Statistical methods in medicine and public health

Short courses by Profs. Lambert and Dickman in Melbourne, February 2018

Register here

Overview

Professors Paul Lambert and Paul Dickman will teach a series of three 1-day courses in Melbourne in
February 2018.

Wednesday 21/2: Beyond the Cox model – flexible parametric survival models and extensions (Lambert).

Thursday 22/2: Statistical methods for population-based studies of cancer patient survival (Dickman).

Friday 23/2: Hands-on with methods from the previous 2 days (Lambert and Dickman).

The content is structured so that participants may choose to attend any one day, any two days, or all three days. The first day will be held at Victorian Centre for Biostatistics (ViCBiostat, 553 St Kilda Rd) and the last 2 days will be held at Cancer Council Victoria (615 St Kilda Rd).

The first two days will consist of lectures and demonstrations of software implementations of the methods (primarily Stata). In day 3 participants will work hands-on with their own laptop and, potentially, their own data.

Day 1 will introduce flexible parametric survival models in detail and will not be restricted to applications in cancer. Day 2 will focus on studying the survival of cancer patients, including an overview of flexible parametric survival models and their application to estimating and modelling relative/net survival.

Day 3 is ‘free format' with no prepared lectures; participants will be provided with sample data sets and exercises covering the material from the first two days, but are also welcome to apply the methods to their own data. The aim of day 3 is for participants to explore aspects of specific interest to them; the faculty will do their best to answer any and all questions.

The focus of all days is on principles and application rather than mathematical detail; a degree in mathematics or statistics is not required, but some prior knowledge of survival analysis methods would be an advantage.

Professors Lambert and Dickman work primarily with Stata and have developed Stata commands to implement many of the methods they have developed. They have experience applying these methods in other software, such as SAS, R, and WinBUGS, and are willing to assist participants who wish to work with these packages but not all of the methods described in the course can be applied in other packages and they do not have the same level of expertise in other packages. Stata will be used to illustrate the methods, but the lectures are understandable without prior knowledge of Stata.

Program

Day 1: Beyond the Cox model - flexible parametric models and extensions

This day will consist of a series of lectures introducing flexible parametric survival models.

Flexible parametric models, also known as Royston-Parmar models, combine the desired features of both the Cox model and parametric survival models. While the Cox model makes minimal assumptions about the form of the baseline hazard function, prediction of hazards and other related functions for a given set of covariates is hindered by this lack of assumptions; the resulting estimated curves are not smooth and do not hold information about what occurs between the observed failure times.

Parametric models offer smooth predictions by assuming a functional form of the hazard, but often the assumed form is too structured for use with real data, especially if there exist significant changes in the shape of the hazard over time. Flexible parametric survival models provide researchers with the possibility to fit parametric models without imposing a restriction on the shape of the baseline hazard. Non-proportional hazards can be handled easily and because one has a parametric expression for the baseline cumulative hazard it is very easy to make smooth predictions of quantities of interest (e.g., hazard function, survivor function, expectation of life).

This day will consist of a series of lectures introducing flexible parametric survival models, illustration of the methods using real data, and demonstrations of how the methods are implemented in Stata. References will be provided for how the models can be estimates in SAS and R, but no demonstration will be provided.

Topics include:

  • A brief review of Cox Model & motivation of parametric models
  • Flexible parametric survival models with proportional hazards.
  • Flexible parametric survival models with time-dependent effects.
  • Simple predictions: survival, hazard and contrasts (differences/ratios)
  • Example applications
  • Attained age as the-time scale
  • Standardised survival curves and related measures.
  • Other predictions (Restricted mean survival time / loss in expectation of life).
  • Brief overview of further extensions:
    • Competing risks (cause-specific and subhazard models).
    • Random effect models.
    • Using splines on the log-hazard scale.

