Πλοήγηση ανά Επιβλέπων "Vasdekis, Vassilis"

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

Α Β Γ Δ Ε Ζ Η Θ Ι Κ Λ Μ Ν Ξ Ο Π Ρ Σ Τ Υ Φ Χ Ψ Ω

Τώρα δείχνει 1 - 19 από 19

Application of linear mixed models in the evaluation of the impact of hypo-chlorhydria in stomach on the physicochmemical characteristics of contents of stomach and upper small intentine of fasted adults
(29-07-2016) Vertzoni, Maria; Athens University of Economics and Business, Department of Statistics; Vasdekis, Vassilis
Evaluate the impact of reduced gastric acid secretion after administration of two acid-reducing agents on the physicochemical characteristics of contents of upper gastrointestinal lumen of fasted adults using Linear Mixed Model methodology. Materials and Methods Eight healthy male adults, fasted from food for 12h, participated in a three-phase crossover study. Phase 1: No drug treatment prior to aspirations. Phase 2: Oral administration of 40mg pantoprazole at ~9am the last three days prior to aspirations and at ~7am on aspiration day. Phase 3: Oral administration of 20mg famotidine at ~7pm prior to aspirations and at ~7am on aspiration day. Samples from the contents of upper gastrointestinal lumen were aspirated for 50 min, after administration of 240ml table water. For each parameter (pH, buffer capacity, chloride ion concentration, osmolality, total protein content, free fatty acid content, total phosphatidylcholine content and cholesterol concentration), data from the three different treatments (Phase 1: Control, Phase 2: Pantoprazole and Phase 3: Famotidine), from two different locations (stomach and end of duodenum) and time points were obtained from the same subject and, therefore, were assumed to be correlated. Linear Mixed Model analysis was used to model the correlation between treatments and/or between aspiration time points using the MIXED procedure of SPSS Statistics 17.0.
An application of nonlinear mixed models for longitudinal data
(Athens University of Economics and Business, 02-2015) Chatzopoulos, Michalis; Vasdekis, Vassilis
Thesis - Athens University of Economics and Business. Postgraduate, Department of Statistics
Dantzig selector in linear models in n<p problems
(16-02-2017) Georgopoulos, Nicolaos; Athens University of Economics and Business, Department of Statistics; Vasdekis, Vassilis
In many important statistical applications, the number of variables or parameters p is much larger than the number of observations n. In radiology and biomedical imaging, for instance, one is typically able to collect far fewer measurements about an image of interest than the unknown number of pixels. Examples in functional MRI and tomography all come to mind. High dimensional data frequently arise in genomics. Gene expression studies are a typical example: a relatively low number of observations (in the tens) is available, while the total number of genes assayed (and considered as possible regressors) is easily in the thousands. Other examples in statistical signal processing and nonparametric estimation include the recovery of a continuous-time curve or surface from a finite number of noisy samples.
Dealing with missing values on variables using multiple imputation methods to Cox regression analysis
(2020) Papadimitriou, Nikolaos; Παπαδημητρίου, Νικόλαος; Athens University of Economics and Business, Department of Statistics; Vasdekis, Vassilis
In the field of Survival Analysis, where the complete case analysis is the common method, we exclude cases with missing values. In order to take advantage of the whole dataset, we propose multiple imputation methods to cope with missing data. To implement these methods, a fully observed variable is necessary to exist in the dataset. This fully observed variable offers closest to the real values estimations of the other variable with missing values. More specifically, the proposed multiple imputation methods in this thesis are the following: the semi-parametric predictive mean matching and the non-parametric nearest neighbor multiple imputation. In order to evaluate the performance of the aforementioned methods, Cox regression analysis is employed. In the end, the methods are compared in terms of efficiency, robustness and consistency.
Detection of variance homogeneity violation in mixed effects models
(06/21/2019) Lamprou, Dimitra D.; Λάμπρου, Δήμητρα Δ.; Athens University of Economics and Business, Department of Statistics; Merkouris, Panagiotis; Psarakis, Stelios; Vasdekis, Vassilis
In this thesis, we study Mixed-effects Models which are widely used in longitudinal studies. Firstly, we give the definition of Linear Mixed Effects Models, explaining the fixed and random effects and describing a two-stage model concept. Subsequently, we write the log-likelihood function for the response in model and estimate parameters. Then we give the definition of Generalized Linear Mixed Models. Afterwards, we propose two tests for checking the homogeneity of the covariance structure assumption across subjects. The first test is the model based test whose covariances matrices are computed from the fitted model and the other is the empirical test whose the empirical variation is computed from the estimated random effects. We used simulation in order to observe how the two tests are performed for detecting violations of the homogeneity assumption in the mixed-effects models. In the application, we analyze the data from 27 rats and examine the performances of the test in this example.
