Tutors
Nikos Pantazis
Objective and Scope
The main objective of the course is to familiarize students with the fundamental principles of the family of Generalized Linear Models (GLMs) and their practical application. The course also focuses on specific models within the GLM framework that are widely used in the Health Sciences. More specifically, emphasis is placed on the criteria for selecting the appropriate model according to the type of data under analysis Generalized Linear Models, as well as on the implementation, interpretation of results, and assessment of model goodness-of-fit.
Learning Outcomes
Upon successful completion of the course, students are expected to have acquired an in-depth understanding of the most widely used GLMs available in commonly used statistical software packages (such as Stata, SAS, and R). In particular, students should be able to: a) select the most appropriate model according to the design of the biomedical study, b) apply the model to real-world data, c) interpret model results in a clear and accessible manner for other health scientists, d) evaluate model goodness-of-fit, and e) critically assess the appropriate use of statistical models in published biomedical studies. Finally, students should understand the fundamental principles underlying the development of any model belonging to the family of Generalized Linear Models.
Course Content
- Introduction to the Structure of Generalized Models:
Review of the basic principles of normal regression models. Review of maximum likelihood theory. The exponential family of distributions. Basic structure of generalized linear models. Comparison of nested GLMs. Assessment of GLM goodness-of-fit. Residual analysis. Algorithms for fitting generalized linear models.
- Models for Binary and Binomial Data – Logistic Regression: Logistic, probit, and complementary log-log (c-log-log) link functions. Parameter estimation methods. Application of linear models to binomial data. Assessment of confounding effects, interactions with continuous and categorical variables, and interpretation of results. Linear trends in proportions. Examples of final model selection.
- Assessment of Logistic Regression Models: Model goodness-of-fit assessment: Deviance, Pearson X², and Hosmer–Lemeshow statistic. Definition of residuals. Residual plots. Definition and identification of outliers. Identification of influential observations. Handling influential observations. Assessment of binomial distribution assumptions.
- Models for Matched Case-Control Studies:
Advantages of matching. Conditional likelihood. Basic analysis of matched data. Conditional logistic regression models. Assessment of model’s goodness of fit and interpretation of model’s results.
- Multinomial and Ordinal Logistic Regression Models: multinomial (πολυωνυμικής) και ordinal (διατεταγμένης) λογιστικής εξάρτησης
Basic principles and applications. Overdispersion: possible causes, models for overdispersed data, and comparison of methods addressing overdispersion. Offset term (varying follow-up time). Models for rates. Interpretation of results. Assessment of assumptions.
- EPoisson Regression Poisson:
Basic principles and applications. Overdispersion: possible causes, models for overdispersed data, and comparison of methods addressing overdispersion. Offset term (varying follow-up time). Models for rates. Interpretation of results. Assessment of assumptions.
- Υπερδιασπορά (Overdispersion): Πιθανοί λόγοι. Μοντέλα για δεδομένα με υπερδιασπορά. Σύγκριση μεθόδων που αντιμετωπίζουν την υπερδιασπορά.
Μέθοδοι Διδασκαλίας
Το περιεχόμενο του μαθήματος καλύπτεται σε 52 ώρες, οι οποίες διαχωρίζονται σε 13 δίωρα θεωρίας (διαλέξεων) και σε 13 δίωρα πρακτικής άσκησης σε ΗΥ.
Μέθοδος Αξιολόγησης
3 Εργασίες και τελική γραπτή εξέταση. Για να θεωρηθεί επιτυχής η ολοκλήρωση της παρακολούθησης του μαθήματος απαιτείται ο/η φοιτητής/τρια να έχει λάβει τουλάχιστον προβιβάσιμο βαθμό στην γραπτή εξέταση. Υπό αυτή τη συνθήκη, ο τελικός βαθμός υπολογίζεται ως: (μέσος όρος βαθμολογίας εργασιών)*0,4+(βαθμός γραπτής εξέτασης)*0,6.
