71446 research outputs found
Sort by
Vertical Contracting and Integration in the Pharmaceutical Market
No full textThis dissertation consists of three chapters studying vertical contracting and vertical integration involving insurers, pharmacy benefit managers (PBMs), and pharmacies in the context of Medicare Part D. The first chapter examines two recent vertical mergers in the pharmaceutical market and how they affected premium competition. I find evidence consistent with incentives of a vertically integrated firm for raising rivals’ costs in one merger, and evidence consistent with cost savings and efficiency gains from vertical integration in the other. The second chapter investigates the impacts of pharmacy network changes on medication adherence and mortality. We find that losing in-network pharmacy access leads to an immediate drop in adherence to statins of up to 4.1 percentage points during the first two quarters, and long-term decreases in adherence rates of over 2.5 percentage points. Additionally, all-cause cumulative mortality rates rise by 0.9 percentage point during the first year after pharmacy access disruptions, followed by an additional increase of 1.5 percentage points by the end of the second year. The mortality impacts are particularly pronounced for female, low-income, and sicker patients. The third chapter examines offsetting effects of stable pharmacy access on medical spending. I find limited to no effects of pharmacy network changes on inpatient visits and stays at skilled nursing facilities and moderate increases of 1.9-3.5 percent in spending on outpatient service categories including hospital outpatient services, durable medical equipment, and outpatient drugs. Pharmacy network disruptions also result in increases in utilization of hospice care.Economic
Trustworthy Machine Learning for Medicine
No full textMachine learning is achieving significant breakthroughs in various applications, including medical research, from analyzing genomic data to accelerating drug discovery and designing personalized treatment plans for patients. However, as machine learning is applied to such high-stakes domains where model errors and biases can adversely impact human lives, there is a growing focus on build- ing models that not only have high accuracy but that are also trustworthy. This dissertation studies three areas of trustworthy machine learning – interpretability, robustness, and safety alignment – and addresses key challenges in each area. In the area of interpretability, we develop a theoretical framework to understand the mathematical properties of explanation methods, elucidating their commonalities and differences, explaining why different methods can generate disagreeing explana- tions, and providing a principled approach to select among methods. In the area of robustness, we develop algorithms to efficiently estimate a model’s average-case robustness, enabling an accurate and efficient characterization of real-world model behavior for large-scale applications. Lastly, in the area of safety alignment, we develop a novel benchmark dataset and evaluate and improve the medical safety of large language models, finding that publicly-available medical large language models do not meet medical safety standards and that fine-tuning them on safety demonstrations can improve their safety while preserving their medical knowledge. Altogether, this research advances the conceptual understanding and practical application of trustworthy machine learning, especially in the medical domain, and paves the way for future research.Biomedical Informatic
Objective, Your Honor: Exploring the Complex Interplay of Risk Assessment Instruments, Judicial Decisions, and Prediction
No full textA wide array of algorithmic risk assessment tools are used throughout the United States for the purposes of pretrial reform, promising a more objective, less disparate, and comprehensive approach to pretrial decision making. Using statistical methods on historical data, these tools aim to predict pretrial outcomes of interest such as failure to appear in court and new criminal arrest in order to determine whether to release a defendant into the community before their trial. Yet, using historical data to influence future predictions introduces its own set of biases, raising questions about the potentially disparate outcomes of these tools. This thesis looks closely at the Public Safety Assessment (PSA), one of the most commonly used pretrial risk assessment tools, as a case study to explore the validity, accuracy, and fairness of actuarial risk assessment instruments (RAIs). Previous validation studies have focused primarily on the predictive accuracy and fairness of the PSA.
