The time will be divided equally between classroom lectures and practical sessions. All faculty will be present at the practical sessions; exercises with worked solutions will be provided but these sessions are also intended as a forum for participants to talk to the faculty about aspects of particular interest to them. We have several group rooms available and can organise informal mini-lectures or group discussions on topics of particular interest.
Presenters: Paul Lambert, Paul Dickman
Introduction: A brief revision of concepts and classical models used in survival analysis (including a review of Poisson regression and Cox regression).
Flexible parametric models (FPM): The majority of survival analyses make use of the Cox proportional hazards model, with researchers generally being less keen on using parametric survival models due to the restrictive assumptions about the shape of the underlying hazard/survival functions. Flexible parametric models make use of restricted cubic splines to fit models for survival data. One of the advantages of the models is the ease at which time-dependent effects (e.g. for non-proportional hazards) can be fitted. A further advantage of parametric models is the ability to make various useful predictions, for example time-dependent hazard ratios, differences in hazard functions or differences in survival functions. The aim of this lecture is to provide an understanding of how to fit and interpret flexible parametric models, and to demonstrate a variety of useful predictions from the models, including a variety of ways to quantify differences between groups.
Presenters: Mark Rutherford, Anna Johansson
APC-models: Age–period–cohort (APC) models provide a framework for modeling trends in incidence and mortality rates. This lecture will centre on analyzing, interpreting and presenting time trends according to the effects of age, calendar period and birth cohort. Concepts like drift, identifiability and splines will be discussed. The emphasis will be on practical applications to real cancer incidence data, and the use of the apcfit package to fit age, period and cohort effects via restricted cubic spline models in Stata.
Outcome-selective sampling designs: Key concepts in sampling from cohorts to generate epidemiological studies with a sample of controls will be addressed. The emphasis will be on the case-cohort design and the nested case control design. The history of these designs and a comparison between them will be discussed. Examples from applications in real studies will be presented.
Presenters: Sandra Eloranta, Michael Crowther
An introduction to survival in the presence and absence of competing risks and its associated methodology:
Competing risks: Key quantities such as the cause-specific, marginal and subdistribution hazards, as well as their survival analogs (cause-specific cumulative incidence, Kaplan-Meier function etc) will be explained. The focus of the course will be to provide a conceptual understanding of when statistical methods for competing risks are required (given a specific research question), what methodology and software are available and how results are interpreted.
Multi-state models: Multi-state models allow rich insights into complex disease pathways, where a patient may experience many non-fatal/intermediate events, and we wish to the investigate covariate effects for each specific transition between two states, not just for example, on the first event, or a terminal event. This course will introduce the basic concepts of multi-state models, including both Markov and semi-Markov models, and describe the calculation of useful quantities such as transition probabilities. Current methodological approaches will be described along with available software.
Presenters: Michael Crowther
Joint modelling: The joint modelling of longitudinal and survival data has been an area of growing interest in recent years, with many applications in cancer and cardiovascular disease. The models can provide both an effective way of conducting an analysis of a survival endpoint (e.g. time to death), influenced by a time-varying covariate measured with error, or alternatively correct for non-random dropout in the analysis of a longitudinal outcome (e.g. a biomarker). In this session, I will introduce joint modelling through real applications, describing the methodological framework and underlying assumptions, estimation, model building and predictions. The practical will use the stjm package in Stata.
+ Half-day seminar on Real World Evidence studies in practice (programme here)