To Be Assigned

Permanent URI for this collectionhttps://digital.lib.washington.edu/handle/1773/20986

Items in this collection will be assigned to permanent collections within the ResearchWorks Archive.

Browse

Now showing 1 - 20 of 22

Continuous and Short Fiber Reinforced Composites: Optical Coherence Tomography and Mechanical Characterization
(2026-04-20) Saadat, Sanaz; Sadr, Alireza
In this dissertation, Dr. Sanaz Saadat investigated the polymerization behavior and mechanical performance of short fiber and continuous fiber–reinforced dental composites using Optical Coherence Tomography (OCT) and three-point bending (flexural strength) testing. Optimizing polymerization behavior and mechanical performance remains a major challenge in dental composite restorations. This research investigates the effects of short fiber and continuous fiber reinforcement on polymerization behavior and flexural strength. Real-time Optical Coherence Tomography (OCT) imaging of deep Class I restorations revealed that continuous fiber reinforcement reduced polymerization shrinkage induced gap formation, with the greatest effect observed in conventional composite formulations. Flexural strength 3-points bending test further emphasized that combining short fiber reinforced composites with the continuous fiber achieved higher strength and exhibited more favorable, less catastrophic failure patterns than conventional composites. Overall, the incorporation of continuous fiber increased both the flexural strength and deformation capacity of the tested composite systems, and combined use of short and continuous fibers improved interfacial adaptation, lowered shrinkage stress, and enhanced overall structural performance, supporting continuous fiber reinforcement as an effective approach for increasing the durability of composite restorations.
Cross-Detector Descriptor Fusion: Scale Control and Spatial Alignment for Local Feature Matching
(2026-04-20) Sossi, Frank Thomas; Olson, Clark F
Cross-Detector Descriptor Fusion:Scale Control and Spatial Alignment for Local Feature Matching Frank Sossi Chair of the Supervisory Committee: Committee Chair Professor Clark Olson Computing & Software Systems Local feature descriptors are fundamental to many computer vision applications including SLAM, structure from motion, and image retrieval. This thesis evaluates two approaches to improving local feature matching: using multiple detectors as a quality filter for keypoint selection, and fusing complementary descriptors to combine their strengths. We show that spatial intersection between different keypoint detectors acts as a quality filter. When different detection methods, whether SIFT and SURF or SIFT and KeyNet, both identify a keypoint at the same location, this consensus indicates a distinctive feature. Descriptors computed at intersection keypoints consistently outperform those on single de- tector sets, with HardNet achieving 82.1% mAP on SIFT-KeyNet intersection, a 25% relative improvement and the best single descriptor result in our study. In order to evaluate color descriptors we re-implemented a color version of the HPatches patch benchmark, allowing us to evaluate color aware descriptors. Using this dataset, we show that fusing the color histogram descriptor HoNC with learned CNN descriptors yields substantial improvements: HoNC+SOSNet concatenation achieves 50.6% mAP on patch matching, outperforming all individual descriptors. HoNC’s strong discriminative capability (high verification to matching ratio) complements the CNN’s matching optimized represen- tations. Cross family fusion (SIFT+CNN) requires pre-fusion L2 normalization to ensure equal contribution from each descriptor; with proper normalization, SIFT+HardNet achieves 46.0% mAP on patches. Keypoint scale is also a dominant factor: filtering to large scale keypoints yields 39% relative improvement for SIFT and 21% for CNN descriptors. We develop DescriptorWorkbench, an open source evaluation framework, and conduct over 100 experiments. The results show that keypoint quality determined by detector con- sensus and scale has greater impact on matching performance than descriptor algorithm choice alone.
Multi-modal profiling reveals epithelial hierarchy disorganization underlying field cancerization in head and neck cancer
(2026-04-20) Mills, Claire Burke; Beronja, Slobodan
Local recurrence rates in head and neck squamous cell carcinoma (HNSCC) exceed 50%, potentially due to field cancerization, a phenomenon where tissue that appears normal microscopically actually harbors molecular changes that can give rise to new tumors. The prevailing model assumes these "at-risk" tissues share driver mutations with adjacent tumors, but whether field cancerization depends solely on mutations remains unknown. To comprehensively characterize this phenomenon, we profiled matched tumor, normal-adjacent field (NAF), contralateral normal, and blood samples from HPV-negative HNSCC patients using whole genome sequencing, targeted driver gene sequencing, single-nucleus multiome profiling, and functional cell-fate tracing assays. While histologically normal NAF tissues did harbor expanded mutant clones, these mutations rarely overlapped with paired tumors and were not shared across patients, suggesting features beyond mutations are altered in NAF. Single-cell multiomics revealed that NAF epithelium loses its normal organizational hierarchy: basal stem cell markers persist in cells that should be differentiating, while terminal differentiation programs fail to activate. Proliferating cells appeared in tissue layers normally restricted to post-mitotic cells, a pattern typically seen only in overt dysplasia. Additionally, we identified a distinct "emerging state" cell population enriched in NAF that expresses tumor-associated markers and transcriptionally bridges normal epithelium and tumor, which we spatially identify using immunofluorescence staining of key markers. Together, our findings demonstrate that field cancerization operates through epithelial hierarchy disorganization, disrupting differentiation and generating cells with tumor-like features, all within tissue that appears histologically normal.
A Unified Token and Parameter Compression Pipeline for High-Resolution Vision–Language Models
(2026-04-20) Sultana, Tasnia; Ali, Mohamed
High-resolution vision–language models achieve strong performance on fine-grained visual reasoning tasks, but their deployment remains costly due to large visual token counts and heavy language backbones. This work investigates how to build small and efficient multimodal models while preserving high-resolution reasoning ability. We propose a training-free unified compression pipeline that reduces inefficiency at both the token and parameter levels. At the token level, we introduce HiRED–Merge, which combines attention-guided token budgeting with neighbor-aware norm proportional token merging. The method merges only spatially adjacent tokens that survive attention-based selection, helping preserve local structure and reduce information loss from aggressive token dropping. At the parameter level, we apply GLU-aware structured MLP pruning to the language backbone, removing coupled neuron pairs while maintaining dense computation and model structure. A 20% pruning reduces a 7B model to approximately 6B parameters. Experiments on ScienceQA, TextVQA, DocVQA, ChartQA, and MME show that our pipeline improves throughput, memory efficiency, and scalability while maintaining competitive accuracy. These results enable practical deployment of high-resolution vision–language models under limited computational resources.
