The typical research of Alzheimer's disease includes a series of cognitive states. Multi-state models are often used to describe the history of disease evolvement. Competing risks models are a sub-category of multi-state models with one starting state and several absorbing states. Analyses for competing risks data in medical papers frequently assume independent risks and evaluate covariate effects on these events by modeling distinct proportional hazards regression models for each event. Jeong and Fine (2007) proposed a parametric proportional sub-distribution hazard (SH) model for cumulative incidence functions (CIF) without assumptions about the dependence among the risks. We modified their model to assure that the sum of the underlying CIFs never exceeds one, by assuming a proportional SH model for dementia only in the Nun study. To accommodate left censored data, we computed non-parametric MLE of CIF based on Expectation-Maximization algorithm. Our proposed parametric model was applied to the Nun Study to investigate the effect of genetics and education on the occurrence of dementia. After including left censored dementia subjects, the incidence rate of dementia becomes larger than that of death for age < 90, education becomes significant factor for incidence of dementia and standard errors for estimates are smaller. Multi-state Markov model is often used to analyze the evolution of cognitive states by assuming time independent transition intensities. We consider both constant and duration time dependent transition intensities in BRAiNS data, leading to a mixture of Markov and semi-Markov processes. The joint probability of observing a sequence of same state until transition in a semi-Markov process was expressed as a product of the overall transition probability and survival probability, which were simultaneously modeled. Such modeling leads to different interpretations in BRAiNS study, i.e., family history, APOE4, and sex by head injury interaction are significant factors for transition intensities in traditional Markov model. While in our semi-Markov model, these factors are significant in predicting the overall transition probabilities, but none of these factors are significant for duration time distribution.

In this article, we make inferences for the Weibull parameters and propose two test statistics for the comparison of two Weibull distributions based on interval-censored data. However, the distributions of the two statistics are unknown and not easy to obtain, therefore a simulation study is necessary. An urn model in the simulation of interval-censored data was proposed by Lee (1999) to select random intervals. Then we propose a simulation procedure with urn model to obtain approximately the quantiles of the two statistics. We demonstrate an example in AIDS study to illustrate how the tests can be applied to the infection time distributions of AIDS.

El análisis de supervivencia trata de la evaluación estadística de variables que miden el tiempo transcurrido hasta un evento de interés. Una particularidad que ha de considerar el análisis de supervivencia son datos censurados. Éstos aparecen cuando el tiempo de interés no puede ser observado exactamente y la información al respecto es parcial. Se distinguen diferentes tipos de censura: un tiempo censurado por la derecha está presente si el tiempo de supervivencia es sabido mayor a un tiempo observado; censura por izquierda está dada si la supervivencia es menor que un tiempo observado. En el caso de censura en un intervalo, el tiempo está en un intervalo de tiempo observado, y el caso de doble censura aparece cuando, también, el origen del tiempo de supervivencia está censurado.La primera parte del Capítulo 1 contiene un resumen de la metodología estadística para datos censurados en un intervalo, incluyendo tanto métodos paramétricos como no-paramétricos. En la Sección 1.2 abordamos el tema de censura noinformativa que se supone cumplida para todos los métodos presentados. Dada la importancia de métodos de optimización en los demás capítulos, la Sección 1.3 trata de la teoría de optimización. Esto incluye varios algoritmos de optimización y la presentación de herramientas de optimización. Se ha utilizado el lenguaje de programación matemática AMPL para resolver los problemas de maximización que han surgido. Una de las características más importantes de AMPL es la posibilidad de enviar problemas de optimización al servidor 'NEOS: Server for Optimization' en Internet para que sean solucionados por ese servidor.En el Capítulo 2, se presentan los conjuntos de datos que han sido analizados. El primer estudio es sobre la supervivencia de pacientes de tuberculosis co-infectados por el VIH en Barcelona, mientras el siguiente, también del área de VIH/SIDA, trata de usuarios de drogas intra-venosas de Badalona y alrededores que fueron admitidos a la unidad de desintoxicación del Hospital Trias i Pujol. Un área completamente diferente son los estudios sobre la vida útil de alimentos. Se presenta la aplicación de la metodología para datos censurados en un intervalo en esta área. El Capítulo 3 trata del marco teórico de un modelo de vida acelerada con una covariante censurada en un