This research goes beyond validation to explore how the presence of the PSA recommendation influences judicial decision-making by using data from a randomized control trial (RCT) conducted in Polk County, Iowa in 2018. Using a variety of statistical analyses including Chi-Squared tests, logistic regression, ROC curves, and fairness evaluations, we find that little to no significant correlations exist between the features used to compute the recommendation scores and the outcomes observed. We also determine that the recommendations generated by this tool have little correlation with the judges’ decisions and that the tool is significant and correlated with outcomes for majority demographics groups, but less so for others. The results of this study highlight the consequences of using historical data to inform prediction, and how even demographic-blind evaluations can lead to disparate outcomes. These results raise questions about the usage of data-driven decision making tools in the criminal justice space, and highlight the limited benefits of supposed criminal justice reform measures, providing a critical analysis of RAIs in the pretrial reform space.Computer Scienc
Intertext.AI: Augmented Close Reading for Classical Latin using AI for Intertextual Exploration
No full textTechnological tools for computational language analysis, including applications for ancient languages like classical Latin, continue to improve as artificial intelligence (AI) models become more advanced. However, the subjective process of literary analysis is still primarily manual, relying on commentaries and secondary scholarship. Identifying and interpreting meaningful textual connections, or intertextuality, is labor-intensive and can often present a difficult learning curve for less experienced classicists. Although existing digital platforms offer complex searches by various linguistic qualities such as syntax and phrase similarity, they often do not enable comparisons of the results in their broader contexts, which can provide deeper insights that short excerpts do not reveal. Thus, this thesis proposes the use of AI and visualizations that prioritize contextualization to facilitate the discovery of intertextual correspondences using complex quantitative representations of texts. We introduce a novel web interface, Intertext.AI, that integrates Latin BERT (Bamman and Burns 2020), a machine learning language model trained on classical Latin texts, into contextually rich text visualizations to assist classicists in searching for potential intertextual connections. To evaluate the system, we tested its ability to find allusions attested in classical scholarship, investigated a case study on how a reader may use the interface to enhance their close reading, and conducted a user study with 19 participants who explored potential correspondence between two pairs of Latin poems. Intertext.AI identified over 80% of attested connections from excerpts of Lucan’s Pharsalia, demonstrating the system’s technical efficacy at detecting allusions. Further, while participants did not identify significantly different types or quantities of connections when using Intertext.AI or other tools, they overall found it easier to find and justify potential intertextuality with Intertext.AI, reported higher confidence in their observations identified through the interface, and preferred having access to it during the search process. These findings thus suggest that Intertext.AI facilitates meaningful intertextual discovery and interpretation by
fostering literary comparison.Computer Scienc
Measuring Apoptotic Sensitivity Through Oligonucleotide Mediated Peptide Delivery
No full textApoptosis is a process of cell death triggered by cellular stress or extrinsic signals. In apoptosis, pro-death proteins like Bcl-2 interacting mediator of cell death (BIM) activate pore-forming proteins BAX and BAK at the mitochondria, leading to mitochondrial outer membrane permeabilization (MOMP) and downstream activation of apoptotic caspases. BH3 profiling is a method used to measure the signals required to initiate this cell death pathway. Cells are treated with peptides mimicking the Bcl-2 homology 3 (BH3) domain of pro-apoptotic proteins. By measuring responses to these peptides, BH3 profiling can compare peptide sensitivity, or “apoptotic priming,” across cell types, genotypes, or treatments. However, the requirement for membrane permeabilization limits physiological relevance and prevents longitudinal analysis.
