Indiana University-Purdue University Indianapolis (IUPUI) / A popular non-invasive brain activity measurement method is based on the functional magnetic resonance imaging (fMRI). Such data are frequently used to study functional connectivity (FC) defined as statistical association among two or more anatomically distinct fMRI signals (Friston, 1994). FC has emerged in recent years as a valuable tool for providing a deeper understanding of neurodegenerative diseases and neuropsychiatric disorders, such as Alzheimer's disease and autism. Information about complex association structure in high-dimensional fMRI data is often discarded by a calculating an average across complex spatiotemporal processes without providing an uncertainty measure around it. First, we propose a non-parametric approach to estimate the uncertainty of dynamic FC (dFC) estimates. Our method is based on three components: an extension of a boot strapping method for multivariate time series, recently introduced by Jentsch and Politis (2015); sliding window correlation estimation; and kernel smoothing. Second, we propose a two-step approach to analyze and summarize dFC estimates from a task-based fMRI study of social-to-heavy alcohol drinkers during stimulation with avors. In the first step, we apply our method from the first paper to estimate dFC for each region subject combination. In the second step, we use semiparametric additive mixed models to account for complex correlation structure and model dFC on a population level following the study's experimental design. Third, we propose to utilize the estimated dFC to study the system's modularity defined as the mutually exclusive division of brain regions into blocks with intra-connectivity greater than the one obtained by chance. As a result, we obtain brain partition suggesting the existence of common functionally-based brain organization. The main contribution of our work stems from the combination of the methods from the fields of statistics, machine learning and network theory to provide statistical tools for studying brain connectivity from a holistic, multi-disciplinary perspective.

In dieser Dissertation werden neue Regressionsansätze für multivariate longitudinale oder funktionale Daten entwickelt, die eine flexible Modellierung von interessierenden Kovariableneffekten ermöglichen. Die Abhängigkeit innerhalb und zwischen den verschiedenen Zielgrößen wird über latente multivariate Gauss-Prozesse modelliert. Die Regressionsansätze folgen einem zweistufigen Verfahren, in dem in einem vorgelagerten Schritt multivariate funktionale Hauptkomponentenanalysen eingesetzt werden, um sparsame empirische Basen für die Gauss-Prozesse zu konstruieren. Drei verschiedene Regressionsmodelle werden für verschiedene Arten multivariater longitudinaler oder funktionaler Daten entwickelt. Das erste Projekt führt das zweistufige Verfahren für multivariate normalverteilte funktionale Daten ein, die eine gekreutzte oder genestete Datenstruktur aufweisen können. Das Regressionsmodell ist im frequentistischen Rahmenmodell der funktionalen additiven gemischten Modelle eingebettet und wird durch Anwendungen auf Bewegungsdaten und Sprachdaten illustriert. Das zweite Projekt entwickelt ein bayesianisches Regressionsgerüst für multilevel multivariate funktionale Daten, die verschiedenen punktweisen Verteilungen folgen. Das erlaubt es, verschiedene Datentypen, wie etwa binäre, Zähl- oder kontinuierliche funktionale Daten gleichzeitig zu modellieren, was durch eine Anwendung auf Berliner Verkehrsdaten veranschaulicht wird. Das dritte Projekt vereint multivariate longitudinale normalverteilte Daten mit einer Ereigniszeit-Zielgröße in einem gemeinsamen bayesianischen Modellierungsansatz. Solche Modelle werden oft im medizinischen Bereich verwendet, beispielsweise wenn der Fokus der Analyse auf der Schätzung der Assoziation zwischen longitudinalen Messungen von Biomarkern und dem Überleben von Patienten mit chronischen Lebererkrankung liegt. / In this thesis, novel regression approaches for multivariate longitudinal or functional data are developed, which allow to flexibly model the covariate effects of interest. The dependency within and between the different outcomes is modeled using latent multivariate Gaussian processes. The regression approaches adopt a two-step procedure where, in a preliminary step, multivariate functional principal component analyses are employed to generate parsimonious empirical bases for the Gaussian processes. Three different regression models are developed for different types of longitudinal or multivariate functional data. The first project establishes the two-step procedure for multivariate Gaussian functional data which can exhibit a crossed or nested multilevel structure. The regression model is embedded in the frequentist functional additive mixed model framework and is demonstrated by applications in movement data and speech production data. The second project develops a Bayesian regression framework for multilevel multivariate functional data that follow different pointwise distributions. This allows to simultaneously model data of different types such as binary, count, or continuous functional data, which is illustrated by an application to Berlin traffic data. The third project combines multivariate longitudinal Gaussian data with a time-to-event outcome in a Bayesian joint modelling approach. Such models are often used in medical contexts where the main point of interest lies in estimating the association between longitudinal measurements of biomarkers and e.g. the survival of patients as in the presented application to a chronic liver disease. All projects are accompanied by simulation studies to assess the estimation accuracy and the models' limitations.