intervalo. Puntos importantes a tratar son el desarrollo de la función de verosimilitud y el procedimiento de estimación de parámetros con métodos del área de optimización. Su uso puede ser una herramienta importante en la estadística. Estos métodos se aplican también a otros modelos con una covariante censurada en un intervalo como se demuestra en el Capítulo 4.Otros métodos que se podrían aplicar son descritos en el Capítulo 5. Se trata sobre todo de métodos basados en técnicas de imputación para datos censurados en un intervalo. Consisten en dos pasos: primero, se imputa el valor desconocido de la covariante, después, se pueden estimar los parámetros con procedimientos estadísticos estándares disponibles en cualquier paquete de software estadístico.El método de maximización simultánea ha sido implementado por el autor con el código de AMPL y ha sido aplicado al conjunto de datos de Badalona. Presentamos los resultados de diferentes modelos y sus respectivas interpretaciones en el Capítulo 6. Se ha llevado a cabo un estudio de simulación cuyos resultados se dan en el Capítulo 7. Ha sido el objetivo comparar la maximización simultánea con dos procedimientos basados en la imputación para el modelo de vida acelerada. Finalmente, en el último capítulo se resumen los resultados y se abordan diferentes aspectos que aún permanecen sin ser resueltos o podrían ser aproximados de manera diferente. / Survival analysis deals with the evaluation of variables which measure the elapsed time until an event of interest. One particularity survival analysis has to account for are censored data, which arise whenever the time of interest cannot be measured exactly, but partial information is available. Four types of censoring are distinguished: right-censoring occurs when the unobserved survival time is bigger, left-censoring when it is less than an observed time, and in case of interval-censoring, the survival time is observed within a time interval. We speak of doubly-censored data if also the time origin is censored.In Chapter 1 of the thesis, we first give a survey on statistical methods for interval-censored data, including both parametric and nonparametric approaches. In the second part of Chapter 1, we address the important issue of noninformative censoring, which is assumed in all the methods presented. Given the importance of optimization procedures in the further chapters of the thesis, the final section of Chapter 1 is about optimization theory. This includes some optimization algorithms, as well as the presentation of optimization tools, which have played an important role in the elaboration of this work. We have used the mathematical programming language AMPL to solve the maximization problems arisen. One of its main features is that optimization problems written in the AMPL code can be sent to the internet facility 'NEOS: Server for Optimization' and be solved by its available solvers.In Chapter 2, we present the three data sets analyzed for the elaboration of this dissertation. Two correspond to studies on HIV/AIDS: one is on the survival of Tuberculosis patients co-infected with HIV in Barcelona, the other on injecting drug users from Badalona and surroundings, most of whom became infected with HIV as a result of their drug addiction. The complex censoring patterns in the variables of interest of the latter study have motivated the development of estimation procedures for regression models with interval-censored covariates. The third data set comes from a study on the shelf life of yogurt. We present a new approach to estimate the shelf lives of food products taking advantage of the existing methodology for interval-censored data.Chapter 3 deals with the theoretical background of an accelerated failure time model with an interval-censored covariate, putting emphasize on the development of the likelihood functions and the estimation procedure by means of optimization techniques and tools. Their use in statistics can be an attractive alternative to established methods such as the EM algorithm. In Chapter 4 we present further regression models such as linear and logistic regression with the same type of covariate, for the parameter estimation of which the same techniques are applied as in Chapter 3. Other possible estimation procedures are described in Chapter 5. These comprise mainly imputation methods, which consist of two steps: first, the observed intervals of the covariate are replaced by an imputed value, for example, the interval midpoint, then, standard procedures are applied to estimate the parameters.The application of the proposed estimation procedure for the accelerated failure time model with an interval-censored covariate to the data set on injecting drug users is addressed in Chapter 6. Different distributions and covariates are considered and the corresponding results are presented and discussed. To compare the estimation procedure with the imputation based methods of Chapter 5, a simulation study is carried out, whose design and results are the contents of Chapter 7. Finally, in the closing Chapter 8, the main results are summarized and several aspects which remain unsolved or might be approximated in another way are addressed.

