|
Background
Event history analysis denotes a set of statistical methods used to analyse and describe life times, durations and more complex event history data. There have been made great advances in Event history analysis over the last three decades. Nevertheless the field still remains dominated by the classical methods for single event times (Kaplan-Meier estimator, logrank test, Cox regression), and existing methodologies are not always easily adapted to the more ambitious research questions and richer data structures of contemporary research:
- Biomedical event history data are becoming ever more complex and often involve multiple transitions between health states, or the occurrence of recurrent events, in parallel with time-dependent partially observed stochastic marker processes for disease progression. A common approach is to subdivide into sequential event times which are separately analysed using standard survival analysis methodology. Methods for a joint analysis of such data are emerging, but are still in their infancy.
- In recent years there has been an increasing interest in the ambitious task of deducing causality from statistical data. Different approaches have been proposed: graphical models, predictive causality, and counterfactual causality. As causality is based on a notion of the past influencing the present and the future, event history modeling should play a more central role in the causality literature than the case is today.
- The influence of genetic factors on morbidity and mortality has been studied in recent years using multiplicative frailty models for familial association. The relatively simple shared frailty models have been extended to models with shared and separate frailty components. However, there can now be detailed genetic information available at the individual level, and this has not properly been built into the existing models for association in event histories.
- Cohort sampling methods like the nested case-control and case-cohort designs have made it possible to reduce costs of covariate collection and checking in studies of large epidemiological cohorts. But methodological developments are still needed, e.g., for sampling of clusters (e.g. families) and for the integrated analysis of the detailed data on a case-control sample and the more rudimentary data on the remaining cohort.
- Modern high-throughput technologies produce data of an extraordinary high dimension, for example in microarray experiments and genome-wide SNP genotyping. When such genomic data are used as predictors for survival – alone or in conjunction with more traditional risk factors – both the high dimension of the covariate space and the measurement errors in the genomic data need to be taken into account. Additional complications may be due to missing data and/or replicated measurements of gene expressions.
Aims
These examples clearly illustrate the need for further methodological developments in Event history analysis. A main aim of the workshop was to gather key international researchers in Event history analysis and related fields in order to get an overview of the state of the art concerning methodologies for analyzing complex event history data, and to identify and discuss important application areas and research directions.
Organization
The workshop was part of the project "Statistical analysis of complex event history data" at the Centre for Advanced Study, and it was organized in collaboration with the NOREVENT working group.
|