Racial Disparities in Health Outcomes Across Criminal Legal System Contacts: A Systematic Literature Review
(2026-04-20) Mortimer, Leah; Spigner, Clarence
Objective: This systematic literature review identifies studies that found racial disparities in health consequences associated with the criminal justice system contact and analyzes the magnitude of inequities in health outcomes related to those racial disparities.Methods: Following the PRISMA 2020 Statement, research articles were identified using the Web of Science database through the search terms “crim* and health and race.” Published original research articles, early access reports, and review articles published in the US, in English, from January 2000 through June 2024 were included. Snowball referencing of selected studies was used to identify additional studies along with non-peer reviewed (“grey”) literature sources to search. Analysis of selected studies reported the time-period of data collection, study type, data source, population, covariates, health outcome, contact with the criminal legal system, racial groups considered in analysis, and information on racial disparity results/conclusions for each selected study. Results: The Web of Science search returned two thousand and ninety-six (2096) articles after deleting duplicates. Of these, fifty-one (51) were considered eligible for full text review and twelve (12) met all inclusion criteria. Snowball referencing yielded an additional six (6) peer-reviewed articles. The grey literature database search yielded an additional two (2) articles for inclusion. In total, twenty (20) articles are included in this review. Included articles identify racial disparities associated with criminal legal system contact across physical health outcomes (8), mental health outcomes (6), general self-rated health (3), and mortality (4). Discussion: Racial disparities in health outcomes associated with criminal legal system contact were reported across disease, injury, chronic conditions, cardiovascular risk, cellular aging, mental health conditions, depressive symptoms, and general self-rated health with black people tending to be more affected than white people. Limitations: Limitations of the evidence include unconsidered covariates, failure to distinguish between multiple racial groups, limited generalizability, a lack of female participants, a reliance on the same national datasets, and an underrepresentation of certain health conditions. Implications: Understanding the racial disparities in health outcomes linked to criminal legal system contact—across various health impacts and types of contact—can help better inform and target interventions that both reduce criminal legal system involvement and improve health among the populations most at risk.
Comprehensive assessment and quantification of incoherent speech using natural language processing
(2026-04-20) Xu, Weizhe; Cohen, Trevor TC
Coherence is a linguistic feature that is defined as the orderly and interconnected flow of ideas. The disruption of coherence is a linguistic anomaly that is commonly observed in a group of psychiatric disorders known as schizophrenia spectrum disorders (SSD), where disorganized thoughts manifest as incoherent speech. While early detection of symptoms can potentially lead to better outcomes, manual assessment of symptom severity can be time-consuming and require specialized expertise. Therefore, symptom evaluation through automated coherence assessment methods is desired. However, gaps remain in prior research on this area, namely 1) most prior work focuses on the estimation of local coherence (coherent transitions between adjacent semantic units) via computation of cosine values between vector representations of sequential semantic units. The estimation of global coherence (sustaining a theme or topic throughout a narrative) has received much less attention; 2) the impact of automated speech recognition (ASR) errors receives little attention. Prior work mainly focused on using manual transcript data; 3) there is limited exploration on using language model perplexity to assess coherence, especially given the recent advancement of large language models (LLM). This work bridges the gaps through the following contributions: 1) Two new global coherence assessment methods were developed based on centroids of embeddings (vector representation of semantic space). We found that the global coherence methods align better with human judgment than local coherence methods. 2) A time-series feature extraction pipeline is used to replace the aggregation step in coherence assessment pipelines. We found that by using this method, coherence evaluation process is resistant to the impact of ASR errors in the text input. 3) Two sentence-level perplexity-based coherence methods were developed, and we revealed that combining perplexity features with traditional coherence scores (proximity features because they are based on cosine similarity) resulting in better prediction models than using proximity or perplexity features alone. 4) The innovations and classical approaches were combined into the Comprehensive Coherence Calculator (CCC), a software package that can perform comprehensive coherence analysis with a myriad of configurations. With these contributions, fully automated coherence assessment pipeline can be established to offer patients easy monitoring at home, clinicians necessary information to provide better care and researchers an objective quantitative basis for the study of semantic coherence.
NeuroPathPredict (NPP): A data-driven paradigm to map the distribution of Alzheimer’s disease neuropathology.
(2026-04-20) Madan, Raghav; Gennari, John H; Crane, Paul K
Alzheimer’s disease (AD) is characterized by the progressive accumulation of misfolded proteins, primarily tau neurofibrillary tangles, amyloid-β plaques, and TDP-43 inclusions, across various brain networks. Despite a century of histopathological insight, quantitative understanding of how these pathologies spatially unfold remains limited. Classical frameworks such as Braak and Thal staging distilled sparse regional observations into ordinal categories, yielding reproducible heuristics for disease progression but obscuring fine-scale gradients and inter-individual heterogeneity. Modern imaging and network-based diffusion models have extended these ideas to the living brain; however, they remain constrained by coarse parcellations, strong mechanistic assumptions, and the lack of direct calibration against histological ground truth. The field, therefore, lacks a rigorous, data-driven framework that can translate quantitative neuropathology into spatially continuous, anatomically interpretable maps across the brain. To address this gap, I developed NeuroPathPredict (NPP), a modular, open-source system that integrates quantitative histopathology with high-resolution neuroimaging and spatial statistics. NPP comprises three primary pipelines. (1) QNPtoVox transforms Halo-derived, tile-wise tau burden measurements into voxel-level maps co-registered to the MNI ICBM 2009b template via ex vivo MRI, preserving anatomical orientation and enabling cross-participant comparability. (2) The Integrated-Brain Information System (I-BIS) creates a multilayer “brain GIS” by unifying various structural, functional, and vascular atlases into a shared 0.5 mm volumetric grid. It produces thousands of biologically relevant covariates, such as distances, densities, and neighborhood features, which help contextualize each voxel regarding white-matter tracts, networks, and tissue types. (3) The NPP modeling framework combines these data using universal kriging with external drift, a geostatistical method that links anatomy-informed predictors with residual spatial autocorrelation to infer continuous tau fields from sparse observations, providing both predictions and spatially explicit results uncertainty. Applied to ten donors from the Adult Changes in Thought (ACT) study, NPP demonstrates that anatomically enriched models outperform non-spatial baselines and recover mesoscale gradients of tau burden consistent with known vulnerability patterns along association tracts and functional networks. The results show that spatial autocorrelation persists beyond measured covariates, validating the use of kriging in brain space and underscoring the value of integrated anatomical context. More broadly, NPP establishes a reproducible computational framework that transforms post-mortem histology into standardized, voxel-wise maps suitable for cross-modal validation with MRI and PET, for testing theories of selective vulnerability, and for modeling the spatial interplay of multiple pathologies. By integrating digital pathology, neuroimaging, and spatial statistics, this work enhances the ability to reconstruct, predict, and ultimately comprehend the brain-wide spatial dynamics of neurodegeneration.