Lorsqu'on étudie la démence à partir de données de cohorte, les sujets sont suivis par intermittence ce qui donne lieu à des temps d'apparition de la démence censurés par intervalle et ont un risque important de décès, d'où un nombre non négligeable de sujets qui décèdent sans avoir été diagnostiqués déments. Le modèle adapté à l'étude de la démence dans ce contexte est un modèle illness-death dans lequel les sujets initialement non malades peuvent transiter vers l'état décédé directement ou en passant par l'état malade. La vraisemblance du modèle permet en particulier de tenir compte du fait que les sujets décédés sans diagnostic de démence ont pu passer par deux chemins différents entre leur dernière visite et leur décès. Elle ne se factorise pas comme dans le cas où les différents temps de transition sont connus exactement ; tous les paramètres sont donc estimés conjointement. Or, une pratique courante lorsqu'on s'intéresse aux facteurs de risque de démence consiste à considérer uniquement la transition de l'état non malade à l'état malade. Afin de pouvoir appliquer les techniques d'analyse de survie classiques, les sujets décédés sans diagnostic de démence sont artificiellement censurés à droite à leur dernière visite. La première partie de cette thèse permet de montrer que cette approche, contrairement à l'approche illness-death, peut induire des biais dans l'estimation des effets des facteurs de risque. Le fait de modéliser le décès en plus de la démence permet aussi d'exprimer des quantités directement liées au décès comme des espérances de vie ou le risque absolu de démence au cours de la vie entière. Dans la deuxième partie de cette thèse, nous nous efforçons de dégager toutes les quantités pertinentes d'un point de vue épidémiologique qui peuvent être exprimées dans un contexte illness-death. Elles peuvent être estimées en plus des différentes intensités de transition et des effets des facteurs de risque à l'aide du paquet R SmoothHazard, développé au cours de cette thèse. Enfin, la dernière partie de cette thèse consiste à prendre en compte l'hétérogénéité de nos données. Nous introduisons des effets aléatoires sur les trois transitions du modèle illness-death afin de prendre en compte des facteurs de risque partagés par les sujets appartenant à un même groupe. / In dementia research, difficulties arise when studying cohort data. Time-to-disease onset is interval censored because the diagnosis is made at intermittent follow-up visits. As a result, disease status at death is unknown for subjects who are disease-free at the last visit before death. The illness-death model allows initially disease-free subjects to first become ill and then die, or die directly. Those two possible trajectories of the subjects who died without dementia diagnosis can be taken into account into the likelihood. Unlike the case where transition times are exactly observed, the latter do not factorizes and parameters of the three transitions have to be estimated jointly. However, when studying risk factors of dementia, a common approach consists in artificially ending follow-up of subjects who died without dementia diagnosis by considering them as right censored at the last time they were seen without disease. The first part of the present work shows that this approach (unlike the illness-death modeling approach) can lead to biases when estimating risk factor effects of dementia. Modeling death in addition to disease also allows to consider quantities which are closely related with risk of death, like lifetime risk of disease or life expectancies. In the second part of this work, we detail all the quantities which are of epidemiological interest in an illness-death model. They can be estimated, in addition to the transition intensities and the effects or risk factors, using the R package SmoothHazard which has been implemented during this thesis. Finally, in the last part of this work, we consider shared frailty regression models for the three transitions of the illness-death model.

