scientific article; zbMATH DE number 1550322

Authors

Publication date 10 January 2001

discrete S-I-R Markov model Gibbs sampler Kalman filter

Applications of Markov chains and discrete-time Markov processes on general state spaces (social mobility, learning theory, industrial processes, etc.) (60J20) Inference from stochastic processes and prediction (62M20) Research exposition (monographs, survey articles) pertaining to biology (92-02) Medical epidemiology (92C60) Epidemiology (92D30)

Review text

The author and his co-workers have devoted considerable time and effort to the modelling of many aspects of HIV transmission, and this book summarizes their current state of thinking. As in any such venture, the huge wealth of biological and sociological details involved necessitates some judicious simplification, and the way in which this is accomplished largely determines the success of the project. The compromise struck here is a nice one; the underlying structure of the nonlinear transmission process is retained, and the formulation is stochastic, so that parameters can be estimated and the precision of the estimates properly assessed. The epidemic is described using a discrete S-I-R Markov model, distinguishing many stages of infectivity, with innovations that are multinomially distributed, conditionally on the current state variables. A Kalman filter procedure combined with a Gibbs sampler is used for estimation and prediction. Mixing within the population is modelled in terms of a few groups with different patterns of behaviour, coupled by preferred mixing; thus the details of the underlying relationship graphs are largely but not entirely ignored. It is to be expected that the approach will yield reliable results of considerable practical use. The book unfortunately has a number of weaknesses. References are collected by chapter, and are thus often duplicated; furthermore, they are not alphabetically ordered. The result is very irritating. The book is also beset by niggling errors (it begins with Errata), many of which are merely typographical. However, on closer inspection, some are more than that -- for example, the misapprehension about the correlations between the state of a process and its own innovations, given in (a), p. 87, and repeated in (4), p. 242. Indeed, the independence of the innovations said to be necessary in (1), p. 242, is not satisfied by the processes under consideration, an approximation being made (but not acknowledged) hereabouts. More practically, I could not find the parameter values assumed for inter-group mixing in the San Francisco homosexual HIV example, the under-reporting errors discussed on p. 243 are likely to show strong negative lag 1 correlations, and I have little confidence in the estimates of the incubation time distribution tabulated on p. 251, at least for periods exceeding 15 years or so. This is a great pity. As already mentioned, the approach seems to offer a well-judged blend of theory and practice, and I expect it (when tidied up) to deliver consistently valuable results.

Recommendations

Cited in

(15)

This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4523447)