Interpretable Machine Learning for Biomarker Identification in RNA Seq Cancer Data
(2026-02-05) Newton, Jeremy; Kim, Wooyoung
Existing research on RNA Seq gene expression biomarkers has provided various methods to select a small list of genes as cancer biomarkers from a large number of gene expression data. Previous methods for identifying potential gene expression cancer biomarkers have focused on statistical analysis, but other methods have incorporated machine learning, often including Interpretable Machine Learning (iML) techniques. On 16 cancer types from TCGA data, we used inherently interpretable machine learning models: Logistic Regression, Random Forest, and Linear Support Vector Machine to narrow down subsets of potential genes as biomarkers using the trained models' feature importance rankings. We subsequently applied model-agnostic iML techniques, such as Shapley Additive Explanations (SHAP) and Permutation Importance, to narrow down the subsets even further. We compared classification performance between machine learning models trained on iML selected features with features selected by statistical methods, and biomarkers from external research. We found that iML biomarker selection methods lead to comparable or better classification performance on these datasets than the biomarkers from outside research, or from statistical analysis alone. Mutual Information estimation (MI) was a surprisingly useful technique for initial feature selection, and iML techniques improved the MI selected features for classification. We cross-checked potential biomarkers with biomedical annotations and gene pathway analysis, finding some support for the validity of the biomarkers.
User-Guided Deep Multiple Clustering
(2026-02-05) Yao, Jiawei; Hu, Juhua
Multiple clustering is based on the observation that a dataset can often be partitioned in more than one meaningful way (for example by color or by shape). However, most existing deep methods still optimize a single partition or produce several partitions without making clear which underlying factors they capture, and they often separate representation learning from the clustering objective. This can lead to results that do not match the aspects users care about. This dissertation proposes a user-preference guided framework for deep multiple clustering that aims to obtain partitions that are both diverse and aligned with user interests, and is organized into four contributions that start from data-driven ways of identifying relevant factors and progress to methods that explicitly incorporate user intent and practical system considerations. The first contribution, AugDMC, uses targeted data augmentations as aspect selectors together with a self-supervised, prototype-based objective with stabilization, to learn representations that preserve distinct factors of variation and support multiple interpretable partitions without manual feature engineering. The second contribution, DDMC, introduces dual-level disentanglement tailored to clustering: a variational EM procedure links coarse and fine grained factor discovery (E-step) with a clustering-aware objective (M-step), narrowing the gap between learning “good features” and obtaining “good partitions”. The third contribution, Multi-MaP, aligns frozen CLIP encoders with a user’s high-level concept by introducing learnable textual proxies and constraining them with concept-level and LLM-derived reference-word signals. Building on this, the fourth contribution, Multi-Sub, is a framework for concept conditioned subspace proxies. It first builds a low dimensional subspace that is guided by text, using reference words suggested by an LLM, and then learns a proxy for each image inside this subspace. Representation learning and clustering are optimized together, so the method no longer needs contrastive concepts specified by the user and it also avoids the extra cost of a two stage pipeline. On publicly available visual multiple-clustering benchmarks such as ALOI, Stanford Cars, CMUface, Flowers, Fruit/Fruit360, and Cards, these methods consistently improve NMI and RI and yield partitions that better reflect user intent, with ablation studies validating each design choice. Taken together, the results illustrate how incorporating user preferences, structuring the representation space, and jointly optimizing representations and clusters can make multiple clustering systems better aligned with users’ actual goals in practice.
Deep clustering to identify subgroups of multivariate trajectories in longitudinal biomedical datasets
(2026-02-05) Vemuri, Bhargav; Tarczy-Hornoch, Peter
Unsupervised patient subgrouping in longitudinal biomedical datasets enables the discovery of distinct temporal phenotypes that capture heterogeneity in disease progression, treatment response dynamics, developmental trajectories, telemonitoring, and more. One-stage multivariate time series (MVTS) deep clustering methods are well-suited to this task because they (1) jointly model multiple longitudinal variables and (2) integrate missing data imputation, representation learning, and clustering into a unified framework. Recent state-of-the-art MVTS deep clustering approaches include Variational Deep Embedding with Recurrence (VaDER; de Jong et al., 2019) and Clustering Representation Learning on Incomplete time-series data (CRLI; Ma et al., 2021). In this work, we apply CRLI in two real-world longitudinal biomedical contexts and evaluate its performance against VaDER using 20 synthetic MVTS datasets of our own design. Our overarching question was: how and when are one-stage MVTS clustering methods (VaDER, CRLI) useful in biomedical research data exploration? In Aim 1 (Assessing the ability of CRLI to detect meaningful trajectories in a sparse, irregular, biased real-world dataset), we explored CRLI’s capacity to detect multivariate trajectories in the electronic health record (EHR). Temporal EHR data is marred by irregular measurement intervals, high missingness, and multiple biases (selection, measurement, time-related). We assessed how well CRLI handles these hurdles in the context of identifying GLP-1 medication (semaglutide, dulaglutide, etc.) treatment response subgroups in the NIH All of Us Research Study. We showed that (1) CRLI can be used to identify post-treatment multivariate response trajectories in the EHR and (2) this is possible despite a small cohort (n=336) and infrequent measurements. In Aim 2 (Assessing the ability of CRLI to detect meaningful trajectories in a high-dimensional, multimodal, prospective dataset), we applied CRLI to another real-world data source, the Adolescent Brain Cognitive Development (ABCD) Study, a longitudinal observational cohort with a prespecified assessment protocol, including a consistent follow-up schedule and a high retention rate (98.9%). This dataset allowed us to explore physical health trajectories (pubertal hormones, anthropometrics) as we did in Aim 1, but also mental health trajectories, as measured by 8 Child Behavior Checklist (CBCL) syndrome scales. We calculated cluster associations with mental health outcomes to better characterize cluster differences. We showed that (1) given longitudinal and static variables, CRLI identified longitudinal trajectories that had non-uniform associations with static variables, providing a basis for testable clinical hypotheses, and (2) CRLI identified clusters that could not have been identified with a single timepoint or single variable alone. In Aim 3 (Assessing the ability of CRLI and VaDER to detect trajectories in synthetic datasets under diverse data constraints), we designed a framework using the mockseries Python package that let us rapidly generate unique MVTS datasets by sampling from a range of values for various datasets characteristics (time series length, noise, missingness, number of clusters, number of samples). We also incorporated the ability to modify time series variable properties (trend, rate of change, seasonality) by designing 5 distinct variable styles inspired by biomedical trends we observed in Aims 1 and 2 and the literature. We reported VaDER and CRLI performance on 4 external clustering validation indices (purity, RI, ARI, NMI) across 20 synthetic datasets. We showed that (1) practitioners should be wary of novel methods that do not report performance on adjusted metrics (ARI, AMI), (2) 2D visualizations are an invaluable interpretability tool, especially when there are too many longitudinal variables to understand on an individual basis, and (3) while CRLI generally outperforms VaDER, neither method achieved across-the-board ARI dominance. Cross-cutting contributions that emerged across the aims were as follows: (1) we observed that internal clustering validation indices (Calinski-Harabasz, Silhouette, Davies-Bouldin, S_Dbw validity) were rarely concordant, making the selection of optimal cluster number in Aims 1 and 2 complicated, (2) cohort selection criteria that required a minimum number of repeat measurements across multiple longitudinal variables resulted in final cohorts that may not have generalized well to the population and/or an external validation dataset, (3) method performance in Aim 3 as measured by Adjusted Rand Index (external clustering validation index) was subpar compared to other indices that have been reported in the literature, casting doubt on trustworthiness of clusters identified in previous Aims, and (4) visual (qualitative) inspection and interpretation of identified clusters is a necessary complement to quantitative clustering result evaluation (by internal and/or external clustering validation index) for a holistic understanding of trajectory differences between clusters.
