Background and Purpose: Colorectal cancer (CRC) is the third most common cancer, the second leading cause of cancer death, with lower survival rates at later stages. Adherence to CRC screening can prevent the development of cancerous polyps and reduce incidence. Area-level characteristics, such as access to healthcare and deprivation, can create barriers to timely screening, increasing the risk of developing CRC. The degree to which area-level characteristics versus individual-level characteristics are responsible for CRC outcomes, including incidence and stage at diagnosis, are not well-understood. Specifically, deficits in the use of spatial statistical techniques has led to a lack of clarity in the current literature. This study aimed to overcome these deficiencies by identifying and utilizing the optimal measurement for area-level access to healthcare and deprivation, employing robust spatiotemporal and multilevel analytic methods to assess their effects on CRC incidence and late-stage diagnosis in Pennsylvania (PA) at the block group-level from 2008 to 2017. The results of this research will more accurately map areas of high predicted CRC relative risk for targeted public health interventions to reduce the burden of CRC over time. The following three study aims were used to address the research problem: Aim 1: Identify the best predictive measure of access to healthcare for estimating CRC incidence risk at the block group-level in PA from 2008 to 2017. Q1: What is the best measure of access to care for estimating risk of CRC incidence? H1.1: The most comprehensive measurement, Multi-Modal 2SFCA, is optimal for predicting CRC incidence compared to unidimensional distance, availability, and other 2SFCA measures. H1.2: Weighting access to healthcare measures for individual insurance coverage improves predictive performance of CRC incidence. Aim 2: Ascertain the relative risk from area-level deprivation on CRC incidence at the block group-level in PA from 2008 to 2017.Q2: How does area-level deprivation affect CRC incidence? H2.1: WQS will demonstrate the relative importance of an extensive array of SES variables for CRC incidence. H2.2: Higher deprivation will be positively associated with risk of CRC incidence. Aim 3: Determine the individual-level likelihood of being diagnosed with late-stage CRC based on place of residence across PA from 2008 to 2017.Q3: How does place of residence affect the likelihood of developing late-stage CRC incidence after adjusting for individual-level characteristics and covariates? H3.1: PA residents living in areas of worse deprivation and low access to care have a higher likelihood of being diagnosed with late-stage CRC. H3.2: The likelihood of late-stage CRC varies significantly by individual characteristics. Methods: This research used ecologic and cross-sectional study designs to perform secondary data analysis of the cancer registry and publicly available data. The geographic units were block groups in PA (N = 9,740), accessed from the US Census Bureau. The sample included screening age-eligible PA residents, 45-75 years, diagnosed with a primary incident case of CRC from 2008 to 2017 (N=34,250), identified via the PA Cancer Registry. Out-of-state residents at diagnosis and high-risk individuals were excluded. Nine block groups were uninhabitable with no population and thus excluded. Primary exposure variables (i.e., area-level access to healthcare and deprivation) were calculated using the PA Cancer Registry, a provider database, the US Census Bureau’s polygon and network shapefiles, and American Community Survey. Ecologic covariates (see below) were derived from the American Community Survey, the Behavioral Risk Factor Surveillance System, and the USDA’s Rural-Urban Commuting Areas. The PA Cancer Registry provided individual data for patient demographics, tumor characteristics, and insurance coverage. Exploratory spatial, temporal, and spatiotemporal analyses of the CRC data were performed before Aims 1 to 3. Aim 1: CRC cases were aggregated by block group to represent a count of CRC incidence. Area-level access to healthcare measures was calculated using providers’ addresses, population-weighted block group centroids, and road/rail networks (i.e., driving, walking, and public transit). Measures included great-circle distance, driving distance to the nearest provider by miles/time, physician-to-population ratio, enhanced two-step floating catchment area (2SFCA), variable 2SFCA, and multi-modal 2SFCA. Four 15-minute catchment sizes were tested (range = 15-60-minutes). A weighted version of each 2SFCA measure for insurance coverage was calculated. Predictive performance was assessed with model fit statistics from 29 hierarchical Bayesian spatiotemporal Poisson regression models. All models included CRC screening adherence, rurality, age, race, education level, unemployment, and poverty level. Aim 2: CRC cases were aggregated by block group to represent a count of CRC incidence. Area-level deprivation indicators (n=39) were calculated from the American Community Survey’s five-year pooled estimates for demographic, social, economic, and housing characteristics and represented at the census tract or block group-level. Weighted Quantile Sum regression generated an area-level deprivation index, weighting each indicator by its relative relationship with CRC incidence. A hierarchical Bayesian spatiotemporal Poisson regression with conditional autoregressive priors and a first-order autoregressive time series process was used to estimate the relative risk of CRC. The ecologic covariates included in the model were area-level access to healthcare from Aim 1, CRC screening adherence, rurality, age, and sex. Aim 3: Three binary outcome variables represented localized vs. regional, distant, and regional and distant CRC at diagnosis. Aim 1 and 2’s area-level access to healthcare and deprivation measurements were used for this study’s primary exposure variables. The data was split into three time periods (2008-2009, 2010-2013, and 2014-2017) to analyze CRCS coverage mandates from the Affordable Care Act for private insurers in 2010 and Medicare in 2014. Using binomial distributed outcomes, three two-level generalized linear mixed models using hierarchical Bayesian methods with conditional autoregressive priors were run for each time period. Results: There were 34,250 eligible incident cases with 0-6 cases per block group (N=9,731) each year and an average of 3.5 cases per block group for the pooled study period. From 2008 to 2017, the pooled CRC incidence rate was 7.45 cases per 1,000 for 45 to 75 year olds in PA. Scan statistics found the highest CRC burden was in Philadelphia (northeast, west, and south), Pittsburgh, and rural areas in southwest PA (e.g., Westmoreland County and Fayette County) and northcentral PA (e.g., Lycoming County, Clinton County, and Centre County). In PA, yearly crude CRC rates decreased slightly over the ten years (0.80 to 0.72, Δ =-.08), though not empirically tested. Aim 1: The best fitting model used the Multi-Modal 2SFCA, which included aggregated physician-to-population ratios within 45-minutes from the provider facility for population-weighted block group centroids via driving, walking, and public transit of the same distance. Access was generally worst in rural areas and best in urban/suburban areas. Block groups with access one standard deviation above the state median had 27% decreased CRC risk. Weighting for insurance coverage improved a measure’s predictive ability for shorter travel times (i.e., 15-minutes and 30-minutes). Aim 2: Of a 39 indicator deprivation index, nine were statistically significant and three were related to SES (i.e., median household income, the percent of the block group without a high school degree, or living in a house without heating). However, the most important significant indicators belonged to geography and income domains, collectively representing 71% of the relative influence of the index. The area-level deprivation index was significant and positively associated with CRC incidence at the block group-level in PA from 2008 to 2017 (RR: 1.33, 95% CI: 1.32–1.34). Aim 3: After accounting for individual age, race, and insurance coverage, the relationship between area-level access to healthcare and deprivation and late-stage CRC became non-significant. While no area-level effects were significant, several individual-level features had consistent significant findings across outcomes and time periods. At the individual-level, having government insurance and being uninsured had significant positive relationships for all outomes and time periods. Age, and race had significant inverse relationships with late-stage CRC diagnosis. Conclusions: In summary, this study addressed the limitations of previous research by employing innovative measurement techniques, such as the Multi-Modal 2SFCA and Weighted Quantile Sum regression, and rigorous spatiotemporal methods to assess the impact of area-level access to healthcare and deprivation on CRC incidence and late-stage diagnosis. The findings highlight the importance of considering walking and public transit access to healthcare in relation to CRC incidence. Additionally, the study demonstrated the effectiveness of the WQS method in calculating an accurate area-level deprivation index, which enhanced the prediction of CRC incidence and identified high-risk areas for targeted interventions. However, individual-level characteristics, particularly insurance coverage, were found to be more influential in predicting the stage at which CRC was diagnosed than area-level effects. Regardless, using inferences and similar methods from this dissertation improves disease mapping and resource allocation for CRCS outreach, supports evidence for policy, and helps guide the development of tailored public health interventions to ultimately reduce the burden of CRC. / Epidemiology