My thesis develops an alternative method for delivering BH3 peptides to living cells without membrane permeabilization. We used a trimethoprim (TMP)-inducible degron system based on a destabilized dihydrofolate reductase (DHFR) tag that enables rapid, post-translational control of protein stability. For system validation, Citrine-DHFR fusion protein was utilized, and TMP treatment induced Citrine expression in over 90% of cells at concentrations as low as 625 nM. In parallel, we optimized transfection conditions across HeLa lines with varying apoptotic sensitivity. Through in silico design, site-directed mutagenesis and Golden Gate cloning were used to replace Citrine with BIM and integrate the sequence into a new plasmid construct to allow future experimental validation. This system supports new extensions of BH3 profiling, including rapid nucleotide changes to expressed peptides and downstream applications such as single-cell RNA sequencing in unpermeabilized cells.Biomedical Engineering A
Essays on Innovation, Strategic Alliances, and Disruption in the Global Container Shipping Industry
No full textThis dissertation explores the dynamics of innovation, strategic alliances, and disruption in the global container shipping industry through three essays. The first chapter examines how innovation in vessel size affect market structure and welfare, highlighting a trade-off between cost efficiency and concentration. The second investigates the formation and stability of global shipping alliances, quantifying the gains from global fleet reallocation and the risks of enhanced market power. The third estimates importer demand for speed and reliability in freight transport during supply chain disruptions, revealing a significant willingness to pay for transit speed and predictability. Together, these essays provide new insights into the evolving dynamics of container shipping and offer policy guidance on innovation, competition, and resilience.Economic
Land, Landlords, and Local Capital in Housing Markets
No full textThis thesis consists of three essays on the housing market, with a focus on how spatially induced constraints affect prices, investment, and supply. The first two chapters study household investors, who collectively own about half of all US residential rental units: their characteristics, impact on local markets, and the spatial dispersion in the returns they earn. Taken together, my results suggest an underappreciated role for landlords' investment decisions in shaping local prices, and strong barriers to arbitrage across space in this sector. In the first chapter, I show that higher wealth or propensity to invest among local household investors increases local housing prices and lowers returns. I document several new facts about these investors in the US, including that they are highly home-biased. I develop a novel model that fits these patterns and in which risk-averse investors price local rental properties in implicit local capital markets and under spatial equilibrium. A series of reduced-form tests confirm the model's comparative statics, using IV and shift-share designs. The second chapter complements the first by showing the extent to which rental yields diverge across space—both across and within cities—and that this dispersion primarily reflects dispersion in returns, not rent growth. The third chapter examines the effect of highway construction on the housing market, including housing supply and land prices, and the extent to which land availability mediates these effects, focusing on the Interstate Highway System (IHS). I find that highways drive population growth where land is abundant and lower prices where it is scarce. I then quantify the welfare and population effects of the IHS's housing market channel, finding large, though heterogeneous, gains. These findings highlight the potential housing market benefits of infrastructure projects when subsequent development is possible.Economic
Tracking Joint Kinematics and Muscle Kinetics with Multimodal Wearable Sensors
No full textSmart wearables are becoming increasingly popular, taking forms such as watches, glasses, and even rings. Health tracking has emerged as one of their key functions, offering unobtrusive and continuous monitoring beyond the capabilities of traditional clinical tools. Current smart wearables primarily focus on general health metrics, such as heart rate, blood oxygen level, and calories burned. However, injuries and conditions often arise at the tissue level, specifically within muscles, bones, or joints. Tissue-specific tracking can thus provide more detailed and actionable health insights, making it a transformative feature of next-generation smart wearables. This dissertation aims to bridge the gap between wearable technologies and tissue-specific measurements by developing and evaluating a range of wearable sensing systems to track joint movements and muscle force outputs, paving the way for more precise and personalized health monitoring.
This thesis begins by developing wearable joint kinematics estimation strategies. First, we present a soft sensing shirt designed to track 3D shoulder kinematics during both cyclic and random arm movements. The shirt incorporates eight textile-based capacitive strain sensors with high linearity, low hysteresis, and effective shielding against human parasitic capacitance, making it ideal for wearable applications. In a single-participant study, the sensing shirt achieved accurate tracking with an ensemble-based machine learning algorithm, yielding root mean square errors (RMSEs) below 4.5° for joint angle estimation and normalized root mean square errors (NRMSEs) below 4% for joint velocity estimation. Next, we propose a frame alignment method for inertial measurement units (IMUs), which are commonly validated against ground truth optical motion capture systems (OMC). Fair comparisons between IMU and OMC measurements require accurate frame alignment between the two systems. Unlike existing methods that address the local and global frame misalignments as separate issues, we present an assumption-free, data-based approach that simultaneously aligns both local and global frames via quaternion-based least squares optimization. Tested with data from six participants, this method achieved alignment errors under 1.5° and effectively isolated IMU drift during long-duration dynamic movements. Lastly, we develop a wearable system for long-duration 3D shoulder kinematics tracking by fusing soft sensor and IMU signals. Unlike existing wearable systems, our approach requires only a short calibration period and no external lab-based equipment. In tests with six participants, the system uses just 2.5 minutes of random arm movement calibration to achieve RMSEs of 4.5° across all degrees of freedom for over an hour of continuous functional activities, including desk work, dancing, walking, activities of daily living, and strength training. Furthermore, we demonstrate the system’s potential for real-world applications by showing that a simplified configuration using four soft sensors (instead of eight) and a shorter 90-second calibration could still achieve RMSEs of 5.1°.
