Course Level: Beginner
Format Requirements: This is not a statistical course and so detailed statistical skills are not a pre-requisite. Nevertheless, we will be discussing statistical techniques and so basic familiarity with regression analysis will be assumed. The focus on the use of survival analysis for decision modeling will assume that participants are familiar with the basic concepts of model building and understand the distinction between decision trees, Markov Models, Discrete Event Simulation etc. in addition to didactic sessions, we do plan to finish the course with a debate with the participants to explore how, having been exposed to the material and arguments presented in the course, they would anticipate that survival data would be used. In particular, which modeling form (partitioned survival analysis, Markov modeling, DES) is best suited to modeling survival data, and how they think the techniques associated with survival analysis could be best exploited to inform model
Background: This course considers the way in which survival or time to event data is used to inform decision modeling. This is intended to be a practical course focusing on the use of survival analysis, not a statistical course. Statistical considerations will be covered, but the focus is on implementation of survival models and the utility of survival analysis to inform models. This can be for censoring adjustment, extrapolation, or the calculation of transition probabilities. In particular, a form of modeling used extensively with oncology data, known as partioned survival models (closely related to Q-TWIST) will be introduced and compared to more common models in our field (Markov of DES models).
Description and Objectives: Survival data are ubiquitous in medical research and medical decision modeling often requires the modeling of the full lifespan of patients in order to calculate outcomes for comparative effectiveness such as life-years and Quality Adjusted Life Years (QALYs). Perhaps the clearest overlap between the use of survival analysis and decision modelling is in the area of oncology, where the life expectancy of patients is such that empirical studies (often RCTs) are designed that follow patients for a good proportion of their remaining life expectancy. The analysis of these data by the medical statistical community makes extensive use of non-parametric methods such as Kaplan-Meier curves and Cox Proportional Hazards models. By contrast, the techniques used in the medical decision analysis community require techniques for extrapolation typically resulting in more extensive use of parametric methods. In particular, ‘partitioned survival analysis has emerged as a popular form of modeling for oncology data (closely related to Q-TWIST). In the proposed course, we will review the use of survival data from the perspective of the decision modeler, with the objectives of improving the understanding of the participants around:
- The standard use of survival analysis in medical statistics
- The rationale for a different approach to support modeling
- The introduction of partitioned survival models as a particular form of modeling (and its relationship to Q-TWIST)
- How survival models can inform parameter estimation for Markov and DES models
- Importance of capturing heterogeneity in decision models