Harbour porpoises (Phocoena phocoena) are the only cetacean residents found year-round in Swedish waters. Since in situ monitoring of cetaceans can be difficult, invasive and often costly, strandings can be used as a cost-effective alternative to continuously collect data on these elusive animals. In this study, spatiotemporal patterns, and their possible underlying causes, of harbour porpoise stranding reports in Swedish waters are investigated over the ten-year period of 2014–2023. When making spatial comparisons, for management purposes, the ten-year period is divided into two, 2014–2018 and 2019–2023. Data on 854 stranded harbour porpoises were analysed from the coasts of the Skagerrak, Kattegat, and Baltic Seas. Both significant spatial and temporal patterns could be identified, with strandings peaking in July to September and with hotspots occurring along most of the Swedish west coast, with the most frequent hotspots located around Öresund and especially the area around Kullen peninsula. The spatial patterns of strandings found in this study reflect data on porpoise abundance, prey abundance, and gillnet fisheries effort. The latter is known to be one of the primary causes of porpoise mortality. Furthermore, the coverage of the Swedish stranding network is analysed. While coverage of the stranding network overall has increased over the period, some areas still lack sufficient coverage, including the coast of Falkenberg, southern Gothenburg, northern Halmstad and certain areas around Lommabukten, north of Malmö. With this, we conclude that harbour porpoise strandings in Sweden show distinct spatiotemporal patterns that can be used as baselines for management and monitoring of these small cetaceans.