Made available in DSpace on 2011-05-04T12:42:04Z (GMT). No. of bitstreams: 0 Previous issue date: 2010 / A hipertensão arterial é um dos mais importantes fatores de risco para o desenvolvimento das complicações cardiovasculares, cerebrovasculares e renais. Embora seja facilmente detectável, o controle dos níveis tensionais constitui um enorme desafio da saúde pública. O objetivo desta tese foi analisar dados longitudinais de uma coorte de pacientes hipertensos resistentes. Em estudos longitudinais, o principal foco de interesse é na mudança ocorrida ao longo do tempo; seja ela avaliada como tempo até o evento ou como medidas repetidas tomadas durante o período de acompanhamento. O presente trabalho foi organizado em três artigos onde foram apresentadas essas duas abordagens. No primeiro artigo, foi realizada uma Análise de Sobrevida, tendo como desfecho eventos cardiovasculares fatais e não fatais, em mulheres hipertensas da coorte. Foi verificado que para atingir o objetivo de diminuir a morbidade e a mortalidade cardiovascular nessa população, as decisões deveriam ser baseadas no controle da pressão devigília obtida na Monitorização Ambulatorial da Pressão Arterial (MAPA) e não no controle dapressão de consultório. No segundo artigo, foram usadas as medidas da pressão arterial (PA) obtidas na MAPA em sua forma resumida usual (médias de PA 24h, vigília e noturna). Ospacientes hipertensos pseudorresistentes apresentaram trajetória ascendente, indicando a necessidade de acompanhamento desses pacientes a intervalos inferiores a um ano. Além disso, não foi observada redução dos valores do índice de massa corporal e da circunferência da cintura nesses pacientes. O terceiro artigo abordou a evolução temporal dos valores do descenso noturnopressórico nos pacientes da coorte, além de estimar as probabilidades brutas de transição entre as categorias do descenso noturno, em MAPAs sucessivas. Apesar de não ultrapassar o limite de normalidade de 10 por cento, houve uma queda acentuada nos valores percentuais do descenso noturnodos pacientes dippers ao longo do tempo. A probabilidade estimada de permanência no estado dipper foi de 52 por cento, enquanto que no estado non dipper esse valor foi de 46 por cento. Nesses dois artigos foram usados Modelos Aditivos Generalizados Mistos, que incorporam efeitos aleatórios, umavez que a variação intra-paciente foi expressiva. A incorporação de métodos estatísticos mais sofisticados faz jus à qualidade e custo de coleta das informações longitudinais. Com base nesses três artigos, concluiu-se que o uso da MAPA é primordial no acompanhamento de pacientes hipertensos resistentes, pois permite detectar as variações ao longo do tempo na evolução clínica. / Hypertension is one of the most important risk facotors for cardiovascular, cerebrovascular and renal diseases. While it is easy to detect, blood pressure control is a major public health challenger. The objective of this thesis was to examine longitudinal data fron a cohort of resistant hypetensive patients. In longitudinal studies, the main focus of interest is on changes over time, either evalueated as time-to -event or as repeated measures taken durin the foloow-up. this thesis was organized in tree articles which presented these two approaches. In the fist article, in the suvival approach, we modeled the time free of fatal and nonfatal cardiovasculr event, in hypertensive women of the cohort. It was found that to achieve the goal of decreasing cardiovasculr morbidity and mortality in this population, decisions should be based on the control of daytime Ambulatory Blood Pressure (ABP) and not on the control of office blood pressure. In the second article, it was used blood pressure (BP) measurementes from ABPM in its usual summary form (mean 24h, daytime ande nighttime). Pseudoresistant hypetensive patients showed an upward trend, indicating th need to monitor them more than once a year. Moreover, there was no reduction in body mass index and waist circunference values in these patientes. the third paper dealt with the evolution of nocturnal blood pressure values in these cohort patients, as well as estimated the crude probabilities of trnsitions between the nocturnal dip categories in sucessive ABPM. Although the limit of normality of 10% was not excessd, a sharp drop in nocturnal dip values was observed on dippers patients oves time. the maindtenance probbiliy in dippes status was estimted in 52% whereas in non dipper status figured in 46%. In these two articles generalized additive mixed models that incorporte random effects were used, since the intr-patient vrition was significant. the incorpotation of moro sophisticated statistical methods is justified by the quality and cost of longitudinal informations. Based on these three articles, it was conclued tht the use of ABPM is essentil in monitoring patients with resistant hypertension, since it allows to detect chnges over time in the clinical outcome.