Survival Analysis

Tutors

E. Kritsotakis

C. Thomadakis

 

Objective and Scope

The Survival Analysis module covers essential topics for analysing time-to-event data or survival times. The emphasis is on practical aspects of applied data analysis in real-world datasets, without overlooking the underlying statistical theory. The module begins by illustrating the special features of survival data and introducing the basic mathematical functions central to describing and analysing survival times (the survival, hazard, and cumulative hazard functions), discussing their interrelationships and links to the cumulative distribution and probability density functions. A revision of the basic theory of Maximum Likelihood Estimation and the Delta method for variance estimation takes place early to ensure effective application of these methods in the context of survival analysis. We then explain methods for summarising survival data using non-parametric techniques to estimate survival functions such as the Kaplan-Meier and the Nelson-Aalen estimators, and consider the derivations of variances and confidence intervals. We proceed with methods for comparing survival time distributions between two or more groups and discuss the derivation of the log-rank test, its underlying assumptions, and its relation to other tests, such as the Wilcoxon and Peto-Peto test. The module then turns to regression models for survival outcomes, emphasising the formulation and application of the Cox proportional hazards model (estimation, testing, prediction and interpretation of model coefficients), and ways to assess model validity in practice. Some more general issues in regression modelling are discussed, including model fit, linearity assessment, flexible modelling, and variable selection. We then extend the Cox model to incorporate time-dependent explanatory variables. Alternative proportional-hazards models based on parametric distributions (exponential, Weibull) are also examined. Finally, we introduce methods for summarising times-to-event for different event causes (competing risks) and describe models for cause-specific survival data. Class materials include lecture presentations, readings (journal papers and book chapters), practical exercises and an individual assignment using the R software to implement survival analyses of real-world clinical and epidemiological studies. Stata code is also provided as an optional study.

Εκπαιδευτικοί στόχοι του μαθήματος

Μετά την επιτυχή ολοκλήρωση του μαθήματος, οι φοιτητές θα πρέπει να έχουν αποκτήσει βαθιά γνώση των πλέον διαδεδομένων μεθόδων για την ανάλυση δεδομένων επιβίωσης. Συγκεκριμένα, οι φοιτητές θα πρέπει: (α) να κατανοούν τις θεμελιώδεις έννοιες και τη σημειολογία που χρησιμοποιούνται στην ανάλυση επιβίωσης, (β) να μπορούν να εφαρμόσουν και να ερμηνεύσουν τα αποτελέσματα από τα κύρια μοντέλα ανάλυσης επιβίωσης με απλό και σαφή τρόπο, (γ) να μπορούν να αξιολογούν την καλή εφαρμογή του μοντέλου, (δ) να κατανοούν τις βασικές αρχές που διέπουν το σχεδιασμό μιας μελέτης που περιλαμβάνει δεδομένα επιβίωσης, και (ε) να έχουν βελτιώσει τις υπολογιστικές τους δυνατότητες όσον αφορά τη χρήση Η/Υ για τη στατιστική ανάλυση δεδομένων.

Bibliography βιβλιογραφία

  1. Collett: Modelling Survival Data in Medical Research
  2. Cox and Oakes: Analysis of Survival Data
  3. Kleinbaum: Survival Analysis: A self-learning text
  4. Klein & Moeschberger: Survival Analysis: Techniques for censored and truncated data
  5. Allison: Survival Analysis Using the SAS System