Epilepsy is a disease affecting millions of people worldwide. Despite over 50 years of research, the mechanisms that generate and sustain ictal discharges, a key neural hallmark of seizures, remain unknown. While once thought to be caused by hypersynchronous neuronal firing, we now recognize that the activity underlying ictal discharges is much more complex. With the development of microelectrode arrays (MEAs) suitable for use in humans, it is possible to observe neural activity at fine spatiotemporal scales in human patients with epilepsy. However, the diversity of seizure characteristics and limited patient population has led to a number of conflicting observations and theories. The purpose of this work is to elucidate mechanisms underlying ictal discharges in humans by applying statistical analyses and computational modeling to MEA recordings from human patients with epilepsy. We approach this aim in two projects. In the first project, we unify two seemingly conflicting theories surrounding cortical sources of ictal discharges. According to the ictal wavefront theory, ictal discharges are seeded at an expanding narrow front of high neuronal firing that delineates the boundary between regions of cortex with compromised functionality, and surrounding territory where the seizure is observable in electrical recordings, but cortical function remains intact. A second theory posits that discharges are predominantly seeded from a stationary localized cortical source. The two theories are based on observations from MEA recordings of seizures in two different small cohorts of patients. In this project, we analyze and model the discharge propagation patterns in a combined dataset from both cohorts. We show that discharges are seeded at the ictal wavefront in addition to other–possibly stationary–locations. In the second project, we characterize spatiotemporal patterns in the secondary transients of complex ictal discharges. Electrographic recordings of ictal discharges often have complex waveforms. Existing analyses focus on the spatiotemporal dynamics of the first, high-amplitude transient. In this project, we establish that ictal discharges often comprise multiple transients separated by ≈60 ms. Surprisingly, and contrary to our initial hypothesis, we find that individual transients within a complex discharge may propagate with different speeds, suggesting that different mechanisms are involved in the propagation of different transients.

Modelling hydrologic processes is important for understanding how the water cycle works in different environments. Cities which undergo constant changes are subject to flood hazards resulting from severe rainfall. This paper aims to simulate severe rainfall, visualize the results, incorporating both spatial and temporal dimensions, and to make future recommendations for further studies on flood visualization. Visualizing the results from a rainfall simulation using GIS provides urban planners and others the means to view the dynamics of the surface runoff. At the same time, it makes accessible advanced querying and analytical tools. A hydrological model for the study area in Gävle, Sweden was used to simulate a 100-year rainfall. Through FME, the data was reduced, time-stamped and combined to a shapefile. Both 2D software, ArcGIS, and 3D software, ArcScene, were used for creating an animated flood visualization. This study shows that although 2D tested better by a group of planners and water professionals, the 3D was still considered more intuitive. The heightened sense of realism from 3D outweighs its drawbacks, and further studies are required to test different methods of 3D visualization.