Advancing Variant Interpretation: A Gene-Specific Framework for Prioritization, Prior Estimation, and Calibration to Enhance Evidence Strength and Clinical Significance Classification
(2026-02-05) Chen, Yile; Gennari, John
The rapid growth of clinical sequencing has led to an accelerating number of Variants of Uncertain Significance (VUS), now comprising a substantial fraction of reported germline findings. While functional assays and computational Variant Effect Predictors (VEPs) contribute valuable evidence, current frameworks often treat genes uniformly, overlook gene-level heterogeneity in pathogenicity prevalence, and rely on uncalibrated or globally calibrated predictor scores. These gaps limit the consistency, accuracy, and clinical actionability of variant interpretation under ACMG/AMP guidelines. There is a pressing need for approaches that incorporate gene-specific context, integrate diverse evidence sources, and improve the calibration of computational evidence to strengthen variant classification. This dissertation introduces gene-specific informatics frameworks to improve functional assay prioritization, pathogenicity prior estimation, and the calibration of Variant Effect Predictors (VEPs), with the goal of reducing the burden of VUS in genomic medicine. By integrating statistical modeling, positive–unlabeled learning, domain-aware clustering, and adaptive calibration strategies, the work strengthens the ACMG/AMP Bayesian framework for context-aware variant interpretation. First, a gene prioritization model identifies genes where functional assays would yield the greatest clinical impact by jointly optimizing VUS “movability,” correction of potential misclassifications, and gains from computational predictors, highlighting high-value genes such as \textit{TSC2}. Second, gene-specific pathogenicity priors are estimated using a refined PU-learning method (DistCurve), supported by a complementary domain-based clustering approach for genes with limited labels. Third, a gene-aware calibration framework converts raw VEP scores into calibrated PP3/BP4 evidence strengths through a dynamic decision-tree workflow that selects the optimal strategy per gene. This gene-specific approach outperforms global calibration and, together with a per-gene mixed-predictor selection strategy, improves the accuracy and consistency of variant classification. Together, these contributions establish a context-aware decision-support ecosystem that better directs functional assay investment, provides robust statistical foundations for Bayesian interpretation, and improves the reliability of computational evidence. The resulting framework enhances the accuracy, consistency, and clinical actionability of genomic variant classification.
Developing Informatics Frameworks for Evaluating Deep Learning Algorithms for Mammography
(2026-02-05) Ramwala, Ojas Ankurbhai; Gennari, John H; Lee, Christoph I
Deep learning algorithms have played a major role in advancing AI for mammography-based breast cancer screening. Studies have shown that AI tools can achieve, and in some cases exceed, the performance of breast imaging radiologists. Integrating deep learning algorithms into clinical workflows has the potential to streamline mammography interpretation, aid early cancer detection, and enhance risk prediction. Nevertheless, in spite of promising initial performance across a broad range of tasks in mammography interpretation and breast cancer screening, their adoption in clinical settings remains limited. A major contributing factor is the lack of methods to thoroughly evaluate the safety, reliability, clinical utility, and trustworthiness of AI models in mammography, thereby creating a critical gap between algorithm development and real-world implementation. Therefore, we have developed informatics frameworks to support a comprehensive and systematic evaluation of AI algorithms prior to their clinical adoption, enabling stakeholders to critically assess model generalizability and interpret the underlying inference mechanism that drives model predictions. AI models may have limited generalizability in new clinical settings or within specific demographic subpopulations. We developed an open-source framework, ClinValAI (Clinical Validation of AI), to support health systems in implementing a cloud-based infrastructure for rigorous external validation of AI algorithms before clinical adoption. ClinValAI enables secure, privacy-preserving external validation by protecting patient imaging data and developers’ intellectual property while offering scalable, customizable workflows that accommodate the diverse computational demands of multiple AI algorithms. We demonstrate ClinValAI’s utility by performing a large-scale external validation of multiple FDA-cleared commercial AI algorithms for breast cancer detection using mammography exams from seven U.S. regional breast cancer registries. By comparing those algorithms and evaluating their performance against radiologists’ assessments, our study highlights the benefits and risks of adopting AI tools in clinical workflows. ClinValAI provides a holistic framework for validating medical imaging models and has the potential to advance the adoption of accurate, generalizable AI models in mammography-based breast cancer screening. Even when AI algorithms demonstrate strong generalizability, their 'black box' nature obscures meaningful insights into their inference mechanism, undermining trust and transparency. We address this challenge specifically for the Mirai model for mammography-based breast cancer risk prediction. We developed a method, FOCUS (Feature-space OCclusion for Understanding Saliency), to assess the contribution of different mammogram imaging regions to Mirai’s prediction. We created interpretable visualizations, FOCUS Maps, to identify the mammogram patch that most strongly influences Mirai’s risk scores. We observe that Mirai's risk estimates are primarily driven by localized imaging features and are significantly influenced by sites where cancer was subsequently detected. FOCUS Maps may assist radiologists in localizing suspicious regions for intensive imaging evaluation, such as a focused diagnostic ultrasound, bolstering its clinical utility in potentially guiding personalized screening and early intervention strategies. Although Mirai may be detecting early signs of malignancy, we also identified other localized, non-lesion imaging patterns that drive its predictions. Our explainability technique can help radiologists assess whether clinically meaningful features are associated with AI model predictions for breast cancer risk. Overall, by developing informatics frameworks for external validation and model explainability, our work supports a comprehensive evaluation of the generalizability and trustworthiness of AI tools, potentially enabling the clinical adoption of AI tools to improve healthcare outcomes and promote health equity.