Estimation of the skeleton of directed acyclic graphs with PC and PenPC algorithms
(07/13/2018) Vontas, Enias; Athens University of Economics and Business, Department of Statistics; Ntzoufras, Ioannis; Psarakis, Stelios; Vasdekis, Vassilis
Causal questions and relationships are important in many scientific fields. When we are able to discern such a relationship we can have a much clearer view of the solution of the problem at hand. Directed Acyclic Graphs (DAGs) can communicate such causal relationships among its nodes, which represent random variables, and they can do it in a compact and easy to understand way. An important first step in the estimation of the underlying DAG produced from random variables is the estimation of its skeleton. The methods explored in this thesis are two: the PC-stable and PenPC algorithms. In the first case, the authors proposed an algorithm that would require testing for sparse graphs as few independence relations as possible. In the second case, the authors made an important contribution. They added a step in the beginning of the PC-stable algorithm where for each node we estimate its adjacent nodes through penalized regression. Their second step is a modified PC-stable algorithm. In the end, PenPC has higher sensitivity and specificity compared to PC-stable algorithm.
Evidence synthesis: from meta-analysis to network meta-analysis with an application in patients with COPD
Thano, Adriana; Θάνο, Αντριάνα; Athens University of Economics and Business, Department of Statistics; Ntzoufras, Ioannis; Vasdekis, Vassilis
Evidence synthesis methodologies become essential as more and more analyses are available for a specific research question. This dissertation has been focused on the evidence synthesis methods in healthcare, using randomized control trials (RCT) as a source of evidence. The first method described is the meta-analysis, an overall analysis to pool the treatment effect of two specific treatments being compared directly. The meta-analysis technique has two effect models, the fixed and the random effects, which their differentiation relies on a fundamental assumption over the uncertainty sources; the latter assumes between-study variance in additional to the within-study variance, which is the only source of variability in the fixed effect model. Furthermore, the indirect treatment comparisons (ITC) overcomes the limitation of the meta-analysis, making feasible the comparison of treatments without the requirement of them to be directly compared in an RCT. The ITC uses a common comparator, a treatment which has been compared with the other two treatments of interest, if both indirect and direct evidence are available a pooled estimation can be performed. The ITC and pooled effect methodologies can be considered as mixed treatment comparisons (MTC), however, since they are based on trivial mathematical equations they cannot exploit the geometry of the network made by the treatments connected. The last and most important evidence synthesis tool that has been presented is the network meta-analysis, the extension of meta-analysis. A network of multiple treatments, connected directly or indirectly by multiple studies is analyzed simultaneously by fixed or random effects. The dissertation is organized in two parts; the theory of these methods, accompanied with examples in the Bayesian and frequentist prospective for continuous outcomes, and an extensive application in network meta-analysis in patients with COPD, using a publication performed by Mapi [1]. The main scope of this thesis has been to present both in theory and application all the main steps of evidence synthesis and compare the estimations among different approaches and models. As a conclusion, the Bayesian and frequentist approaches deemed to result in approximately same estimations, with the random effects estimations in both cases providing more uncertainty around them.
