Environmental Engineering Reference
In-Depth Information
of community area under fruit cultivation (another potential source of pesticides
exposure) was included in the analyses as dichotomous confounder (<5% vs. ≥5%
of community area).
All analyses were preformed in SAS Institute [23], stratifi ed by gender and cancer
type. The regression analysis includes cancer rate as dependent variable, and age,
wine growing area, rural/urban setting, and fruit cultivation. All analyses were strati-
fi ed by gender and diagnosis. The results of our initial Poisson regression indicated a
possible problem with overdispersion, which is partly due to heterogeneity between
communities with respect to unobserved risk factors. We therefore opted to assume
a negative binomial distribution for the dependent variable, which allows to estimate
a dispersion parameter
k
for the variance (variance = expected value·(1 +
k
·expected
value)) and includes the Poisson distribution as a special case (
k
= 0). The nega-
tive binomial distribution emerges naturally if expected counts (Poisson parameters)
vary among communities according to a gamma distribution. The interpretation of
RRs stays the same as for Poisson regression. However, results do not substantially
differ. For a few rare cancers, the ML fi tting algorithm did not converge using the
negative binomial distribution. In these cases, estimates from Poisson regression are
reported.
Side Analysis of Standardized Incidence Ratios (SIRs) Using German
Incidence Rates as Reference
Even in communities with a small area under cultivation, cancer incidence might be
elevated, potentially leading to an underestimation of RRs in communities with me-
dium or large area under cultivation. In an additional analysis, we therefore calculated
SIRs regardless of the incompleteness of the Rhineland-Palatinate cancer registry. The
SIRs were separately computed for winegrowing communities with small, medium,
and large area under cultivation using estimated German incidence rates. The expected
numbers of cancer (E) for the time period 2000-2003 were compared with the ob-
served numbers (O), calculating SIR as the ratio between the observed and expected
numbers. Exact 95%-confidence intervals (CIs) based on the Poisson distribution of
O were calculated.
Results of any analysis based on small numbers are diffi cult to interpret. There-
fore, only those results based on at least 10 cases in the respective referent group and
10 cases in both comparison groups combined are reported here.
Tables 2 and 3 present incidence RR for cancer in males and females for wine-
growing communities with medium (>5 to ≤20%) and large (>20%) area under cul-
tivation compared to communities with small (>0 to ≤5%) area under cultivation.
Signifi cantly increased RR are observed for non-melanoma skin cancer (C44 ICD-
10) among men (RR = 1.32 (95% CI 1.20-1.45) for medium and RR = 1.39 (95% CI
1.25-1.54) for a large area under cultivation) as well as among women (RR = 1.40
(95% CI 1.27-1.54) for medium and RR = 1.38 (95% CI 1.23-1.53) for a large area
under cultivation).
Search WWH ::
Custom Search