36214 research outputs found
Sort by
Investigations of Influenza Vaccine Effectiveness: Assessing Repeat Vaccination and Illness Duration among Breakthrough Cases
No full textThesis (Ph.D.)--University of Washington, 2025Background: Influenza causes significant morbidity and mortality every year. Our best line of defensive against influenza is vaccination; however, despite decades of study, questions remain about the full impact of the vaccine. Influenza viruses undergo antigenic drift, requiring annual updates to the vaccine composition. Consequently, the effectiveness of the influenza vaccine differs year to year. One knowledge gap is understanding the impact of prior vaccination on current season vaccine effectiveness. For the first aim, we considered the antigenic distance hypothesis, which postulates negative interference on a given season's vaccine effectiveness when prior and current season vaccine viruses are similar and positive interference when the prior season vaccine virus and current season circulating viruses are similar, as a framework to describe how prior vaccination can negatively or positively impact the current season's A(H3N2) vaccine effectiveness. Another gap in knowledge is how vaccines might impact length of illness among breakthrough cases. For the second aim, we examined vaccine effectiveness (of all virus types) on illness duration among individuals who tested positive for influenza. We focused on individuals who sought ambulatory care and explored the potential for routinely collected data to serve as a proxy for illness duration.Methods: We used data from the United States Influenza Vaccine Effectiveness Network (US Flu VE Network), a collaborative of hospitals and universities across the United States and the Centers for Disease Control and Prevention that estimated influenza vaccine effectiveness every year since the 2004–2005 season. Annual vaccine effectiveness was estimated using a test-negative design. Each year, consenting individuals presenting to ambulatory care were enrolled into the study and tested for influenza. For the first aim, we focused on seasons 2016–2017, when the vaccine A(H3N2) virus was updated from the prior season, and 2017–2018, when the vaccine A(H3N2) virus remained the same as the prior season. For influenza-positive individuals we determined whether the genetic clade of infecting viruses matched the prior season's vaccine virus and categorized individuals into match/mismatch groups. Multiple imputation was used to estimate clade designations for individuals without genetic characterization data. We estimated adjusted odds ratios (aOR) and vaccine effectiveness (VE) by season and match/mismatch status of infecting viruses and the prior season's vaccine virus. For the second aim, we calculated illness duration based on survey responses to questions about returning to normal activities. We estimated vaccine effectiveness against illness 7 days (versus recovering in <7 days) among influenza-positive individuals from seasons 2013–2014 to 2018–2019. We conducted random forest and area under the receiver operating characteristic (ROC) curve (AUC) analyses to assess whether data from medical records (e.g., International Classification of Diseases codes) could be used to accurately classify illness duration.
Results: In the first aim, we did not observe negative interference in 2016–2017, when the vaccine viruses were not the same, as expected per the antigenic distance hypothesis. In 2017–2018, negative interference was not observed for cases with infecting viruses that did not match the prior season's vaccine virus, but was observed for cases with matching clades (between repeat vaccinees and current season only vaccinees, aOR=1.27; 95% CI: 1.01, 1.61), which was inconsistent with the antigenic distance hypothesis. In the second aim, we did not demonstrate a substantial difference in recovery time between vaccinated and unvaccinated influenza-positive individuals (VE=10%; 95% CI, 0%, 10%). In subgroup analyses, significant vaccine effectiveness estimates were observed in seasons 2015–2016 and 2018–2019, as well as for B/Victoria viruses across all seasons. The top variables for predicting illness duration 7 days was site, age, virus type, antiviral prescription, and chest x-ray order. However, the ROC/AUC (AUC=0.65; 95% CI: 0.63, 0.67) results showed poor performance of the random forest model to accurately classify illness duration.