Expanding the application of the I-squared statistic to characterize heterogeneity in treatment effects within cluster randomized trials, multi-center randomized trials, and meta-analyses using individual patient data
(26-09-2023) Saloustros, Alexandros; Σαλούστρος, Αλέξανδρος; Athens University of Economics and Business, Department of Statistics; Psarakis, Stelios; Demiris, Nikolaos; Vasdekis, Vassilis
Ο έλεγχος της ομοιογένειας της επίδρασης της θεραπείας στις μετα-αναλύσεις στοχεύει στο να προσδιορίσει εάν η επίδραση της θεραπείας διαφερει σημαντικά μεταξύ διαφορετικών μελετών. Στις μετα-αναλύσεις με συγκεντρωτικά δεδομένα (aggregate data) είναι συνήθες να ποσοτικοποιείται η έκταση της ομοιογένειας της θεραπευτικής επίδρασης χρησιμοποιώντας μια εύκολα κατανοητή έννοια που ονομάζεται I-τετράγωνο. Επιπλέον, είναι σημαντικό να εξεταστεί εάν το αποτέλεσμα μιας αγωγής διαφέρει μεταξύ των συστάδων σε μια τυχαιοποιημένη κλινική δοκιμή ή μεταξύ των κέντρων σε μια τυχαιοποιημένη πολυκεντρική δοκιμή. Όταν διεξάγονται δοκιμές που χρησιμοποιούν διασταυρούμενη (cross over) σχεδίαση και άλλα τυχαιοποιημένα σχέδια, όπου οι συστάδες ή τα κέντρα εκτίθενται και στις δύο συνθήκες αγωγής και ελέγχου, προκύπτει η ανάγκη εξέτασης διαφορών των φαρμακευτικών επίδρασεων. Στα πλαίσια αυτής της έρευνας, αξιολογούμε μια εναλλακτική μεθοδο μέτρησης για το I-τετράγωνο, το οποίο βοηθά στον προσδιορισμό της έκτασης της ετερογένειας στην επίδραση της θεραπείας μεταξύ των συστάδων ή των κέντρων σε τυχαιοποιημένες δοκιμές. Επιπλέον, παρουσιάζουμε την εφαρμοσιμότητα αυτής της μεθοδολογίας στην εκτίμηση της ομοιογένειας του αποτελέσματος της αγωγής για μετα-ανάλυση χρησιμοποιώντας ατομικά δεδομένα ασθενών.
Generalised linear mixed models with the new proc nlmixed procedure
Papakosta, Panagiota; Athens University of Economics and Business, Department of Statistics; Vasdekis, Vassilis
In this Thesis, we study the Generalised Linear Mixed Models with application in SAS with the new Proc Nlmixed Procedure. Firstly, we talk about Fixed Models and afterwards we pass in the definition of Mixed Models, explaining the Random Effects Models, Covariance Pattern Models and Random Coefficient Models. We more specifically, continue in the Normal Mixed Models, giving their definition and the models we also meet, as the Covariance Pattern Models with their Covariance Structure. Next, we define the Generalised Linear Mixed Models. We continue, by giving the New Proc Nlmixed Procedure in SAS and analyzing Adaptive Gaussian Quadrature Method, the approximation to the Log-Likelihood Function and the Quasi-Newton Optimization Algorithm. Additionally, we give an example of Logistic Regression with correlated binary data, in Proc Nlmixed Procedure. Finally, in the last chapter we introduce correlated binary data from an Arthritis Clinical Trial and analyze them with the Proc Nlmixed Procedure
Goodness of fit tests for random effect models with binary responses estimated viah-likelihood
(23-05-2016) Korre, Antonia K.; Athens University of Economics and Business, Department of Statistics; Vasdekis, Vassilis
The objective of this study is to introduce goodness of fit statistics for thesystematic part of random effect models. These statistics are formed as scoretests and are based on partitioning the observations into mutually exclusivegroups. Weighted versions of these statistics are also introduced which arebased on the correlation between an appropriately adjusted candidate covariatefor entrance into the model and the model residuals. In addition tothis, weighted cumulative and moving sum processes are also provided witha goal to detect, in a more effecient way, various departures from the nullmodel, as compared to the usual cumulative and moving sums. The estimatingprocedure that is used throughout this study is the h-likelihood, sothe quantities needed for the above mentioned statistics are derived basedon this estimating approach. The simulations were designed so as to examinethe performance of the tests under different factors, such as sample size,magnitude of random effects’ variance, type of the model misspecification,etc. The results indicate the superior performance of the weighted tests ofany type (either those based on partitioning or those based on cumulativesums), there are cases, however, were some unweighted tests displayed a similarpower as their weighted analogous. The proposed tests are illustratedwith the use of two real datasets.
Joint mean-covariance models with applications to longitudinal data: unconstrained parameterisation
Georgountzou, Athanasia; Γεωργούντζου, Αθανασία; Athens University of Economics and Business, Department of Statistics; Vasdekis, Vassilis
Longitudinal data is a special type of multivariate data that occur when counts, measurements or categorical responses are obtained from one or more experimental units or subjects through time. A number of various approaches for representing longitudinal data in terms of a statistical model have been developed. The data used in this thesis are from Kenward (1987) that reports an experiment in which cattle were assigned randomly to two treatment groups A and B, and their weights were recorded to study the effect of treatments on intestinal parasites. Thirty animals received treatment A and another 30 received treatment B. They are weighed 11 times over a 133-day period; the first 10 measurements on each animal were made at two-week intervals and the final measurement was made one week later. The measurement times were common across animals and were rescaled to t=1,2,…,10,11. No observation was missing so this is a balanced longitudinal dataset. Our goal is to find which model among these presented in this thesis fits best to the two different groups of our data. We compare among joint mean- covariance models (Pourahmadi, 1999) and mixed effects models.