Can I Afford College? How Students’ Perceived Ability to Pay Shapes Postsecondary Outcomes and the Role of Policy
(2026-02-05) Terry, Ellyn; Hill, Heather
Low- to moderate-income children who are qualified for college are far less likely to attend college and graduate than high-income children due, in part, to concerns about the cost of college (Advisory Committee on Student Financial Assistance, 2010). While many financial aid resources can make college affordable to many low-income children, the complexity of aid programs and eligibility requirements makes it difficult to understand the cost of higher education, whether it's affordable, and which colleges are affordable (Goldrick-Rab & Kolbe, 2016). When students lack clarity on the affordability of college, they may think they cannot afford it and decide not to pursue higher education. While prior research on financial aid and college access acknowledges the importance of student’s perceived ability to pay for college in shaping educational outcomes, it has not empirically evaluated this construct as a distinct mechanism, separate from actual ability to pay, and has rarely considered how policy can influence it. This dissertation addresses this gap by exploring how high school students’ perceived ability to pay for college shapes their postsecondary outcomes and the role that policy can play in shaping high school students’ perceived ability to pay, thereby affecting their college-going expectations, and college enrollment. Chapter 1 of this dissertation models the relationship between a student’s socioeconomic status (SES) when in high school, perceived ability to pay for college, and their later post-secondary outcomes (i.e. enrollment and degree attainment) using a structural equation model and a large nationally representative longitudinal survey spanning 2009 to 2021. Findings indicate that perceived ability to pay mediates approximately 38% of SES’s effect on students’ college-going expectations and between 8% and 19% of its effect on postsecondary enrollment and degree attainment, depending on the outcome measured. The findings suggest that public policies aiming to improve postsecondary outcomes among low-SES students must address not only financial barriers but also students’ perceptions of their ability to pay. Chapter 2 of this dissertation examines the implementation of a new child development policy in California, which aims to support college-going among low-income students by providing universal, progressive education savings accounts. Using a mixed-methods approach—including administrative data analysis, surveys, and interviews with program participants and staff—I explore how CalKIDS is being implemented, how students perceive the program, and what barriers may be limiting account claiming and fund use. Findings reveal that while the process of claiming accounts is generally easy for students, limited program awareness remains a primary obstacle to broader reach, largely due to constrained marketing resources and limited staff. Additionally, program administrators raised concerns that fund utilization among the eligible population is significantly lower than expected. Investigating this, I find that young adults who are eligible to use their funds are not experiencing issues with withdrawing funds—instead, many students choose to save their funds for future use, perceiving CalKIDS as a form of emergency savings for later in their educational journey, particularly when transferring to four-year institutions. Last, I highlight several program design features which create implementation challenges and may reduce the value of the program—such as logistical delays in processing distribution requests and confusion about eligibility rules—and provide recommendations for improvement. Despite these limitations, I find that students overwhelmingly view CalKIDS as a helpful and motivating source of support. Chapter 3 of this dissertation examines the effect of a new FAFSA completion policy in California that aims for every high school senior to complete a FAFSA or the state financial aid form. This policy represents a major effort to increase college-going by ensuring that students learn their specific net cost of college, after accounting for financial aid. Using a panel of nearly all California high schools from 2017–18 through 2023–24, I estimate the policy's effects on FAFSA submissions, financial aid awards, and college enrollment using two quasi-experimental methods. Specifically, I use difference-in-differences and instrumental-variable designs to identify the causal impact of the policy. Results show that the policy significantly increased FAFSA and Cal Grant submissions, but had no effect on Cal Grant awards, and only modest positive effects on college enrollment. This research aims to provide critical insights for state, local, or national policymakers seeking to improve educational access for young adults from low-income backgrounds. By analyzing the role that perceived ability to pay for college plays in shaping college-going expectations and enrollment, Chapter 1 illuminates how policies and interventions can improve academic outcomes by targeting students’ perceived ability to pay. By conducting a process evaluation of a CDA policy, Chapter 2 improves our understanding of how students perceive the benefits of CDA programs, identifies areas for strengthening CDA program implementation in California, and informs the work of agencies nationwide that implement CDAs. Last, by examining the effect of a FAFSA completion policy in Chapter 3, this research illuminates how such policies can increase college enrollment by helping students understand the net cost of college after financial aid is applied. Ultimately, by understanding the complex interplay between socioeconomic status, perceived ability to pay for college, educational attainment, and policy interventions, it is possible to help promote a more equitable and inclusive society where every child has the opportunity to climb the economic ladder.
