The CUSUM method is frequently used by epidemiologists for detecting a shif
t in the incidence of rare health events. An exact solution for a Poisson p
rocess has been proposed by other works, but potential applications have be
en limited by the fact that the size, the structure of the at-risk populati
on and the baseline rate may not be constant during the period of surveilla
nce. Furthermore, for practical use tables of critical values are available
only for some expected values and in any case less then 9. This paper prop
oses an approximate CUSUM procedure, based on the normal approximation to a
Poisson process, which may be an efficient solution of the problems previo
usly pointed out. Analyses of simulated and actual data sets illustrate the
usefulness of the proposed procedure. An application to mortality data for
respiratory diseases in a north Tuscany area, characterized by the presenc
e of chemical plants, is showed. No shift in the mortality rate during the
1980-1989 period was detected compared with the 1970-1979 period, in contra
st with the result obtained with the standard CUSUM method for a Poisson va
riate. Copyright (C) 1999 John Wiley & Sons, Ltd.