|Current Workshop | Overview||Back | Home | Search ||
Multistate Models: Bridging Biostatistics, Demography and Econometrics
Description and Aim
Multistate models describe the life course in terms of transitions individuals experience as they go through stages of life. Depending on the specific application, these states may represent health states, stages of disease, occupation or education. Multistate models have been successfully used in a wide variety of applied sciences, but it is probably fair to say that the most fruitful areas of application are medicine, demography and economics. Important examples of applications of multistate models are stem cell transplantation (with disease relapse and death as endpoints and graft-versus-host disease as intermediate states) in medicine, modeling of life course events like marriage or migration in demography, and participation in the labor market in economics. The majority of multistate models considered are at the individual (micro) level; in demography these are also used to obtain projections at the population (macro) level.
Motivated by different contexts and by different research questions, methodology for multistate models in these three major disciplines has developed fairly separately, with less than ideal involvement or partnership across fields. This may have been due to “cultural” differences, differences in terminology and notation which tends to make papers from other fields hard to read, and perhaps a lack of awareness of the similarities. Yet the fact is that the similarities in terms of the questions asked, the problems to be solved, and in the end often also the methodology used, are far greater than the distinctions. To enhance progress, there is a need to blur disciplinary boundaries.
This workshop brought together experts in multistate models from these three disciplines, with three aims in mind. The first aim was to increase awareness of both well established and state of the art methodology in these three disciplines, and to exchange ideas. The second aim was to take away barriers, broaden views and to promote collaboration between scientists from these fields. The third aim was to contrast and compare the relative merits of approaches from biostatistics, demography and econometrics in answering the same research question based on the same data.
Roughly speaking, the workshop consisted of three parts, running in parallel. For the first part, three special topic sessions were organized around the following topics of special interest: “Expected length of stay”, “Goodness of fit and non-Markov models”, “Heterogeneity”. These sessions were very helpful, because each of the three disciplines presented a review of the state of the art. In-depth discussions followed, usually extending over coffee breaks and/or lunch. The second part consisted of research talks. Usually, these talks also covered one or more of the special topics, so there were useful cross-references between these talks and the special topics, as well as a lot of discussion. The last part was the use of data from the Survey of Health, Ageing and Retirement in Europe (SHARE) project (http://www.share-project.org/) for the purpose of addressing the research question: “what is the impact of health status at age 50 on age at retirement and what are the associated costs?” The research question had been chosen in such a way that each of the three special topics plays a role. While this last part was fruitful from a methodological point of view because differences between approaches with respect to underlying assumptions were highlighted (eg, whether or not to account for interval censoring), it was generally felt that the quality of the SHARE data was not sufficient to justify drawing substantive conclusions
Summary and conclusion
Overall, the workshop was very effective in contrasting (or, often in pointing out similarities between) statistical approaches to multistate models in the three disciplines biostatistics, demography and econometrics. Have we managed to bridge the three disciplines? The gap between biostatistics and demography became smaller. This is demonstrated in the interest among demographers in methods biostatisticians use for estimating multistate models (non-parametric method versus piecewise constant model) and the interest among biostatisticians in estimating state occupation times, which is a core problem in demography (e.g. life expectancy). The gap between econometrics and the other two disciplines could not be reduced as much as desired. Using SHARE data was a unique experiment that is worth repeating.