Harnessing Language Models for Automated Detection of Depression Severity and Suicide Risk
(2026-02-05) Ren, Xinyang; Cohen, Trevor A
Depression is one of the most common mental disorders globally, and can carry an increased risk of adverse outcomes including suicide. Suicide is one of the leading causes of death worldwide and many more individuals attempt it or experience suicidal thoughts. Compounding these severe public health problems is a longstanding shortage of mental health professionals. There are too many patients for available professionals to monitor effectively, presenting opportunities for the use of technology to expand their capacity. Natural language processing (NLP) methods have been widely applied to psychologically-related text analysis tasks to draw relationships between text and the thoughts and feelings of the person who generated it, as indicators of their mental status. Two prominent areas are the detection of depression symptom severity and suicide risk. In this work, I investigated how language models can be harnessed to automatically detect depression symptom severity and suicide risk. To get the numerical representation of the text for further analysis, one common method is to extract contextual embeddings from language models. Contextual embeddings are word representations that take surrounding context into account, which can better represent the complexities of linguistic expressions than models that represent the same word the same way, regardless of their context. However, there is limited research involving clinical populations that utilize contextual embeddings from state-of-the-art language models to detect linguistic indicators of depression and suicide risk. Moreover, certain patient-generated data sources that can reveal mental status, notably text-based therapy sessions, Google search logs, and YouTube activities, remain underexplored. Relevant research has primarily concentrated on electronic health record (EHR) data and social media posts, which are subject to certain limitations. Furthermore, despite the rapid development of large language models, their clinical application remains challenging due to high computational costs and ethical concerns. To fill these gaps, I have developed a series of methods for automatic depression symptom severity and suicide risk detection or prediction utilizing state-of-the-art language models with under-explored data sources. Specifically, I have analyzed the use of contextual embeddings of first-person singular pronouns as predictors of depression symptom severity. Positive classification results on a PHQ-9-derived binary outcome were obtained when applying the methods to the deidentifiedpsychotherapy messages. To explore the use of individualized web searches for suicide risk assessment, I have evaluated the effectiveness of anomaly detection methods in identifying search pattern changes that precede a suicide attempt using Google search data. The proposed framework for semantic feature construction, which consists of initial filtering with a small language model followed by adjudication with a more advanced large language model to assess relevance to suiciderelated constructs, provides a computationally efficient, tractable approach that can be applied to web search logs at scale. The methods were further applied to study participants’ YouTube activity data, which were combined with Google search logs in order to enhance anomaly detection performance. This work demonstrates the potential of effectively using language models for automatic prediction of depression symptom severity and detection of suicide risk using real-world datasets. It helps bridge the gap between advances in NLP and the growing need to enhance mental health service capacity, offering scalable computational tools for timely risk detection and intervention.
Leveraging Large Language Models for Clinical Information Extraction in Radiology Reports
(2026-02-05) Park, Namu; Yetisgen, Meliha
Medical imaging plays a central role in diagnosing, monitoring, and managing a wide spectrum of diseases, including cancer, cardiovascular disorders, neurological conditions, and musculoskeletal abnormalities. Radiologists interpret complex imaging data and summarize their findings in narrative reports, which remain largely unstructured. The rapid expansion of imaging utilization has led to an overwhelming volume of such reports, posing significant challenges for clinical decision support. Their unstructured format limits automated analysis, secondary use, and integration into downstream clinical workflows. This dissertation addresses two major barriers to the effective use of radiology reports in data-driven clinical systems: the absence of publicly available, large-scale annotated corpora of radiology reports with detailed clinical findings suitable for training supervised models, and the limited application of machine learning approaches, particularly large language models (LLMs), to real-world clinical tasks at scale. To overcome these challenges, the research is organized around three core aims: developing a corpus of radiology reports annotated with detailed clinical findings and designing an advanced information extraction framework optimized for radiologic text; evaluating the performance of diverse machine learning approaches, with emphasis on LLMs, for the practical task of identifying follow-up imaging recommendations; and constructing a large-scale repository of incidental findings (incidentalomas) derived from the model outputs and proposing an NLP-based framework for automated incidentaloma detection to enhance clinical decision-making. Collectively, this work contributes a high-quality annotated dataset for radiologic text analysis and demonstrates the feasibility and utility of large language model approaches for transforming unstructured radiology reports into structured clinical intelligence, advancing the integration of medical imaging data into precision healthcare.
Strengthening Respectful Maternal and Newborn Care: Advancing Measurement and Implementation Science Across Contexts
(2026-02-05) Mehrtash, Hedieh; Sherr, Kenneth
Respectful maternal and newborn care is an essential component of high-quality health services, yet women across diverse settings continue to report mistreatment, poor communication, and unmet expectations during childbirth. These concerns are especially pronounced in fragile and humanitarian contexts, where health system constraints can undermine both clinical quality and women’s experiences. Despite growing global emphasis on respectful care, validated measures that capture key aspects of women’s experiences remain limited, and little implementation science evidence exists for how respectful care can be operationalized globally, particularly in Middle Eastern health systems. This dissertation addresses these gaps through three studies: development and validation of mistreatment measures in the occupied Palestinian territory (oPt), validation of a multidimensional satisfaction with childbirth care scale using multi-country data from a World Health Organisation (WHO) study, and a qualitative study exploring health worker perspectives on implementing a labor companionship intervention in three Middle Eastern hospitals. The first study developed concise, domain-specific mistreatment scales using Item Response Theory (IRT) applied to survey data from 745 postpartum women in the occupied Palestinian territory (oPt). For each domain, a full item set (all items) and a brief item set (reduced subset) was evaluated to determine whether shorter item sets retained comparable psychometric properties. Physical abuse and stigma items were excluded as a result of less than 1% responses, while domains related to poor rapport, failure in professional standards of care, and health system conditions and constraints showed strong item discrimination and adequate model fit. Each of these domains produced a brief scale consisting of 3 to 4 mistreatment items, retaining strong psychometric performance. The brief scales performed comparably to the full item set of the community survey, and importantly, odds ratios for associations with dissatisfaction with care were similar for both full and brief scales. For example, poor rapport was associated with 2.8 times higher odds of dissatisfaction using the full item set and 2.7 times higher odds using the brief item set. Failures in professional standards demonstrated a similar pattern, with odds ratios of 2.3 for the full item set and 2.2 for the brief item set. These consistent effect sizes support the feasibility of integrating short, validated mistreatment measures into routine monitoring and accountability systems in constrained settings. The second study validated a satisfaction with childbirth scale using data from 2,672 postpartum women in Ghana, Guinea, Nigeria, and Myanmar. Exploratory and Confirmatory Factor Analyses confirmed a two-factor structure representing Interpersonal Satisfaction and Structural Satisfaction, with Cronbach’s alpha values of 0.82 and 0.71 respectively. Mistreatment exposures were strongly associated with dissatisfaction across both domains. Women who experienced any mistreatment had 1.5 times higher odds of structural dissatisfaction and 2.8 times higher odds of interpersonal dissatisfaction. Specific forms of women-reported mistreatment demonstrated notable associations with the satisfaction scales. Physical abuse increased the odds of dissatisfaction by 1.5 times for structural and 1.9 times for interpersonal experience. Verbal abuse increased the odds by 1.5 and 2.9 respectively. Denial of companionship produced odds ratios of 2.1 for structural dissatisfaction and 3.1 for interpersonal dissatisfaction. Women reporting lack of health worker responsiveness to their needs showed the strongest association, with 11.1 times higher odds of interpersonal dissatisfaction. These patterns provide strong criterion validity and demonstrate that the validated satisfaction scale can meaningfully differentiate women’s experiences across diverse health systems. The third study examined health worker perceptions of implementing a labor companionship model in tertiary hospitals in Egypt, Lebanon, and Syria using qualitative methods guided by the Consolidated Framework for Implementation Research (CFIR). This framework informed exploration of how individual, inner setting, outer setting, and process-level determinants shaped health worker experiences. Health workers described indicated that heavy workloads, overcrowded labor wards, limited staffing, and inconsistent facility policies were major barriers to providing respectful, person-centered care. Despite these challenges, health workers consistently recognized the value of labor companionship for improving communication, reducing anxiety, and enhancing women’s emotional support. Health workers emphasized the need for clearer guidance on lab, supportive supervision, adequate staffing, and training. These findings show that implementing respectful care is shaped by both individual motivation and structural conditions highlighting the importance of the multi-level strategies required for sustainable implementation in complex health system environments. Taken together, the three studies in this dissertation advance the measurement and implementation of respectful maternal and newborn care across diverse settings. The validated mistreatment and satisfaction scales provide robust, context-sensitive tools for monitoring women’s experiences, while the implementation findings identify actionable strategies to strengthen provider behavior, institutional accountability, and feasibility of implementing respectful care interventions. This body of work offers an integrated, evidence-based framework for promoting dignity, equity, and person-centered maternity care across contexts.