In addition to joint kinematics tracking, this thesis contributes to wearable sensing methods for muscle kinetics. First, we present a strategy for estimating corresponding joint torque from muscles with different architectures during various dynamic activities using wearable A-mode ultrasound. By tracking changes in muscle thickness with single-element ultrasonic transducers, we use muscle deformation data to estimate elbow and knee torque, achieving NRMSEs below 7.6% and coefficients of determination (R^2) exceeding 0.92 during controlled isokinetic contractions across 10 participants. We further demonstrate the feasibility of wearable joint torque estimation with 5 participants during dynamic real-world tasks, including weightlifting, cycling, and treadmill and outdoor locomotion. Lastly, we introduce a wearable muscle fatigue tracking strategy that combines A-mode ultrasound with electrical stimulation. This hybrid approach reveals that muscle deformation from electrically induced contractions correlates strongly with muscle fatigue. Using a muscle deformation index derived from A-mode ultrasound, we reliably track muscle fatigue with correlation coefficients (r) of 0.85 during dynamic, volitional fatiguing contractions on 8 participants using an isokinetic dynamometer.
Together, this thesis advances the design and validation of wearable sensing systems for joint kinematics and muscle kinetics estimation, paving the way for practical applications in real-world, unconstrained environments.Engineering and Applied Sciences - Engineering Science
Electrical pattern formation and function in organogenesis and tissue growth
No full textMonoatomic ions are among the most abundant chemical species in the biological environment. Cells maintain complex protein machinery and expend a large fraction of their energy budget to control the relative distribution of different ions between themselves and their surroundings, as well as between cellular compartments in eukaryotes. We have a deep understanding of ion physiology in the electrical action potentials of excitable cells like neurons and myocytes. However, the universality of ion regulation across all cells suggests that much more can be learned about how cells sense their ionic state in diverse contexts and use it to represent important information for other biological decisions.
Here, I present studies that advance our understanding of the organization and function of cellular ion fluxes at three different spatial scales: 1) a contiguous tissue of genetically similar cells; 2) a physiological functional unit consisting of multiple cell types across multiple organs; and 3) individual molecular machines within single cells. Across these works I exploit all-optical approaches combining light-activated, spatiotemporally localized perturbations with sensors of physiological dynamics in living cells. These tools allow us to dissect causal relationships between biological parameters that have historically been understood to be important to physiology but challenging to directly visualize and control in their native context.
First, we study the earliest spiking activity in the zebrafish heart, imaging the transition from silent to electrically excitable to spontaneously spiking during cardiac development and developing mathematical descriptions for the timing and spatial organization of the earliest heartbeats. In the adult heart, rhythmic activity is generated by specialized pacemaker cells, whereas most other cardiomyocytes have much less spontaneous electrical activity\cite{bartos_ion_2015,liu_electrophysiological_2016}. In general, immature cardiomyocytes have more spontaneous activity and share more molecular similarities with pacemaker cells\cite{liu_electrophysiological_2016}. We observe that the initiation of periodic spiking during embryonic heart development is not a direct consequence of pacemaker cell differentiation and show that it does not involve deterministic encoding of periodicity.
Second, we consider ion physiology in the body vasculature. The vertebrate circulatory system selectively and transiently adjusts blood flow to meet the local demand for specific tissues, for example, in neurovascular coupling and exercise. Several of the mechanisms that enable these responses involve ion flows, including calcium entry and membrane potential hyperpolarization downstream of physiological inputs . One such input is change in shear stress exerted by blood flow on vessels, which is assumed to be driven by a change in heart rate . However, changes in heart rate should systemically adjust flow, and additional mechanisms are required to explain local changes in vascular tone. We use the larval zebrafish to observe multiple different processes executed by different cell types, driving the physiological cascade to a vascular response. We show that, rather than sensing changes in cardiac output, the mechanosensitive ion channel Piezo1 triggers intracellular calcium elevation by directly sensing compression on blood vessels. This constitutes a novel mechanism for converting routine organismal behavior into a local cellular response.