Conclusion: Our results in the first aim were consistent with the antigenic distance hypothesis in one of the two seasons studied. In other studies, the 2017–2018 season was characterized by subclade heterogeneity, which may explain the differential vaccine effectiveness estimates. Furthermore, in the second aim our results did not demonstrate significant differences in illness duration by vaccination status among the medically attended influenza-positive ambulatory care population, though there may be differences in some subgroups. A limitation of both these studies was missing data. We used multiple imputation and sensitivity analyses to address this challenge; however, we acknowledge the limits to any inferences made from these results
Quantifying microenvironmental changes in the developing brain in response to acute and chronic metabolic disrupters
No full textThesis (Ph.D.)--University of Washington, 2025Neurologic diseases are responsible for nearly one-third of all deaths and disability life-adjusted years, many with no effective treatments or cures. Treating the diseased brain is challenging physically, biologically, and clinically: the brain has multiple unique barriers to therapeutics, neurologic disease processes are highly multiplexed and variable, and the presence of pre-existing co-morbidities or prior neurologic conditions changes the disease landscape, complicating or impeding treatment efforts. In this work, we touch on all three challenges. We focus on metabolic disruptions in the form of mitochondrial dysfunction, which is implicated in nearly every neurologic disease and is shown to be a mediator of risk and susceptibility. We use organotypic whole-hemisphere (OWH) brain slice cultures to quantify how mitochondrial dysfunction alters the physical and biological microenvironments of the brain. First, we develop an OWH slice model of mitochondrial dysfunction using the canonical inhibitor rotenone (ROT). We observe region-, dose-, and time-dependent microenvironmental changes that mirror in vivo models. We also show how the extracellular microenvironment, a critical therapeutic barrier, is altered by mitochondrial dysfunction. Next, we characterize how mitochondrial dysfunction from ROT exposure modulates susceptibility to stroke-like injury using an oxygen-glucose deprivation (OGD) model, an effect which has not been previously investigated in vitro. Here, we demonstrate that timing of metabolic disruption relative to the OGD insult worsens tissue recovery. Gene expression analysis and imaging reveal a connection between mitochondrial state and inflammatory responses as one driver for metabolic-related effects on OGD recovery. Our findings highlight a role of pre-existing metabolic deficits in neurological injury and recovery, capture changes in microenvironment features that can impact therapeutic delivery and enable pre-clinical screening platforms that better represent the clinical scenario for patients seeking treatment for neurologic disease
Algorithimic data efficient learning in the era of large model.
No full textThesis (Ph.D.)--University of Washington, 2025In the race towards Artificial General Intelligence, data is the fuel that powers our most advanced models. Vision-Language Models like LLaVA and CLIP are trained on billions of image-text pairs, while Large Language Models (LLMs) like GPT and Claude may process trillions of text samples. Despite the abundance of data, ensuring its quality and effective curation remains more of an art than a science. This process must manage real-world data that is multimodal, noisy, and lacks a guaranteed relationship to target tasks. Furthermore, the process is compounded by the complex training dynamics of neural networks, where the value of each data point depends heavily on the evolving state of model training. Without principled guidance, these challenges often create systematic blind spots, and their impact remains unclear due to a lack of theoretical understanding. My research aims to develop \textbf{theoretical foundations for data curation} through designing \textbf{theory-inspired algorithms} under realistic assumptions and establishing systematic empirical evaluation frameworks to understand the limitations of existing methods including: 1/ target-aware data curation in pretraining 2/label-efficient finetuning 3/ inference-efficient data synthesis and 4/ Interactive learning theories
Exploiting Geometric Constraints for Parameter Quantification in Balanced Steady-State Free Precession MRI
No full textThesis (Ph.D.)--University of Washington, 2025Balanced Steady-State Free Precession (bSSFP) is a magnetic resonance imaging (MRI) sequence well known for its high signal-to-noise ratio efficiency and T2/T1 contrast that suffers from banding artifacts caused by its dependence on static magnetic field inhomogeneities. While phase-cycling is a common remedy to banding, the complex, biphasic nature of the bSSFP signal has historically made its rich phase information difficult to exploit for quantitative tissue and field mapping. This work overcomes this challenge by introducing novel techniques for parameter quantification in bSSFP MRI, and exploiting the geometric constraints of its unique signal profile, which forms a parameterized ellipse in the complex plane. First, a parameter quantification method using four phase-cycled bSSFP acquisitions is established via the geometric property of the signal ellipse's cross-point. An auxiliary circle with a one-to-one correspondence to the bSSFP signal ellipse is considered, facilitating the elucidation of ESM parameters and leading to an analytic ''ellipse unlocking'' method. This cross-point formalism allows for the direct extraction of ellipse parameters, from which quantitative maps of the transverse relaxation time T2 and field-related phase components are generated. Building on the identified information redundancy within the signal ellipse, the second project develops an analytical solution requiring only three phase-cycled