22689 research outputs found
Sort by
Enabling Conditional Generation Without Training: Plug & Play Generative Modelling using Diffusion Models
No full textDiffusion models have achieved state-of-the-art results across various generative modeling tasks, including text, image, video, and audio generation. By iteratively refining standard Gaussian noise into structured signals, these models offer remarkable flexibility. However, their practical capabilities remain constrained by the limited availability of large-scale annotated datasets, particularly for complex and compositional tasks that require adherence to specific rules. These limitations restrict their applicability to relatively simple problems with smaller datasets. In this thesis, we propose a series of methods to reduce the dependency on large-scale data, enabling diffusion models to solve complex inverse problems more effectively. First, we introduce Unite and Conquer, a fully diffusion-based framework that decomposes complex compositional tasks into simpler subproblems, reducing data requirements by a logarithmic factor at the cost of increased computation. By leveraging the product of experts approach from probability theory, we train individual diffusion models to solve these simpler subproblems and then combine their capabilities during inference, enabling the resolution of complex tasks without requiring extensive annotated datasets. Second, we eliminate the need for training multiple individual diffusion models by utilizing a single unconditional diffusion model. By integrating problem-specific physical constraints and inverse dynamics, we introduce Steered Diffusion, which replaces the need for separate models with optimization iterations embedded within the generation process. Third, we enhance Steered Diffusion with DreamGuider, a method that removes its memory overhead and reliance on handcrafted parameters. By incorporating zeroth-order optimization and automatic parameter estimation, DreamGuider enables conditional generation with minimal computational overhead, eliminating manual tuning while improving performance. Finally, we introduce MaxFusion, a method for compositional and controllable text-to-image generation in diffusion models. MaxFusion enhances controllability in complex scenarios where multiple 3D and 2D conditioning inputs must be simultaneously incorporated. Collectively, these contributions advance the conditional generation capabilities of diffusion models, reducing their dependence on large-scale annotated datasets and enabling physics-based, interpretable, and controllable generation
Tuning PID Controllers for HVAC Temperature Modulation of a Pharmaceutical Cleanroom Using MATLAB Simulink
Full text linkManufacturing processes employ control systems to ensure the reliability and quality of goods; these systems are particularly crucial to industries like pharmaceuticals, where stringent environmental conditions are essential. In pharmaceutical cleanrooms, maintaining precise temperature and humidity levels safeguards product integrity and sterility. Proportional-integral-derivative (PID) control is a widely adopted method for regulating HVAC systems in cleanrooms, providing efficient control over environmental parameters.
This study focused on simulating the PID temperature control response of an HVAC system in a pharmaceutical cleanroom using MATLAB Simulink. By analyzing temperature profiles obtained from real-world operations, various PID tuning methods were evaluated for their impact on system stability and transient response. Simulation allowed for the comparison of different tuning strategies without disrupting production processes or altering existing control systems, and further applications could see tuning from simulations directly programmed into site control systems. The findings contributed to enhancing the understanding of PID control dynamics in cleanroom settings, aiming to optimize system performance in pharmaceutical manufacturing
DEVELOPMENT AND VALIDATION OF A MODERN PRESCRIPTION-BASED RISK SCORE TO PREDICT HEALTHCARE SPENDING
No full textPrescription-drug based risk scores offer an important alternative to diagnosis-based risk scoring systems when diagnostic information is unavailable; however, their value depends upon a given score’s ability to capture the current therapeutic landscape. We performed a retrospective cohort study using Merative MarketScan administrative claims data from continuously enrolled, commercially insured adults between January 2021 and December 2022 to develop and validate a modern, prescription-based risk score for predicting healthcare expenditures. Prescription use was described using over 300 binary drug categories, and total healthcare spending was the primary outcome. We characterized the cohort with descriptive statistics and fit several machine learning models that vary in their approaches to regularization and decision-making, including elastic net regression, random forest, and extreme gradient boosting, using 60% of the sample for model development and 40% for validation. Predictor weights were assigned based on beta coefficients from the elastic net regression model, with expenditures modeled on a logarithmic scale. The drug-only risk score achieved an R2 of 0.135 in the validation dataset, improving to 0.173 with the addition of age and sex. These results are comparable to many existing prescription-based risk scores predicting future total healthcare expenditures. Our findings underscore the potential of updated prescription-based risk scores for expenditure prediction and support further research to optimize model performance
Investigating The Association Disinfectant Byproducts In Drinking Water And The Microbiome In Childhood
No full textThe gut microbiome plays a vital role in childhood development and health. Disinfection byproducts (DBPs), including trihalomethanes (THMs) and halo acetic acids (HAAs), formed during drinking water chlorination, may alter gut microbial diversity, but their impact on children remains unclear. This study assessed the association between DBP exposure and gut microbiome alpha diversity among 515 children aged 2–18 years in the Environmental influences on Child Health Outcomes (ECHO) cohort. Water DBP concentrations were estimated by linking residential geocodes to EPA compliance records. Gut microbiome alpha diversity, measured by Shannon index and Observed Features, was assessed through 16S rRNA sequencing. Linear mixed-effects models were applied, adjusting for demographic, socioeconomic, and technical factors, with additional sensitivity and stratified analyses. Higher THM exposure was consistently associated with lower gut microbial diversity across models, although not statistically significant. HAA exposure showed weaker and inconsistent associations. Stratified analyses indicated that socioeconomic status (SES) may modify the effects, with different patterns of association across SES groups. Sensitivity analyses among tap water drinkers supported the robustness of findings but had wider confidence intervals.