Multi-phase nonlinear mixed effects model applied to spirometry data
(06/27/2019) Lika, Aglina; Athens University of Economics and Business, Department of Statistics; Ntzoufras, Ioannis; Tsiamyrtzis, Panagiotis; Vasdekis, Vassilis
The nonlinear mixed effects models are models that can model time to event data and can handle the most cases of temporal trend. Also, they accommodate the scenario of the existence of an interaction effect between some factors and time on the longitudinal response variable. Because of their flexibility they have become popular and have been used worldwide since 1990. The Jeevanantham et.a.l.(2014) proposed an multi-phase nonlinear mixed effects model to model temporal patterns of longitudinal measurements. The formulation and the estimationmethods of them are provided in this thesis and an application of them is illustrated using spirometry data of patients following intravenous or subcutaneous injection using R. For the balanced and unbalanced data sets of the intravenous injection and subcutaneous injection we usethe alternating and the Laplacian approximation method to estimate nonlinear mixed effects models, while to estimate suitable multi-phase mixed effects model for each data set we use only the alternating approximation. From the application is illustrated the usefulness of the modelswhen dealing with nonlinear models and time-varying coefficients.
Nonparametric meta-analysis for diagnostic accuracy studies
(07/10/2018) Tasioula, Marilena Th.; Τασιούλα, Μαριλένα Θ.; Athens University of Economics and Business, Department of Statistics; Vasdekis, Vassilis
In this study we use a non parametric approach for the meta-analysis of diagnostic studies. We are interested in realizing whether this model can be appropriate for the meta-analysis. For this reason we present two more models, the bivariate logistic random-effects model and the plackett copula model. Then we do a simulation study using the non parametric model. Finally we compare these three models by using two examples and we come to some conclusions.
Pairwise likelihood estimation in longitudinal data analysis: a comparison with full likelihood approach
Chatziagapoglou, Lazaros; Χατζηαγάπογλου, Λάζαρος; Athens University of Economics and Business, Department of Statistics; Vasdekis, Vassilis
The importance of longitudinal studies has grown tremendously over the past twenty years due to their unquestionable advantages. While it would appear that they are now considered foundational for drawing causal inference, they are not without limitations. Many times high dimension of the joint covariance matrix results in computational complexity and requires special handling in dealing with this situation. Subject of this article is the presentation of the most prominent ways of analyzing longitudinal continuous data and the proposal of pairwise approach in order to confront computational problems that might appear.
The potential for more robust inference using multiple statistics by combining p-values via permutations
(06/14/2018) Konstantinou, Aliki - Evangelia; Κωνσταντίνου, Αλίκη - Ευαγγελία; Vasdekis, Vassilis
Statistical inference in randomized experiments is usually based on one single test statistic considered as optimal for testing a hypothesis. Several methods that combine multiple tests for the same hypothesis instead of a single pre-specified test have been proposed in the past, suggesting greater statistical power. This thesis aims to implement and test the results of one method that uses multiple p-values from different models via permutations to test the effect of treatment in randomized clinical trials. Simulations and real datasets will be used for this purpose.
Serum uric acid, endothelial function, and small vessel remodeling: a compound association in humans
(01/03/2022) Alexopoulos, Georgios I.; Αλεξόπουλος, Γεώργιος Ι.; Athens University of Economics and Business, Department of Statistics; Vasdekis, Vassilis
The purpose of this study is to assess whether SUA can provide additional information on the severity of resistance artery remodeling than that obtained from common scores used to define the patient’s cardiovascular CV risk,ans also whether the relationship between SUA and resistance artery remodeling is mediated by endothelial dysfunction and reduced NO availability.