Day 2: Statistical methods for population-based studies of cancer patient survival

This day will consist primarily of a series of lectures introducing key concepts in population-based studies of cancer patient survival. That is, methods for estimating cancer patient survival using registry data. Heavy focus will be placed on the estimation and modelling of net survival.

Net survival is the most commonly-used measure of patient survival estimated from registry data, yet it is rarely used in other settings (not even cancer clinical trials). We will define net survival, describe why it is the measure of choice, and discuss its interpretation. We will also describe alternative measure and their relative merits.

  • What is 'population-based cancer survival analysis' and what makes it special compared to other applications of survival analysis?
  • The role of patient survival in cancer control;
  • Net survival; cause-specific survival; relative survival;
  • Relative merits of cause-specific survival and relative survival for population-based cancer registry data;
  • Comparison of methods (with focus on Ederer II and Pohar Perme) for estimating relative/net survival;
  • Interpreting relative/net survival estimates;
  • Age standardisation of relative/net survival, including model-based standardisation;
  • Cohort, complete, period and hybrid approaches to estimation;
  • Modelling excess mortality (relative survival) using Poisson regression and flexible parametric models;
  • Estimation in the presence of competing risks; crude probability of death;
  • Statistical cure; Cure models for relative survival; estimating and modelling the cure proportion; flexible parametric cure models;
  • Estimation of life expectation and proportion of expected life lost;
  • Estimating the number of avoidable premature deaths.

There will be some overlap with the previous day. We will introduce flexible parametric models, but not with the same level of detail as day 1. Many of the methods, including cure models and loss in expectation of life, are estimated in the framework of flexible parametric models but will not be covered on day 1 since the methods are primarily used in the field of population-based cancer survival. Participants with be provided with sample data and exercises with Stata code and worked solutions, but no time will be devoted to hands-on computing until day 3.

Day 3: Hands-on

The first 2 days contain overviews of a range of topics chosen by the teachers. Day 3, in contrast, will be participant-centric, and we will assist participants with whatever interests them. We have a large collection of additional reading material on the topics covered by the course that we are happy to provide to participants and discuss with them.

We will provide an extensive set of exercises with fully-worked solutions, but this day is also an opportunity for participants to discuss their own research projects with us (and with each other). Participants are encouraged to ask questions about their own projects and are welcome to bring their own data. For example, a participant with a particular interest in estimating loss in expectation of life will be encouraged to complete the exercise on that topic under our supervision. If the participant has his/her own data then we will assist in modifying the code to suit.

Paul Lambert

Paul LambertPaul Lambert is Professor of Biostatistics in the Department of Health Sciences at the University of Leicester. Paul also has a position (30% FTE) at the Department of Medical Epidemiology and Biostatistics at K arolinska Institutet. Paul's main research interest has been in developing methods for survival analysis within epidemiological studies, particularly modelling excess mortality (relative survival) models, time-dependent covariate effects, incorporating period analysis in statistical models, and the estimation and modelling of ‘cure' in population-based cancer studies. He is particularly keen on the use of flexible parametric survival models for both standard and relative survival. These offer a number of advantages in terms of communication of results, for example quantifying absolute levels of risk as well as relative risk. He has developed software in Stata to fit cure models for relative survival (strsmix and strsnmix) and also flexible parametric models (stpm2). Paul is coauthor of the book Flexible Parametric Survival Analysis Using Stata: Beyond the Cox Model.

Paul Lambert's homepage: http://pclambert.net/

Paul Dickman

Paul DickmanPaul Dickman is Professor of Biostatistics at the Department of Medical Epidemiology and Biostatistics at Karolinska Institutet. He conducts research in epidemiology and biostatistics with a focus on cancer epidemiology and register-based epidemiology. His primary interests lie in statistical methods for estimating and modelling net survival. He has published widely in the field of cancer patient survival and is coauthor of the Stata strs command for estimating and modelling relative survival. He has taught courses in survival analysis in Italy and England each summer for over 15 years.

Paul Dickman's homepage: http://www.pauldickman.com/

 

Register here