In conclusion, THM exposure may be associated with modest reductions in gut microbiome diversity in children. Although not statistically significant, the consistent trends and SES differences highlight the need for further longitudinal research to better understand DBP-related microbiome alterations during critical developmental periods
HETEROGENEITY OF TREATMENT EFFECTS OF GLUCAGON-LIKE PEPTIDE-1 (GLP-1) RECEPTOR AGONISTS ON WEIGHT LOSS: A RETROSPECTIVE COHORT STUDY
Full text linkBackground: While there is high quality evidence supporting the effectiveness of glucagon-like peptide-1 receptor agonists (GLP-1) for the treatment of obesity, less is known regarding how such effectiveness may vary across different patient subpopulations.
Objectives: To quantify heterogeneity in weight loss response among GLP-1 users, stratified by patient age, sex, race/ethnicity, diabetes, and baseline body mass index (BMI).
Methods: We used TriNetX, a database containing electronic health records from 81 academic medical centers in the United States, to identify adults with at least 6-month continuous GLP-1 use between September 2014 and November 2023. Our primary outcome was percentage change in BMI from baseline, defined as the most recent BMI measurement within 6 months prior to GLP-1 initiation, overall and across different subpopulations of interest. We used linear mixed-effects models with natural splines to examine BMI changes over time, adjusting for individuals’ demographics and clinical characteristics.
Results: Among 23,291 GLP-1 users, the mean baseline BMI was 37.3 ± 7.7 kg/m² and 65.2% were female. During the 6-month follow-up period, 42% achieved ≥5% BMI reduction. There was no statistically significant reduction in BMI between older (≥65 years) and younger adults (-4.1% vs. -4.7%, p=0.132). There were modest, though statistically significant, differences in BMI reduction by sex (females: -5.1% vs males: -3.5%, p<0.001) and race (Black: -4.1% vs White: -4.8%, p<0.001). Individuals with no diabetes experienced substantially greater weight loss than patients with diabetes (-6.7% vs -3.5%, p<0.001). Compared to patients with baseline BMI<30 (-2.0%), patients with BMI 30-35 showed greater reduction (-4.5%, p<0.001), and those with BMI≥35 showed the largest reduction (-5.2%, p<0.001).
Conclusions: Among this large, diverse cohort of adults in the United States, real-world effectiveness of GLP-1 was greater among women, Whites, individuals without diabetes, and those with higher baseline BMI. These findings can be used to inform further research as well as treatment selection among those who may be eligible for pharmacologic treatment of obesity
Health and Quality-of-Life Impacts of Industrial Food Animal Production Among Workers and Community Residents
Full text linkIndustrial food animal production (IFAP) is characterized by large-scale operations that are densely-stocked for the maximization of production and profit. Billions of animals are raised, slaughtered and sold in the United States annually. With these large numbers of animals comes millions of gallons or tons of waste. Emissions from these facilities can contaminate the air, water and soil near these operations, causing a multitude of health and quality-of-life effects. The goal of this dissertation was to quantify and explore the spatio-temporal distribution of the air, surface and health impacts of proximity to IFAP. Each aim utilized either a swine production intensity (SPI) metric, a poultry production intensity (PPI) metric, or both, calculated using the geolocation of each operation with respect to each data point, and the either the steady state live weight or animal count of each operation. We measured gas emissions, methane (CH4), hydrogen sulfide (H2S), and ammonia (NH3), from IFAP waste structure using cars equipped with gas concentration analyzers in North Carolina, the Delmarva Peninsula and in the Baltimore-Washington region. We quantified swine fecal filth inside and outside community members’ homes using Pig-2-Bac, a swine-specific Bacteroidales genomic DNA marker in North Carolina. Lastly, we developed a novel bead-bead salivary immunoassay to measure immune response to 48 antigens (38 pathogens) and applied it to a study of workers in IFAP, neighbors of IFAP and Metro residents in North Carolina. Increasing SPI and PPI increased the odds of gas plume detection and proximity to IFAP biofuels waste infrastructure resulted in large plumes of H2S and CH4. In addition to air pollution, swine fecal filth was detected both inside and outside of community residents’ homes and was associated with increasing SPI. Lastly, immune responses to biological emissions from facilities, like bacteria, viruses and parasites, were higher in workers in these facilities and IFAP neighbors compared to Metro residents. These findings provide more evidence of the impact of emissions from IFAP facilities, as well as necessitate the need for a cumulative impacts approach to understanding the burden of exposure of communities who live near IFAP