Statistical and machine learning regularization techniques in clinical biostatistics: a comprehensive evaluation
(07-06-2024) Σταμάτης, Παναγιώτης; Stamatis, Panagiotis; Athens University of Economics and Business, Department of Statistics; Demiris, Nikolaos; Psarakis, Stelios; Vasdekis, Vassilis
The primary objective of this thesis is to investigate the efficacy of regularization techniques within the domain of clinical biostatistics. A comprehensive exploration of statistical and machine learning methodologies, including Penalization, Early Stopping, and Ensembling, is undertaken. Regularization, as defined, serves to control model complexity by incorporating additional information to address ill-posed problems or mitigate overfitting. Despite its conceptual clarity, the full extent of its applicability and diverse variants remains not entirely elucidated. Leveraging the R software, these techniques are applied to two distinct clinical datasets, both pertinent to prostate cancer research.The first dataset aims to classify patients into benign or malignant tumor categories, wherein Penalization, specifically Ridge Regression, demonstrates superior performance compared to alternative methods, achieving the lowest Misclassification Error (MCE) and highest Area Under the Curve (AUC). Furthermore, the second dataset endeavors to predict the logarithm of prostate-specific antigen (PSA), a significant biomarker, in conjunction with other clinical predictors. Once more, the penalization approach, notably Elastic Net, exhibits notable performance by yielding the lowest Mean Squared Error (MSE) and Mean Absolute Error (MAE). However, the outcomes for machine learning techniques are less promising, potentially attributable to the inherently simple data relationships or issues related to dimensionality. Overall, the study underscores the utility of regularization methods in enhancing predictive accuracy within clinical biostatistics, advocating for their broader adoption and further exploration within this domain.
Statistical methods for analysis under the presence of missing data
(16-07-2016) Stamelakou, Aikaterini; Vasdekis, Vassilis
Missing data are a recurring problem which can cause bias or lead to inefficient analysis, no matter how well a survey questionnaire is designed and no matter how effective is the data collection. These data need a special and meticulous handling in analysis. This is why so many statistical methods have been proposed and developed to address missingness. Some of them are based on deletion of incomplete cases, others try to predict each missing value and then to include the filled in value in analysis, these are called Simple Imputation Methods. Additionally, there is another method, known as Multiple Imputation, which is based on the creation of many imputed data sets by using Data Augmentation. In this thesis, each of these methods will be mentioned. Specifically, the Multiple Imputation method will be the main topic that will monopolize the interest and will be given special emphasis. In the context of this thesis included and an application of Linear Mixed Models in repeated measurements with data that are not complete. Applying different mixed effect models on these data we reach in the appropriate model through the Bayesian Information Criterion. In continue, we apply multiple imputation in our data and then fit the same models in the imputed data this time. Our main goal is to examine the similarities or differences that may have these two data sets
The use of logistic regression models in meta-analysis
(27-04-2020) Dimitrakaki, Alexandra; Δημητρακάκη, Αλεξάνδρα; Athens University of Economics and Business, Department of Statistics; Vasdekis, Vassilis
Το μοντέλο του ενός σταδίου είναι ένα μοντέλο λογιστικής παλινδρόμησης το οποίο μπορεί να χρησιμοποιηθεί στη μετα-ανάλυση δυαδικών μεταβλητών όταν μεμονωμένα δεδομένα συμμετεχόντων είναι διαθέσιμα. Το μοντέλο του ενός σταδίου μπορεί επίσης να χρησιμοποιηθεί όταν μεμονωμένα δεδομένα συμμετεχόντων δεν είναι διαθέσιμα αλλά δίνεται μόνο ο συνολικός αριθμός των γεγονότων και των συμμετεχόντων σε κάθε ομάδα θεραπείας. Σε σύγκριση με άλλες συμβατικές μεθόδους, το μοντέλο του ενός σταδίου δεν βασίζεται στην υπόθεση της κανονικής κατανομής των επιδράσεων. Ως αποτέλεσμα, το μοντέλο του ενός σταδίου μεγιστοποιεί τη σωστή διωνυμική πιθανότητα για αυτά τα δεδομένα. Σκοπός αυτής της μελέτης είναι η περιγραφή του ενός σταδίου μοντέλου λογιστικής παλινδρόμησης και της χρήσης του στη μετα-ανάλυση δίτιμων δεδομένων. Εφαρμογές του μοντέλου του ενός σταδίου στη μετα-ανάλυση μεμονωμένων δεδομένων συμμετεχόντων, στην ανάλυση αραιών δεδομένων, στην ανάλυση υποομάδων, στη μετα-παλινδρόμηση και στη μετα-ανάλυση ακρίβειας διαγνωστικής δοκιμασίας παρατίθενται με παραδείγματα. Επίσης, το μοντέλο του ενός σταδίου συγκρίνεται με άλλες συμβατικές μεθόδους όπως το διμεταβλητό μοντέλο και το μοντέλο της ιεραρχικής συνοπτικής καμπύλης λειτουργικού χαρακτηριστικού δέκτη και κατασκευάζεται η καμπύλη λειτουργικού χαρακτηριστικού δέκτη. Η εργασία αυτή βασίζεται στο άρθρο των Simmonds και Higgins (2016).