Biosimilars in the United States: Market Dynamics and Patient Out-of-Pocket Costs
(2026-02-05) Dayer, Victoria; Sullivan, Sean D
BackgroundThe United States biosimilars market is growing rapidly and has led to significant savings for the health care system and payers. However, market complexities, barriers to entry, and fewer biosimilars in the pipeline than expected has led to concerns about the long-term sustainability of the biosimilars market and therefore threatened savings. We characterized eight physician-administered biologics markets in terms of price declines and reference product market share to better understand the heterogeneity between these markets. We also estimated the effect of biosimilar competition on patient out-of-pocket costs for rheumatology biologics, to understand whether biosimilar competition is truly resulting in savings for patients. MethodsFor Aim 1, quarterly average sales price (ASP) data from the Centers for Medicare and Medicaid Services (CMS) Drug Pricing Files along with volume data from the standard Medicare Limited Data Set (5% sample) for Medicare and the MerativeTM MarketScan® Commercial Database for the commercial population were used as data sources, using price and volume (number of claims) for the following biologics: filgrastim, rituximab, infliximab, trastuzumab, pegfilgrastim, epoetin alfa, bevacizumab, and ranibizumab. We first characterized the markets descriptively, then used linear regression to assess the relationship between reference product market share and each biosimilar entry and to study biosimilar versus reference product price decline over time since biosimilar entry. For Aim 2, the MerativeTM MarketScan® Commercial Database was used to conduct an analysis of patient OOP costs before and after biosimilar competition using the methodology described by Callaway & Sant’Anna. We included infliximab, rituximab, and adalimumab as the drug groups of interest, as they are the rheumatology biologics with biosimilars available. The exposure was biosimilar competition and the outcome was biologic-related OOP costs. ResultsIn Medicare, reference product market share declined consistently, though to varying degrees across molecule groups. In the commercial sector, market share trends were more variable, with some reference product markets remaining stable and others declining steeply. Price trends also differed substantially between payer sectors. In Medicare, biosimilars showed an 18.6% (p<0.01) greater price decline on average than reference products, while in commercial markets the overall result was not significant. Price decline patterns varied widely between the different biologics. For patient out-of-pocket costs, we observed statistically significant savings due to biosimilar competition, with an average savings of over $1,000 in the infliximab and rituximab groups over the study period. When assessing treatment effect by duration of exposure, we observed increasing savings over time, the longer biosimilars were available. In the adalimumab group, there was an increase in patient OOP costs in the year prior to biosimilar availability, which may reflect anticipatory price increases on the part of the reference manufacturer. There was minimal savings in the adalimumab group upon biosimilar entry, and no savings when accounting for one year of anticipation. ConclusionThe first study provides evidence of the differences between biologic markets in both market share and price decline trends, and the challenge of generalizing predictions or drawing generalizable conclusions about biosimilar market behavior. The second study indicates that the availability of infliximab and rituximab biosimilars led to significant savings for patients in addition to the health care system as a whole, with more savings the longer biosimilars are on the market. Together, these studies highlight the importance of policies that incentivize development of biosimilars and support the long-term availability of biosimilars to preserve and increase savings for patients.
Emergent Needs and Interest in New Services among Individuals with Opioid Use Disorder- A Multi-Method Analysis
(2026-02-05) Singh, Samyukta; Banta-Green, Caleb J
Over 21.9 million people worldwide and 2.5 million in the U.S. alone have opioid use disorder (OUD). The opioid epidemic has been exacerbated by the prevalence of illicitly manufactured fentanyl (IMF), which is associated with significant opioid related morbidity and mortality. In 2023, fentanyl use contributed to 78% of overdose deaths. The standard of care treatment for OUD is medication for OUD (MOUD). However, MOUD is persistently underutilized, and there is a large treatment gap. In 2021, only 22.3% of people with past-year OUD reported receiving past-year MOUD. The increasing prevalence of fentanyl underlines the critical need for people to be connected to MOUD, given fentanyl’s potency and addiction potential. The overall purpose of this study was to identify emergent needs and interest in new services among individuals with OUD. The specific aims of the study were to 1) determine the difference in time to onset of moderate-to-severe OUD, among adults who report using fentanyl versus heroin, 2) examine the association between fentanyl compared to heroin as the primary drug of use and interest in methadone and buprenorphine treatment, and 3) evaluate the acceptability of a low-barrier model of care for clients seeking treatment for opioid use disorder at five community clinical sites, from client and staff perspectives. For the first two papers, I used data from a cross-sectional survey that utilized respondent-driven sampling. For the first paper, I used survival analysis to assess the timing of onset of OUD for people using fentanyl compared to heroin. Using a Cox proportional hazards model, people who used fentanyl were found to have a significantly higher hazard of OUD onset, compared to users of heroin (HR=1.49, 95% CI [ 1.12, 1.98], p=0.006). Findings highlight the potency and rapid addiction potential of fentanyl, underlining the need for rapid initiation of MOUD among this high-risk population. For the second paper, I used descriptive analyses and logistic regression to assess interest in methadone and buprenorphine among people who use fentanyl or heroin. The exposure groups were characterized by primary opioid type (fentanyl or heroin) and route of administration (injected or smoked). A substantial majority of individuals across the three groups reported interest in methadone (59-79%) and in buprenorphine (37-47%). Using logistic regression, individuals who primarily reported injecting fentanyl were found to have significantly higher odds of interest in methadone treatment compared to those who primarily reported injected heroin (aOR=2.81, 95% CI [1.28, 6.50], p = 0.012). Interest in buprenorphine was not found to differ significantly by opioid use pattern. High interest in MOUD highlights the need to expand low-barrier MOUD access. Differences in patient preference by primary opioid type and route of administration should be incorporated during shared decision-making and planning care. For the third paper, I used a qualitative approach that included semi-structured interviews to examine the acceptability of a medication-first low-barrier model of care from 27 clients and 15 staff at 5 clinical sites. Clients and staff both endorsed their acceptability of the low-barrier model of care, citing the relational strengths of the model. Further, the contingency management component of the model was welcomed by people who used stimulants, independently or in combination with opioids. Findings provide evidence for the expansion of medication-first, low-barrier models to increase MOUD access among individuals with OUD and/or stimulant use disorder. Together, findings outline emerging needs and interests in OUD services among individuals with OUD, identify populations most at risk of OUD-related morbidity and mortality, inform improvements, and provide evidence for medication-first low barrier models that can increase uptake and retention in OUD care.