THE ASSOCIATIONS BETWEEN MAJOR DEPRESSIVE DISORDER, BREAST CANCER TREATMENT AND OUTCOMES
No full textBreast cancer remains a leading cause of morbidity and mortality among women in the United States, with disease recurrence being a major concern. Emerging evidence suggests that mental health may impact disease progression and survival. Depression, in particular, is reported in up to 60% of women with breast cancer, with the prevalence of clinically diagnosed major depressive disorder (MDD) ranging from 10% to 35%. Despite the potentially high prevalence of depression, the direction and magnitude of the relationship between depression and breast cancer outcomes remains to be clearly defined. Depression has been linked to increased breast cancer mortality in a few observational studies. Studies investigating the association between depression and breast cancer recurrence are inconclusive and have limitations in data ascertainment and small sample sizes.
The goal of this dissertation was to generate robust data on the associations between pre-diagnostic MDD and breast cancer recurrence and mortality (Aim 1) and investigate whether non-adherence to either antidepressants (Aim 2) or endocrine therapies (Aim 3) contribute to disease progression among women with MDD and breast cancer. We used electronic medical records and pharmacy data from the Veterans Affairs (VA) Healthcare System for these analyses given that the VA mandates annual depression screening, reporting national screening rates of 98%.
We observed significantly higher hazards of breast cancer recurrence and mortality among women with MDD compared to those without after adjusting for prognostic factors (Aim 1). Non-adherence to antidepressants or endocrine therapies modified the hazard of recurrence in women with MDD. Non-adherence to antidepressants in the two-years prior to cancer diagnosis was associated with an increased hazard of recurrence compared to women without MDD, whereas those who adhered had a recurrence hazard comparable to women without MDD (Aim 2). Similarly, non-adherence to endocrine therapy following cancer diagnosis was associated with an increased hazard of recurrence compared to women without MDD, whereas those who adhered had a recurrence hazard comparable to women without MDD (Aim 3).
This work provides strong evidence of an association between MDD and disease progression. It also identified non-adherence to antidepressants and endocrine therapies as risk factors for disease progression
Discerning glucose patterns and variability using continuous glucose monitoring in older adults
Full text linkThis dissertation evaluates the clinical utility of continuous glucose monitoring (CGM), questionnaires and short-term biomarkers in the management of diabetes in older adults, an understudied population. It provides insight into the glucose patterns and variability in older adults with and without diabetes, their burden of symptoms of hypoglycemia, and the use of alternate markers of glycemia.
First, we created a Stata package for the analysis of CGM data. Our package generates CGM summary measures for data from all CGM systems and allows the user to flexibly define ranges and generate data visualizations. The package supports rigorous and reproducible analyses of CGM data and was used for all analyses in this dissertation.
Second, we characterized the prevalence of CGM-defined glucose abnormalities in very old adults with and without diabetes. We found hyperglycemia detected by CGM was rare in very old adults with a normal hemoglobin A1c (HbA1c) or HbA1c in the prediabetes range. This suggests that HbA1c adequately captures the burden of hyperglycemia for most older adults for which ARIC participants are representative.
Third, we quantified the burden of self-reported symptoms of hypoglycemia in very old adults with and without diabetes. We found the burden of symptoms of hypoglycemia was similar among persons with and without type 2 diabetes (T2D) and did not differ by high-risk glucose-lowering medication use. These findings raise questions regarding the questionnaire used to obtain symptoms of hypoglycemia in older adults.
Fourth, in adults with diabetes, we evaluated the concordance of biomarkers of glycemia with CGM metrics of short-term glycemic control and glucose excursions. We found that glycated albumin and fructosamine were moderately associated with CGM-defined metrics and performed similarly to HbA1c in detecting target times in and above range. We found that 1,5-anhydroglucitol may be useful as a biomarker of sodium-glucose cotransporter 2 inhibitors adherence in clinical research and trials.
