear which exposure pathways are among the most important contributors. In an ongoing study in Nariobi and Kisumu, Kenya, led by Drs. Kelly Baker and Daniel Sewell, structured observational data have been collected on children’s behaviors. The goal of this research project is to use Bayesian methodology to combine these data with previous studies in order to estimate the probability of becoming infected from various pathogens from a variety of pathways. The results of this project will provide key information for developing effective intervention strategies for reducing diarrheal disease in children living in low- to middle-income countries.
Compare Cancer Rates by Congressional District Previous studies have created estimates of Congressional District (CD) cancer rates and have found higher proportions of rural residents within a CD are correlated with higher cancer mortality-incidence ratios. However, these methods have been reliant on using county-level cancer surveillance data and adjusting based on census block-level population, rather than using cancer data at smaller geographic levels (i.e., ZIP code tabulation areas-ZCTAs) and more rigorous statistical methods to estimate. With the new CD boundaries that have been determined by states following the 2020 census, it is important to assess CD-level rates with these new districts. Policy-relevant data—such as CD-level cancer estimates—can provide critical information to policymakers and advocacy organizations to enhance their efforts and target resources. This project will build upon our current project in which we are using Bayesian hierarchical models to develop ZCTA level age-adjusted cancer rates and visualizing these data using interactive, web-based tools. We will create and visualize estimates at the CD-level to allow researchers to engage with important partners to assess the utility and implementation of these maps for cancer and public health advocacy groups.
An Investigation of Cachexia Syndrome in Cancer Patients Do male and female cancer patients experience similar rate of weight loss? A syndrome accompanying cancer, which affects nearly 50% of all cancer patients and the near totality of all cancer patients with advance disease, is called Cachexia. Cachexia is notable by unintentional weight loss due to a depletion of skeletal muscle and adipose tissue mass. Patients with gastrointestinal cancers are disproportionally affected by cachexia. It has been assessed that patients with pancreatic ductal adenocarcinoma (PDAC) experience the highest incidence of weight loss, with an estimated 70% of PDAC patients affected. The current literature suggests that male and female patients are equally affected. However, this observation has not been supported by recent computerized tomography (CT) image analysis, which tends to suggest that while relative weight loss may be equivalent between male and female PDAC patients, there may be a difference between these two sexes in terms of the location where skeletal mass has been depleted. This project purposes to study gender difference in Cachexia, and whether males tend to lose body weight due to the depletion of skeletal muscle, versus females primarily losing their weight from adipose tissue loss. We will consider three sources of weight loss: skeletal muscle mass, visceral adipose tissue, and subcutaneous adipose tissue. Analytical models that account for within-subject correlations will be used and recommendations will be made following our findings.
Modeling income disparities In most societies, there is variability in the amount of income and wealth obtained by families and individuals. There is also variability in individual beliefs of what the “ideal” income distribution would be. A fairly recent publication by Pluchino et al (2018) proposed a model that looked at how random opportunities or luck can affect wealth distribution more than “talent”. In our project, we would do further simulations to investigate to this model, and expand the model to accommodate issues such as the role of opportunities, systemic bias, and generational wealth.
Predicting Outcomes in a Population in Early Recovery from Alcohol Use Disorder In this project, we’ll focus on a cohort of survey respondents who identify as being in recovery/recovering from Alcohol Use Disorder (AUD). Specifically, we will attempt to build Machine Learning models to accurately predict the occurrence of relapse, and will apply interpretable ML techniques to communicate primary drivers of drinking outcomes. ML models used will include Random Forests, Gradient Boosted Trees, Neural Networks, as well as meta-learning techniques such as stacking.
Why is learning so often difficult to achieve? A typical college student today has more learning resources on their campus than ever, yet they are more likely than ever to have adverse learning experiences during their studies. For many students, adverse experiences begin as only a hint of discomfort due to intellectual content that challenges their prior knowledge or personal values. This discomfort, however, causes some students to ruminate upon negative thoughts and feel negative emotions. Cycling through negative thoughts and emotions leads to overt distal behaviors, such as dropping classes. But discomfort also manifests in covert proximal cognition and emotion—sometimes these feelings remain below the students’ conscious awareness. Fortunately, feelings manifest in students’ physiological responses. This project therefore aims to determine the role that biofeedback plays in sustaining students’ efforts to engage with ideas that induce cognitive and emotional discomfort. The team aims to transform intellectual adversity into a resource for learning instead of an obstacle to it. By combining conceptual change theory with theory from positive psychology the proposed research aims to construct generative learning theory that explains how cognition and emotion interact to form the psychological mechanisms responsible for learning during discomfort. The project will achieve this aim by mobilizing technological innovations in neuroimaging (fNIRS) and wearable sensors (EDA) that detect people’s covert cognitive activity and arousal states—this will address a critical need neglected by research that uses self-report data and learners’ overt talk as the sole metrics used to construct claims regarding learners’ cognitive and emotional discomfort. These technologies also augment people’s capacity to monitor and reflect upon the emergence of their discomfort and thus deliver to learners a psychological buffer that supports their continued learning—transforming a liability into an asset.
Predictive Modeling for Body Fat Percentage Based on Anthropometric Measures The percentage of an individual’s body fat is an important physiologic measure used to characterize fitness and health. However, accurate assessments of body fat percentages can be difficult and expensive to obtain. The dual-energy x-ray absorptiometry (DEXA) scan is considered the “gold standard,” and yet this procedure must be done in a laboratory setting. The main purpose of this project is to develop models for predicting the percentage of body fat in males based on easily obtained anthropometric measures, such as weight and height, and circumferences for neck, chest, waist, hip, thigh, knee, ankle, bicep, forearm, and wrist. One of the challenges in developing such models is that anthropometric measures are inherently collinear, which can result in highly inaccurate model parameter estimates. To formulate our models, we consider approaches based on variable selection and dimension reduction. The models will be based on a dataset comprised of 250 records on male subjects who have had their percentage of body fat accurately obtained. The data was collected at the BYU Human Performance Research Center.
Predicting Time Until Renal Failure for Newly Diagnosed C3G Patients C3 glomerulopathy (C3G) is a group of related conditions that cause renal disease. These kidney problems worsen over time and eventually these patients will progress to end-stage renal disease (ESRD). A natural question among newly diagnosed patients is: “how long do I have until I will need a kidney transplant”? Because C3G is a rare disease, there is not a particularly satisfactory answer to this question, especially if we attempt to account for factors that may improve or worsen prognosis (age, sex, presence of protein in the urine, etc.). Kidney disease researchers at the University of Iowa have been following a cohort of C3G patients for many years in the hopes of better answering this question (and many others). This project is a great opportunity to learn about kidney disease and time-to-event modeling.