Pardon the Interruption: Assessing the Implementation, Operation, and Sustainment of Hospital-Based Violence Intervention Programs in the United States
(2026-02-05) Almquist, Lars; Helfrich, Christian D
Background: Hospital-based violence intervention programs (HVIPs) are public health interventions to prevent violent re-injury. We know little about the experience of HVIPs during the early stage of implementation. Specifically, we do not understand the factors that help or hinder programs from achieving stable operation. A threat of survivorship bias exists, as programs failing to reach operational status are not represented in the literature. HVIPs hinge on the ability of violence prevention professionals to assist the recovery of intentionally injured patients to prevent further spread of violence. Yet, we know little about the tactics these professionals use, or whether the tactics developed by one violence prevention professional transfer to another. While there is published guidance on initial program implementation, there is limited evidence or guidance about factors influencing long-term HVIP sustainability. Methods: Semi-structured interviews were conducted with 18 HVIP leaders regarding the barriers and facilitators to program implementation (Study #1). Interviews were conducted between December 2023 and March 2025 and were organized around a nine-stage blueprint for starting an HVIP, developed by the American College of Surgeons. Leaders were asked to articulate the barriers and facilitators encountered during each stage of implementation in as much detail as possible. Inductive coding was used to identify themes emerging at each stage. Semi-structured interviews were also conducted with hospital-based violence prevention professionals to identify tactics used in everyday work (Study #2). Interviews were organized around 10 “hinge points” on the patient’s recovery continuum deemed integral to program success, identified a priori. Inductive thematic analysis was used to identify individual tactics. The Action Actor Context Target Time (AACTT) rubric helped ascertain essential information about each tactic. Finally, a three-round Delphi study was conducted with leaders of established programs to prioritize factors critical for achieving long-term HVIP sustainability (Study #3). Participants submitted factors influencing their program’s sustainability in Round 1. Responses were synthesized using inductive thematic analysis and returned to participants for refinement in Round 2. Maximum-difference scaling with hierarchical Bayes estimation helped prioritize factors by importance in Round 3. This dissertation addressed these gaps in our knowledge through primary data collection across three studies:Study #1: assessed the barriers and facilitators facing HVIPs actively in the early stages of program implementation. Study #2: identified the tactics violence prevention professionals use to meet the recovery needs of violently injured patients while preventing future violence exposure. Study #3: developed a prioritized list of factors influencing the achievement of long-term HVIP sustainability. Results: Barriers to implementation included insufficient program infrastructure to secure outside investment in an HVIP; challenges hiring violence prevention professionals with criminal histories; and maintaining relationships with community organizations mistrustful of the hospital system. Key facilitators included early identification of executive-level hospital champions; hospitals budgeting for initial HVIP funding; and external capacity-building support to grow program infrastructure. Interviews in Study #2 surfaced 214 tactics used by violence prevention professionals. Tactics addressing the initial bedside encounter (n=49) and trustbuilding with patients and families (n=44) represented both the largest and most diverse share of those identified (n=96). Navigating administrative bureaucracy was particularly challenging and required a distinct set of tactics. (n=39) Comparably few tactics engaged patient retention (n=11) or aftercare as patients exited the program (n=3). Finally, 27 sustainability factors were initially synthesized from 108 submissions by 32 participants in Round 1. Leaders added four factors and removed three during Round 2, with 28 factors rated in Round 3. Participants prioritized frontline violence prevention professionals in six of the first nine factors. Funding-related factors (e.g., government grants, operating support) received moderate priority. Administration (e.g., hospital leadership) and community stakeholders (e.g., community champions) received lower priority. External institutions (e.g., police) received lowest priority. Significance: Study #1 represents the first known attempt to identify and describe the barriers and facilitators influencing early HVIP implementation. These findings may equip nascent HVIPs to recognize and respond to factors that accelerate or hinder implementation. Study #2 is the first known study focusing on the role of the hospital-based violence prevention professional. Dissemination of tactics used to conduct their work will strengthen the skillsets of current HVIP professionals, while enhancing the training of future violence prevention personnel. Findings may support the creation of practical, readily deployable toolkits to translate tactical insight to diverse contexts where HVIPs operate, including HVIPs not yet established. Finally, Study #3 represents the first known study of HVIP sustainability. Priorities for program stability differ from priorities in blueprints for program startup (Study #1). Results may indicate the need for program adaptability during their implementation journey and for HVIP leadership to recalibrate priorities over time.
Disability Resources for Students (DRS) Services Utilization Among School of Public Health Students
(2026-02-05) Wajdik, Chandra Dawn; Spigner, Clarence
Abstract: University students with disabilities have more access to higher education yet significant barriers to full utilization of accommodations exist. We used a survey methodology to assess University of Washington Public Health student opinions of student disability services on campus. I found expected proportions of different genders, ethnicities, and frequency of individuals who self-identify as disabled. The data also revealed that one-third of Public Health students receive accommodations. Importantly, one-third of those who receive accommodations find it difficult to get them. Working with accessibility support staff and faculty can help alleviate this burden.

Browse

Recent Submissions