Abstracts

Camille Maringe, London School of Hygiene and Tropical Medicine

Measuring explained variation in excess hazard models

The availability of longstanding collection of detailed cancer patient information makes multivariable modelling of cancer‐specific hazard of death appealing. In such models, measures of explained variations are rarely reported and discussed.

We propose to report the variation in survival explained by each variable that constitutes a multivariable survival model. We describe how we adapted the ranks explained (RE) measure to the relative survival data setting, i.e. when competing risks of death are accounted for through life tables from the general population. RE is calculated at each event time and we introduce weights for each death reflecting its probability to be a cancer death.

The proportion of variation in survival explained by key prognostic factors is a crucial information toward understanding the mechanisms underpinning cancer survival. Furthermore, time‐varying RE provides insights into patterns of influence for strong predictors. We will illustrate these benefits when modelling cancer survival using population-based cancer registry data.

Francesca Gasperoni, MRC Biostatistics Unit, University of Cambridge

Clustering structure identification in clinical administrative databases through a Semi-Markov multi-state model with a nonparametric discrete frailty

Clinical administrative databases are often characterised by a hierarchical structure (i.e., patients grouped in hospitals) and a longitudinal structure (i.e., repeated hospitalizations). Very few articles in the biostatistical literature dealt with both challenges simultaneously and usually strong assumptions were done. In order to provide a flexible model that can properly take into account the heterogeneity associated with the grouping structure and the complete healthcare path of the recorded patients, we propose a semi-Markov multi-state model, where the transition-specific hazards are modelled through a Cox proportional hazards model with a nonparametric discrete frailty term. The nonparametric frailty is shared among patients hospitalised in the same provider. This novel methodology allows us to detect latent populations of providers (i.e., clusters of providers) for each specific transition on the basis of patients' characteristics, considering the whole healthcare patients clinical history. The proposed method is illustrated through an application to Heart Failure patients recorded in an administrative database from Lombardia, a northern region in Italy.

Paul Lambert, University of Leicester and Karolinska Institutet, Stockholm

Standardised and reference adjusted all-cause and crude probabilities in the relative survival framework

In population-based cancer survival studies the most common measure to compare population groups is age standardized marginal relative survival, which under assumptions can be interpreted as marginal net survival; the probability of surviving if it was not possible to die of causes other than the cancer under study (if the age distribution was that of a common reference population). The hypothetical nature of this definition has led to confusion and incorrect interpretation.

When comparing cancer survival between different population groups then differences between population groups should depend only on differences in excess mortality rates due to the cancer and not due to differences in other cause mortality rates or differences in the age distribution (or other demographics factors).

I will describe using crude probabilities of death and all-cause survival that incorporate reference expected mortality rates. This makes it possible to obtain marginal crude probabilities and all-cause probabilities of death that only differ between population groups due to excess mortality rate differences. These lead to fair comparisons, but are somewhat easier to understand and communicate than the usual measures.

I will discuss how choices have to be made regarding what reference mortality rates to use and what age distribution to standardize to. I will focus on estimation in a modelling framework, but also discuss some ongoing work on non-parametric estimation.  

Membership

Existing members can login below to view all site content. Lost password?

 
 

Other visitors might be interested to learn more about the benefits of membership.

Other events

28 Oct 20Advances in statistical genomics
02 Oct 19New perspectives on studying the effects of treatment on a time to event outcome
10 Jul 19 - 12 Jul 197th Channel Network Conference
03 Jun 19 - 05 Jun 19Workshop on Geostatistical Methods for Disease Mapping
28 Nov 18Advances in Analysis of Count Data (and AGM)

© 2009-2020 Biometric Society, British and Irish Region | Admin | Read our cookie and privacy policy