acquisitions, further exploiting the inherent constraints. This Direct Analytic Solution (DAS) is derived from a linear system during the ellipse-to-circle transformation, formally reducing the data redundancy. However, analysis reveals that DAS is highly noise- and banding-sensitive and fails in specific scenarios. This investigation indicates the limitations of a purely analytical approach with minimal data, thereby motivating the search for a more robust method. To overcome these limitations, a robust numerical method was developed using three acquisitions. This approach imposes a new geometric constraint—that the signal ellipse passes through the origin—to create a simplified model. A regularized joint optimization procedure then solves the resulting nonlinear least-squares problem, yielding artifact-free images and quantitative B0 maps. A practical challenge common to all these quantification methods is the effective, phase-preserving combination of solutions from multi-channel coil data. An accurate, phase-preserving solution combination strategy is therefore critical for robust quantification performance. The Optimal Weighted Average (OWA) method is implemented for this purpose, which uses regional variance weighting to combine these solutions effectively. The optimality of the weighting scheme is confirmed through mathematical derivation, and efficacy in reducing noise is demonstrated with experiments. Collectively, this thesis demonstrates how geometric constraints can be leveraged for robust and highly efficient parameter quantification from phase-cycled bSSFP MRI, achieving accurate results with a minimal number of acquisitions
Out of Harm's Way: Prioritizing Community Assets in Westport, WA
No full textThesis (Master's)--University of Washington, 2025This thesis project develops a three-step methodology to prioritize community assets in hazard-prone areas for community resource-sharing objectives, with a focus on the city of Westport, WA and the surrounding coastal area. The area of study is at risk of islanding, being cut off from the surrounding urban centers due to natural disasters and hazards. The research begins by parsing community-identified assets from prior workshops to create a comprehensive asset inventory aligned with local knowledge and values. Next, assets were mapped against hazard exposure scenarios, including tsunami and sea level rise. Finally, the study applied a multi-criteria analysis (MCA) framework, informed by community-identified themes related to resource sharing, to rank assets by their potential to enhance post-disaster resilience. Results highlight public spaces like parks and the school as priority assets for resilience efforts, given their roles in community life. This study emphasizes the importance of integrating community perspectives and dynamic hazard assessments into resilience planning to enhance local preparedness. The project also outlines opportunities for future research, including community asset scoring
Riverine Biogeochemical Exports from Major Watersheds to the Northwest Patagonian Estuarine Network
No full textThesis (Master's)--University of Washington, 2025River runoff linking terrestrial and marine ecosystems is a significant driver of marine processes (Lewis & Cook, 2023). The effect of river runoff on marine ecosystems is determined by variability in its timing, quality, and quantity. Human watershed perturbations – both direct, e.g. land-use change, and indirect, e.g. climate change – alter all three of these properties. Measuring baseline variability of these properties is crucial to quantifying the effects of human perturbations, which is necessary for tailoring watershed management to limit disruption of marine ecosystem health. This is particularly important in Northwest Chilean Patagonia (NCP), where a semi-enclosed estuarine system highly influenced by river runoff supports much of Chile's $7+ billion fisheries industry. Thanks to a citizen-science effort of unprecedented scale, we were able to characterize baseline inputs of key nutrients to the NCP marine environment for the 2022-2023 water year from the region's five largest watersheds: the Puelo, Yelcho, Palena, Cisnes, and Aysen, from North to South, respectively. Given the pristine nature of NCP's watersheds, these estimates represent one of the world's first regional characterizations of freshwater nutrient inputs to a temperate fjord system in the absence of extensive human intervention. Across all five rivers, we found freshwater concentrations of total nitrogen (TN), total phosphorus (TP), and dissolved organic carbon (DOC) to be lower than those of the receiving subantarctic marine waters, and well below global averages for comparable coastal temperate rainforest (CTR) ecosystems. (For the purposes of this paper, the metrics TN, TP, and DOC will be considered proxies for nitrogen (N), phosphorus (P), and carbon (C).) Conversely, large watersheds were found to be significant contributors of iron (Fe) and dissolved silica (dSi), collectively discharging ~ 79.9 and 291 metric tons during the 2022-23 water year. Taken together, these results indicate a net regulating function of NCP's five largest rivers on marine ecosystems, possibly inhibiting harmful algal blooms by favoring diatoms over dinoflagellates. The results suggest the importance of large watersheds in regulating marine ecosystems, highlight the potential value of watershed conservation for retaining ecosystem services of TN and TP dilution, and provide a baseline characterization of land-sea linkages in a temperate coastal system
Private Choices in Public Health: A Framework for Economic Epidemiological Modeling of Infectious Disease
