Knowledge of the spatial distribution of diseases provides useful info
rmation in etiologic research and in implementation of preventive acti
vities in community health. Spatial autocorrelation analysis is one of
the Various methods that enables to determine spatial clustering of d
iseases. This method has not been applied to lung, stomach and colon c
ancer in Quebec. These cancers are frequent and are associated with en
vironmental factors. The objectives of the study are to determine spat
ial distribution of incidence rates of these cancer sites by sex and t
o help generate etiologic hypotheses. Community health departments (CH
Ds) of residence are considered as risk markers since their population
and environment may be related to the selected cancer sites. Data wer
e obtained from Quebec Cancer Registry. Rates were standardized by the
direct method. Autocorrelation analysis was done through BW coefficie
nt and Moran's coefficient I for correlograms. Results of standardized
rates were compared to those of non standardized rates. Rates yielded
the same results for the BW coefficient. Conversely, results were qui
te different for the correlograms. This implies that results from stan
dardized rates should be kept since the age structure of CHD populatio
ns are different. Important variation in the level of spatial autocorr
elation was found among the six sex-specific cancer sites. For male lu
ng cancer and male stomach cancer first-order neighbouring CHDs showed
similar incidence rates according to a geographic gradient. Female lu
ng cancer exhibited spatial autocorrelation. Absence of spatial autoco
rrelation for colon cancer suggests that CHD is not the appropriate sc
ale for study of this cancer and allows use of conventional epidemiolo
gic methods. These results are discussed in relation to current etiolo
gic hypotheses. Lung and stomach cancers deserve further study.