No description available.

京都大学 / 新制・課程博士 / 博士(工学) / 甲第24614号 / 工博第5120号 / 新制||工||1979(附属図書館) / 京都大学大学院工学研究科材料工学専攻 / (主査)教授 邑瀬 邦明, 教授 宇田 哲也, 教授 作花 哲夫 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DGAM

Multivariate data is prevalent in modern applications, yet it often presents significant analytical challenges. Factor models can offer an effective tool to address issues associated with large-scale datasets. In this dissertation, we propose two novel Bayesian factors models. These models are designed to effectively reduce the dimensionality of the data, as the number of latent factors is typically much smaller than that of the observation vectors. Therefore, our proposed models can achieve substantial dimension reduction. Our first model is for spatiotemporal areal data. In this case, the region of interest is divided into subregions, and at each time point, there is one univariate observation per subregion. Our model writes the vector of observations at each time point in a factor model form as the product of a vector of factor loadings and a vector of common factors plus a vector of error. Our model assumes that the common factor evolves through time according to a dynamic linear model. To represent the spatial relationships among subregions, each column of the factor loadings matrix is assigned intrinsic conditional autoregressive (ICAR) priors. Therefore, we call our approach the Dynamic ICAR Spatiotemporal Factor Models (DIFM). Our second model, Bayesian Clustering Factor Model (BCFM) assumes latent factors and clusters are present in the data. We apply Gaussian mixture models on common factors to discover clusters. For both models, we develop MCMC to explore the posterior distribution of the parameters. To select the number of factors and, in the case of clustering methods, the number of clusters, we develop model selection criteria that utilize the Laplace-Metropolis estimator of the predictive density and BIC with integrated likelihood. / Doctor of Philosophy / Understanding large-scale datasets has emerged as one of the most significant challenges for researchers recently. This is particularly true for datasets that are inherently complex and nontrivial to analyze. In this dissertation, we present two novel classes of Bayesian factor models for two classes of complex datasets. Frequently, the number of factors is much smaller than the number of variables, and therefore factor models can be an effective approach to handle multivariate datasets. First, we develop Dynamic ICAR Spatiotemporal Factor Model (DIFM) for datasets collected on a partition of a spatial domain of interest over time. The DIFM accounts for the spatiotemporal correlation and provides predictions of future trends. Second, we develop Bayesian Clustering Factor Model (BCFM) for multivariate data that cluster in a space of dimension lower than the dimension of the vector of observations. BCFM enables researchers to identify different characteristics of the subgroups, offering valuable insights into their underlying structure.

The small intestine and the liver regulate several physiological functions together including the absorption and bioavailability of drugs and bile and nitrogen homeostasis. It is important to study these two organs together to gain a holistic understanding of their communication with each other. However, there is a lack of culture models that investigate the use of primary cells/tissues from the liver and the intestine to study their interaction and importance in manifestation of drug toxicity. The studies described in this dissertation were conducted using inverted rat intestinal explants obtained from three regions of the jejunum, named as the proximal, medial and distal jejunum. Markers of enterocyte, goblet cell and Paneth cell function in the jejunum followed in vivo – like spatial trends reported for the entire small intestine. Jejunum explants were integrated with hepatocytes to model the intestine-liver axis. Integration of jejunum explants from the proximal region with hepatocytes had a beneficial effect on both hepatocyte urea secretion and jejunum mucin secretion, hinting at communication between these organs in culture. Integrated cultures of the rat jejunum and hepatocytes were used to investigate ethanol toxicity in vitro. Trends in activities of enzymes involved in ethanol metabolism and mucus secretion in integrated cultures with proximal jejunum explants corroborated with in vivo reports on ethanol toxicity. Various metabolites secreted and metabolized in vitro were also identified using mass spectrometry. Spatial trends in concentrations of several lipids including bile acids, lysophosphatidylcholines and fatty acids corroborated with in vivo reports of lipid metabolism. The integrated intestine-liver cultures can be used as a platform for future investigations of drug toxicity, lipid metabolism and inter-organ communication. / Doctor of Philosophy / The small intestine and the liver perform several functions together. The small intestine is responsible for the digestion of food, absorption of nutrients and metabolism of oral drugs. The liver is involved in the metabolism of glucose, protein, lipids and drugs. It is important to study these two organs together to gain a holistic understanding of their communication with each other. However, there is a lack of culture models that investigate the use of cells/tissues directly obtained from animal liver and intestine to study their interaction and importance in manifestation of drug toxicity. The studies described in this dissertation were conducted using tissues obtained from three regions of the jejunum segment of the rat small intestine. Functional markers of various cell types in the jejunum followed in vivo – like spatial trends reported for the entire small intestine. Jejunum tissues were integrated with liver cells to model the intestine-liver axis. Integration of jejunum tissues from the proximal region with liver cells had a beneficial effect on both liver and intestinal markers, hinting at communication between these organs in culture. Integrated cultures of the rat jejunum and liver cells were used to investigate alcohol toxicity in vitro. Trends in activities of enzymes involved in alcohol metabolism and mucus secretion in integrated cultures with jejunum tissues corroborated with in vivo reports on alcohol toxicity. Various metabolites secreted and metabolized in vitro were also identified using mass spectrometry. Spatial trends in concentrations of lipids including bile acids, lysophosphatidylcholines and fatty acids within the jejunum corroborated with in vivo reports of lipid metabolism. The integrated intestine-liver cultures can be used as a platform for future investigations of drug toxicity, lipid metabolism and inter-organ communication.