No full textThesis (Ph.D.)--University of Washington, 2025This dissertation investigates the role of risk compensation in infectious disease dynamics by examining how individuals adjust private preventive behaviors in response to perceived mortality risk, and how these behavioral shifts shape and are shaped by epidemic trajectories. While conventional epidemiological models assume fixed or policy-driven contact rates, real-world transmission depends critically on how individuals perceive and respond to evolving risk. Public health policies, in turn, interact with these private behaviors—amplifying or dampening their effects. The first part of this dissertation provides empirical evidence of risk compensation behavior by analyzing high-frequency mobility data across U.S. counties during the COVID-19 pandemic. Using lagged local mortality as a proxy for perceived risk, the study finds that individuals significantly reduced their mobility in response to rising deaths—particularly in high-contact, discretionary domains. These responses evolved over time and were shaped by public health interventions such as shelter-in-place orders and mask mandates, which often amplified rather than displaced private behavioral changes. The second part develops a dynamic economic-epidemiological model that endogenizes contact rates through mortality-responsive behavioral feedback. Critically, the magnitude of behavioral responsiveness estimated in the empirical analysis is used to calibrate the strength of the feedback loop—bridging observed behavior with modeled transmission. This endogenous co-evolution of behavior and epidemic severity creates a self-regulating system in which private action dynamically responds to real-time risk signals, suppressing transmission as perceived threat intensifies. The model, grounded in a delayed SEIRDS framework, demonstrates how such feedback can flatten epidemic curves, delay transmission peaks, and produce multiple waves. Introducing pandemic fatigue—modeled as declining responsiveness to risk—erodes this adaptive loop, weakening the self-limiting dynamic and enabling persistent spread. Simulations highlight the importance of accounting for behavioral adaptation in forecasting and policy design. Together, these studies offer a unified framework for understanding how private choices interact with public policy and disease dynamics through the lens of risk perception and behavioral adaptation
Towards Multimodal Interactive Intelligence
No full textThesis (Ph.D.)--University of Washington, 2025Great progress has been made in multimodal generative AI models. However, these models still have limitations. For example, they struggle when handling multimodal data and multi-turn interactions. One reason is the lack of training data for these problems. The research community lacks multimodal data on the scale of single modality data and lacks lengthy multi-turn interaction data. Additionally, there are not many well-defined tasks for these problems, preventing researchers from understanding models' performance and leaving them without meaningful optimization goals. In this thesis, we work toward building better multimodal intelligence. We focus on three types of abilities: multimodal understanding, multimodal generation, and grounded multi-turn interactions. For each aspect, we explore the limitations of current models, proposing new tasks and evaluation methods for capabilities that remain beyond the reach of existing models. Identifying these weaknesses, we introduce novel methods and evaluations for multimodal interactive intelligence to address these challenges. This approach enhances existing AI models through AI-AI interactions and human-AI interactions, enabling collaboration across modalities. For multimodal understanding models, we propose BLINK, a benchmark that focuses on core visual perception abilities not found in other evaluations. Most BLINK tasks can be solved by humans in a ``blink'' of the eye, but pose significant challenges for the latest multimodal language models (LMs). To address this weakness, we propose Visual Sketchpad, a framework that allows models to think step-by-step across modalities. This framework empowers LMs to have more diverse interactions, for example, with vision expert models. Such interactions compensate for what existing models miss and greatly enhance models' multimodal understanding abilities. For image generation models, we tackle the long-standing problem that these models do not effectively follow text instructions. We propose TIFA, which uses multimodal LMs to evaluate generated images, providing an efficient evaluation metric that aligns well with human judgment. Moreover, we show that TIFA can work as an effective training signal to improve text-image alignment in image generation. Finally, we focus on grounded dialogue systems. We provide a framework that allows AI agents to be evaluated with either simulated or real users, using end-to-end dialogue-level objectives. To demonstrate the use of this framework, we introduce NavigationBench, a novel task that simulates dialogues between a user and a virtual navigation assistant in a car. It also features a simulated user trained with the latest LM technologies, allowing researchers to simulate multi-turn dialogues for automatic dialogue-level comparisons of AI assistants. Using this framework, we study the performance and verbosity of different agent LLMs
Cost-Effectiveness of Delaying Alzheimer’s Disease Progression with Novel Monoclonal Antibodies
No full textThesis (Master's)--University of Washington, 2025Background: In the US alone, most of over 7.2 million older adults with Alzheimer’s Disease (AD) treat their AD with symptom-relieving therapies. Starting in 2021, three disease-modifying monoclonal antibodies (mAbs) entered the US market as the first disease-modifying treatments for early Alzheimer's disease (AD). Despite years of presence in the US market, the value of these mAbs remains uncertain due to high treatment-related costs and an unclear comparative value of their clinical benefit. Objectives: 1) To estimate the cost-effectiveness of adding aducanumab, lecanemab, or donanemab to the current standard of care (SOC) and 2) to evaluate the value of reducing uncertainties around the mAbs from a US health care perspective and a modified societal perspective.