Dans le cadre des analyses des données de survie, la comparaison de plusieurs groupes d’individus, où l'événement d'intérêt est censuré par intervalle, représente un défi méthodologique. Lorsque le suivi des patients au cours de l'étude n'est pas continu, l'événement d'intérêt pourra survenir entre deux dates d'observation; il est dit censuré par intervalle. Des tests de comparaisons des distributions des temps de survie pour plusieurs groupes, adaptés à la censure par intervalle, ont été développés (tests du score, tests de pseudo log-rank pondérés, tests des rangs). C’est dans ce contexte que nous avons proposé deux nouveaux tests de comparaisons de groupes adaptés à des situations particulières de censure par intervalle. Le premier test concerne une situation où l’hypothèse alternative considère que les fonctions de risque instantané se croisent. Le second test concerne une situation où la population étudiée comporte une fraction non à risque pour l’événement d’intérêt. Ces deux tests ont fait l'objet d'une application sur des données réelles d'immunologie clinique. / In the context of analysis of survival data, the comparison of several groups of individuals, where the event of interest is interval censored, represents a methodological challenge. When the monitoring of patients during the study is not continuous, the event of interest may occur between two observation dates; it is said "interval censored". Tests of comparisons of survival time distributions for several groups, adapted for interval censoring, have been developed (score tests, weighted pseudo log-rank tests, rank tests). In this context, we have developped two new group comparison tests adapted to the particular situations of interval censoring. The first test apply to a situation where the alternative hypothesis considers that the hazard functions cross. The second test concerns a situation where the study population has a fraction not at risk for the event of interest. Both of these tests have been applied to real clinical immunology dataset.

In survival analysis, proportional hazards model is the most commonly used and the Cox model is the most popular. These models are developed to facilitate statistical analysis frequently encountered in medical research or reliability studies. In analyzing real data sets, checking the validity of the model assumptions is a key component. However, the presence of complicated types of censoring such as double censoring and partly interval-censoring in survival data makes model assessment difficult, and the existing tests for goodness-of-fit do not have direct extension to these complicated types of censored data. In this work, we use empirical likelihood (Owen, 1988) approach to construct goodness-of-fit test and provide estimates for the Cox model with various types of censored data. Specifically, the problems under consideration are the two-sample Cox model and stratified Cox model with right censored data, doubly censored data and partly interval-censored data. Related computational issues are discussed, and some simulation results are presented. The procedures developed in the work are applied to several real data sets with some discussion.

A device that is accompanied by an irreversible chemical reaction or physical destruction and could no longer function properly after performing its intended function is referred to as a one-shot device. One-shot device test data differ from typical data obtained by measuring lifetimes in standard life-tests. Due to the very nature of one-shot devices, actual lifetimes of one-shot devices under test cannot be observed, and they are either left- or right-censored. In addition, a one-shot device often has multiple components that could cause the failure of the device. The components are coupled together in the manufacturing process or assembly, resulting in the failure modes possessing latent heterogeneity and dependence. Frailty models enable us to describe the influence of common, but unobservable covariates, on the hazard function as a random effect in a model and also provide an easily understandable interpretation. In this thesis, we develop some inferential results for one-shot device testing data with gamma frailty model. We first develop an efficient expectation-maximization (EM) algorithm for determining the maximum likelihood estimates of model parameters of a gamma frailty model with exponential lifetime distributions for components based on one-shot device test data with multiple failure modes, wherein the data are obtained from a constant-stress accelerated life-test. The maximum likelihood estimate of the mean lifetime of $k$-out-of-$M$ structured one-shot devices under normal operating conditions is also presented. In addition, the asymptotic variance–covariance matrix of the maximum likelihood estimates is derived, which is then used to construct asymptotic confidence intervals for the model parameters. The performance of the proposed inferential methods is finally evaluated through Monte Carlo simulations and then illustrated with a numerical example. A gamma frailty model with Weibull baseline hazards is considered next for fitting one-shot device testing data. The Weibull baseline hazards enable us to analyze time-varying failure rates more accurately, allowing for a deeper understanding of the dynamic nature of system's reliability. We develop an EM algorithm for estimating the model parameters utilizing the complete likelihood function. A detailed simulation study evaluates the performance of the Weibull baseline hazard model with that of the exponential baseline hazard model. The introduction of shape parameters in the component's lifetime distribution within the Weibull baseline hazard model offers enhanced flexibility in model fitting. Finally, Bayesian inference is then developed for the gamma frailty model with exponential baseline hazard for one-shot device testing data. We introduce the Bayesian estimation procedure using Markov chain Monte Carlo (MCMC) technique for estimating the model parameters as well as for developing credible intervals for those parameters. The performance of the proposed method is evaluated in a simulation study. Model comparison between independence model and the frailty model is made using Bayesian model selection criterion. / Thesis / Candidate in Philosophy