An important phenomenon of systems driven far-from-equilibrium is spatiotemporal chaos where the dynamics are aperiodic in both time and space. We explored this numerically for three systems: the Lorenz-96 model, the Swift-Hohenberg equation, and Rayleigh-Bénard convection. The Lorenz-96 model is a continuous in time and discrete in space phenomenological model that captures important features of atmosphere dynamics. We computed the fractal dimension as a function of system size and external forcing to estimate characteristic length and time scales describing the chaotic dynamics. We found extensive chaos with significant deviations from extensivity for small changes in system size and also the power-law growth of the dimension with increasing forcing. The Swift-Hohenberg equation is a partial differential equation for a scalar field, which has been widely used as a model for the study of pattern formation. We found that the magnitude of the mean flow in this model must be sufficiently large for spiral defect chaos to occur. We also explored the spatiotemporal chaos in experimentally accessible Rayleigh-Bénard convection using large-scale numerical simulations of the Boussinesq equations and the corresponding tangent space equations. We performed a careful study analyzing the impact of variations in the domain size, Rayleigh number, and Prandtl number on the system dynamics and fractal dimension. In addition, we quantified the dynamics of the spectrum of Lyapunov exponents and the leading order Lyapunov vector in an effort to connect directly with the dynamics of the flow field patterns. Further, we numerically studied the synchronization of chaos in convective flows by imposing time-dependent boundary conditions from a principal domain onto an initially quiescent target domain. We identified a synchronization length scale to quantify the size of a chaotic element using only information from the pattern dynamics. We also explored the relationship of this length scale with the pattern wavelength. Finally, we analyzed bioconvection which occurs as the result of the collective behavior of a suspension of swimming microorganisms. We developed a series of simulations to capture the gyrotactic pattern formation of the swimming algae. The results can be compared with the corresponding trend of pattern instabilities observed in the experimental studies. / Ph. D.

Data sets of increasing size and complexity require new approaches for prediction as the sheer volume of data from disparate sources inhibits joint processing and modeling. Rather modular segmentation is required, in which a set of models process (potentially overlapping) partitions of the data to independently construct predictions. This framework enables individuals models to be tailored for specific selective superiorities without concern for existing models, which provides utility in cases of segmented expertise. However, a method for fusing predictions from the collection of models is required as models may be correlated. This work details optimal principles for fusing binary predictions from a collection of models to issue a joint prediction. An efficient algorithm is introduced and compared with off the shelf methods for binary prediction. This framework is then implemented in an applied setting to predict instances of civil unrest in Central and South America. Finally, model fusion principles of a spatiotemporal nature are developed to predict civil unrest. A novel multiscale modeling is used for efficient, scalable computation for combining a set of spatiotemporal predictions. / Ph. D.