Methods: A Markov simulation model was developed by incorporating literature and a prior cost-impact model of delaying AD progression. The clinical benefit and cohort’s characteristics mirrored results from phase 3 clinical trials of the three agents. The price of the agents was estimated using wholesale acquisition costs published by Micromedex RED BOOK. The model used a 1-week cycle length and estimated 10-year cost impact and comparative incremental cost-effectiveness ratios (ICERs) among the SOC and the mAbs. Results were described from both a health care sector and a societal perspective. One-way sensitivity analysis was conducted to demonstrate key drivers of cost-effectiveness of the mAbs. Cost-effectiveness acceptability curve and the expected value of perfect information (EVPI) used probabilistic sensitivity analysis to demonstrate the comparative value of the mAbs and the value of resolving uncertainties.
Results: The three agents were estimated to delay progression from MCI to severe AD by 6.2 - 9.2 months. The analysis showed that SOC costs 757,800 from the health care sector and the modified societal perspective, respectively, over 10 years. From the health care sector perspective, the mAbs incurred additional 73K, and 73K, and aducanumab and lecanemab added 64K to the cumulative costs, respectively. Donanemab had the higher comparative ICER at 175,300/QALY. Key driving factors of ICER among lecanemab and donanemab were age at the treatment initiation, patient health state utility values in earlier AD states, and the drug acquisition costs across both perspectives. SOC had more favorable cost-effectiveness at a WTP of 200K/QALY. The EVPI of the decisions around lecanemab versus SOC was estimated to be 190K/QALY and 109 million at the WTP of 275K/QALY for the healthcare sector and the societal perspectives, respectively.
Conclusion: In the model simulation, the novel AD treatments incurred substantial costs with modest clinical benefits, with donanemab adding the highest cost. Assuming the treatment costs based on WAC, only lecanemab was found likely to be cost-effective under a WTP of $200K/QALY from the modified societal perspective. The cost-effectiveness of the mAbs is substantially influenced by the patient’s age at the treatment initiation, patient health state utility values at earlier AD states, and the drug acquisition costs
Empirical Orthogonal Teleconnections as a Framework for Regional Climate Analysis: Insights from Pacific Northwest Temperature Extremes
No full textThesis (Master's)--University of Washington, 2025As the climate warms, understanding whether heat waves will intensify following mean temperature trends or undergo dynamic changes in their progression leading to changes in intensity is critical for regional climate adaptation. This study investigates heat wave dynamics in Washington and Oregon using Empirical Orthogonal Teleconnections (EOT) and climate composites, with a focus on setting historical baselines for use in further study with future projections. We analyze daily 2-meter maximum temperature data during the heat wave season (May 1–September 30) from ERA5 (1951–2020) and CMIP6 (historical forcing, 1981–2010) downscaled with WRF. EOT analysis reveals distinct spatial modes of teleconnectivity of maximum temperature, and this study primarily focuses on four: the Whole Region, Northeast, Southeast, and Coastal domains, which experience heat waves distinctly from one another. The composite analysis of ERA5 fields during days in the top ten percent for 2-meter maximum temperature shows both local forcings and a degree of progression from the Coastal to Whole Region, while the Northeast and Southeast modes represent regions of remote forcing during heat waves. This is supported by the results of the heat wave selection criteria used at the defining spatial point of each mode. This work provides a novel framework for the analysis of temperature regimes on a regional scale, having implications for understanding the synoptic and mesoscale conditions necessary for heat waves in different parts of the region