Smoking can be considered as a "disease" with 4 states: current,
former, experimental and never. The transitions between these states can
be modelled as a Markov Process, or, under simplifying assumptions, as a
Markov Chain. Considering that smoking status is often mis-reported,
so that we observe not the real or "latent" chain, but rather an observed
chain, it is of interest to estimate the parameters of the latent
chain and to test different types of possible latent Markov Chains. This
methodology has been applied to two studies, the "School" smoking study at
UWO and the Waterloo Smoking Prevention Project at Waterloo; some
interesting differences between the two studies have been found (Mannan
and Koval, 2003). A different approach is to "attach" the transitions into
sequences, called trajectories, and to fit mixtures of these trajectories
in an attempt to find a minimal set of trajectories which best explain
the data. This has been used by Driezen(2001) with the Waterloo data.
About the Speaker
John Koval did both his undergraduate and Master's studies at the
University of Waterloo. He read for an M.Phil at Imperial College of the
University of London, and completed a PhD in the Department of Statistical
and Actuarial Sciences at the University of Western Ontario. He has been
in the Department of Mathematics, then the Department of Statistics and
Actiuarial Sciences at Western, before settling into the Department of
Epidemiology and Biostatistics. His research interests include the
distribution of correlated binary outcomes, logistic regression models,
epidemiological regression models, and state transitions models. He
currently holds grants from the Natural Sciences and Engineering Council
and the National Cancer Institute of Canada, and is a co-investigator on a
grant from the Canadian Institutes of Health Research, the latter two
involving smoking among adolescents and young adults, and treatment of
patients with HCV, HBV and HIV by health care providers, respectively.
References
Some references relevant to this talk are
Cameron, R., Brown, K.S., Best, J.A., Pelkman, C.L., Madill, C.L.,
Manske, R.R., and Payne, M.E. (1999). Effectieness of a social influences
smoking prevention program as a functionn of provider type, training
method, and school risk. American Journal of Public Health,
89, 1827-31.
Driezen, P. (2001). The development of youth smoking: clusters of
initiation and regular smoking trajectories. unpublished MSc. thesis,
University of Western Ontario.
Mannan, H.R. and Koval, J.J. (2003). Latent mixed Markov modelling of
smoking transitions using Monte Carlo bootstrapping. Statistical
Methods in Medical Research, 12, 125-46.
Pederson, L.L., Koval, J.J., McGrady, G.A. and Tyas, S.L. (1998). The
degree and type of relationship between psychosocial variables and smoking
status for students in Grade 8: Is there a dose-response relationship?
Preventive Medicine, 27 337-47.