Survival Analysis - Introduction

This resource will introduce key concepts underlying survival analysis and will develop examples of analyses applied to discrete time intervals. Concepts and models will be illustrated using examples and exercises based on R software. 

Researchers are often interested in describing and predicting the occurrence of events, e.g.: the onset of puberty in adolescence; age at time of retirement; life expectancy of different groups of individuals; the first re-occurrence of disease symptoms after an intervention (relapse). Common descriptive and inferential statistical methods are inadequate in describing and explaining event occurrences because of the particular characteristics of the outcome of interest: event occurrence concerns whether and when a target event takes place, thus encompassing different types of information. Furthermore, the study of event occurrence is complicated by difficulties in recording the timing of the target event among study participants, commonly referred to as censoring. 

Survival analysis includes a set of statistical methods to describe and explain occurrence of events, taking into account the complex nature of the information of interest and common issues in event occurrence datasets, such as censoring. Survival analysis methods thus provide  meaningful descriptions of “whether” and “when” an event takes place, as well as explanations concerning what factors and characteristics are associated with differences in the likelihood of the target event taking place. 

The application of survival analysis methods however requires careful consideration of the complexity of the data of interest in order to develop reliable explicatory models. In particular, the data of interest often do not follow commonly-used distributions (e.g., a normal distribution): modelling requires data transformations that may complicate model reading and interpretation significantly. Survival analysis models are also based on sets of assumptions that need to be assessed carefully and tested: assumptions that are unwarranted or implausible can lead to incorrect inference.

Many introductory resources on survival analysis fail to articulate the basilar concepts underlying these methods: consequently, complex models (e.g., Cox regression models) are often presented without providing the critical tools to appreciate the assumptions and implications of the models. To address this issue, the presentations and other material included within these resources will articulate a conceptual framework for describing occurrence of events. Furthermore, these resources will guide the reader in building models of event occurrence based on simpler instances of discrete time event occurrence. These models provide a critical foundation for articulating more complex models applied to continuous time event occurrence.

Tweet    Download transcript    |   Download slides [ 104 Views ]

Tweet    Download transcript    |   Download slides [ 78 Views ]

Tweet    Download transcript    |   Download slides [ 91 Views ]