Third, we examine how cellular ion physiology might influence proliferation and differentiation. We develop approaches to read out the effects of ionic perturbations on ERK/MAPK signaling, a pathway critical for cell cycle progression and various differentiation processes during development. ERK/MAPK signaling has been reported to be sensitive to calcium elevation , transmembrane potential and externally applied electric fields. We isolate physiological events immediately downstream of these perturbations and show that ionic triggers of ERK signaling act through the structural dynamics of specific potassium channels, rather than by universal sensing of any physiological parameter. This may confer different response behaviors on different cell types and could allow spatio-temporal coupling of divergent behaviors in complex developmental processes.
Together, these studies extend our boundaries of where electrophysiological function can occur, of what the systemic function of such activity can be, and of the nature of the biological information it can carry. These efforts provide a framework for exploring new functions of ion fluxes in contexts beyond classical excitability, and offer a glimpse of future work that could further integrate ion physiology into our global view of how cells and tissues sense and adapt to their environment and internal state.Systems Biolog
Construction of a Synthetic Viral ORFeome for Identifying Regulators of Host Immune Responses
No full textAdvances in DNA synthesis and genetic screening technologies enable the systematic
exploration of gene function. Previous screening technologies, such as large-scale ORFeome
libraries and CRISPR-Cas9 libraries either relied on collections of ORFs cloned from the cell or
the synthesis of short segments of DNA corresponding to target genes. Smaller scale ORF
libraries from pathogens, such as Kaposi’s sarcoma herpesvirus and herpes simplex virus 1,
were constructed by amplifying target DNA from the viral genome and subcloning it into an
expression vector. To date, DNA synthesis technology has not been used to construct a largescale
collection of ORFs from human or viral sources.
We generated a large-scale collection of viral ORF products, the viral ORFeome, from
over 600 strains of virus that were known to infect humans or closely related to viruses that
infect humans and representing a potential zoonotic threat. The viral ORFeome contains over
10,000 unique viral ORF fragments with up to five unique barcodes each that can be individually
expressed in mammalian cells for genetic perturbation screens, which we demonstrated by
performing genetic screens for viral ORFs that regulate cellular proliferation. We then employed
this viral ORFeome to identify viral ORFs that modulate the host immune response.
First, we performed a viral ORFeome screen for viral ORFs that regulate the expression
of class I MHC on the cell surface. Class I MHC expression is critical for CD8+ T cell mediated
killing of virally infected cells. We identified several novel regulators of class I MHC cell surface
expression, including MC162R. We demonstrated that MC162R engages the host cell ubiquitin
machinery and redundantly recruits several host ubiquitin E3 ligases to facilitate the
ubiquitylation and degradation of class I MHC.
Second, we sought to examine whether the viral ORFeome could be used to examine
viral inhibition of cell signaling pathways, focusing on the IFNb and IFNg signaling pathways.
IFN signaling is critical for generating an antiviral state in response to viral infection and are
frequently impaired by viral proteins. We identified several novel regulators of IFN signaling,
including families of viral proteins that preferentially regulated either IFNb or IFNg signaling. We
characterized one of these families of viral ORFs, the yatapoxvirus 151R ORFs, as selective
inhibitors of IFNb signaling that interact with IRF9.
Finally, we sought to develop a system for screening for regulators of nucleic acid
sensing. We developed a knock-in reporter system for IFNB1 expression and leveraged the
ability of Sendai virus to potently induce RNA sensing to conduct a screen for viral ORFs that
regulate RNA sensing. Collectively, these results demonstrate the utility of a viral ORFeome in
identifying novel regulators of host cell immune function, which can now be expanded to other
signaling pathways.Medical Science