Our findings revealed that HbA1c adequately captures the burden of hyperglycemia for most older adults, a questionnaire used to capture symptoms of hypoglycemia may not be valid in very old adults with T2D, and fructosamine and glycated albumin are useful alternate markers of glycemia
MULTIDIMENSIONAL STRESSORS: INDIVIDUAL PHYSIO-PSYCHOLOGICAL AND PSYCHOSOMATIC REACTIONS TO STRUCTURAL RACISM
No full textThis dissertation explores how structural racism manifests at levels—individual, interpersonal, systemic, and institutional—within healthcare and higher education contexts, shaping the physiological, psychological, and psychosomatic experiences of students and employees. Chapter 1 introduces the theoretical underpinnings by integrating critical race theory with three-level frameworks from C. P. Jones and Griffith et al., as well as S. P. Harrell’s multidimensional approach. It reviews empirical studies demonstrating how structural racism’s impact is felt through stress, anxiety, other cognitive, and physiological strains that hinder performance and wellbeing. The literature review is organized to examine (a) structural and extra-organizational factors, (b) institutional and intraorganizational influences, (c) interpersonal dynamics, and (d) internalized and individual-level processes. Chapter 2 details the research design, steps employed to investigate these issues within a Northwest U.S. healthcare system. It clarifies the study’s setting, articulates the researcher’s positionality measures for trustworthiness, and lays out the rationale for focusing on healthcare employees and students. This chapter explains how data were collected—through anonymous online surveys—and organized for thematic analysis. Chapter 3 presents the core qualitative findings from 33 participants who self-identified as members of racialized or minoritized groups. Through a systematic thematic analysis, participants reported individual-level psychosomatic and physio-psychological reactions—anxiety, depression, isolation, and sleeplessness—in response to structural and intraorganizational racism. These experiences were linked to decreased focus and productivity, underscoring how institutional stressors influence complex cycles of distress in both academic and professional settings. Chapter 4 introduces an instrument development process grounded in an exploratory sequential mixed-methods research design. Building on the qualitative findings, the chapter illustrates how data-driven insights can inform the creation of measures to assess the influences of structural racism on physio-psychological reactivity. Grounded in established mixed-methods principles, this chapter provides researchers and administrators with a systematic approach to developing tools that can capture the interrelated effects of racism within organizations. This dissertation contributes to understanding how racism-related stress affects institutional structures and individual experiences by combining frameworks, empirical findings, and instrument design to provide a foundation for research and interventions addressing impacts of structural racism
EVALUATING THE QUALITY OF MATERNAL AND NEWBORN CARE IN SUB-SAHARAN AFRICA’S URBAN INFORMAL SETTLEMENTS: A MULTI-COUNTRY MIXED METHODS STUDY
No full textBackground: Sub-Saharan Africa has experienced rapid urbanization, and evidence suggests urban poor women experience worse health outcomes compared to their wealthier counterparts. This multi-country mixed methods dissertation study evaluates the quality of maternal and newborn care in urban informal settlements in Nairobi, Lusaka and Ouagadougou.
Methods: We collected data on experience of care, provision of care and facility readiness across cities. We administered client exit surveys to 1,249 women discharged from childbirth care in facilities serving urban informal settlements and estimated levels and determinants of person-centered maternity care (PCMC) as well as explored courtesy bias in women’s PCMC reporting in facility-based surveys. We conducted 64 semi-structured in-depth interviews with health providers of maternity care in these facilities and studied provision and attributes of PCMC. We conducted health facility assessments and used existing data to estimate facility readiness for labor and delivery care, and small and/or sick newborn care in 194 facilities, comparing those serving and not serving informal settlements. We also tested associations between facility readiness levels and women’s self-reported PCMC scores.
Results: The mean PCMC score ranged from 57.1% in Lusaka to 73.8% in Ouagadougou, with key gaps reported in communication and autonomy. Health systems factors (facility type, managing authority, and receipt of postnatal care before discharge) were statistically significantly associated with PCMC. There was no evidence of courtesy bias in women’s self-reported PCMC in facility-based surveys. Several factors influenced provision of PCMC in informal settlement facilities, including staff shortages, inadequate equipment, excessive workloads, and high levels of provider stress and burnout. Facilities serving urban informal settlements had statistically significantly lower facility readiness for maternal and newborn health services; we did not detect significant associations between facility readiness and PCMC.
Conclusion: This study offers critical evidence on the quality of care in urban informal settlements: urban poor women are disadvantaged as the facilities serving them have low readiness for critical health services. Health providers in these facilities face key structural, systemic and behavioral constraints adversely affecting their provision of PCMC. Multisectoral approaches are needed to address poor quality maternal and newborn care among the urban poorest