|
|
|
|
|
|
|
|
|
|
Abstract
Purposes
(1) Determine the
association of multiple cancers with smoking, focusing on cancers with
an uncertain association; and (2) illustrate quantitative bias analysis
as applied to registry data, to adjust for misclassification of smoking
and residual confounding by alcohol and obesity.
Methods
New Zealand 1981 and
1996 censuses, including smoking questions, were linked to cancer
registry data giving 14.8 million person-years of follow-up. Rate ratios
(RR) for current versus never smokers, adjusting for age, sex,
ethnicity and socioeconomic factors were calculated and then subjected
to quantitative bias analysis.
Results
RR estimates for
lung, larynx (including ear and nasosinus), and bladder cancers adjusted
for measured confounders and exposure misclassification were 9.28 (95 %
uncertainty interval 8.31–10.4), 6.14 (4.55–8.30), and 2.22
(1.94–2.55), respectively. Moderate associations were found for cervical
(1.82; 1.51–2.20), kidney (1.29; 1.07–1.56), liver cancer (1.75;
1.37–2.24; European only), esophageal (2.14; 1.73–2.65), oropharyngeal
(2.30; 1.94–2.72), pancreatic (1.68; 1.44–1.96), and stomach cancers
(1.42; 1.22–1.66). Protective associations were found for endometrial
(0.67; 0.56–0.79) and melanoma (0.72; 0.65–0.81), and borderline
association for thyroid (0.76; 0.58–1.00), colon (0.89; 0.81–0.98), and
CML (0.66; 0.44–0.99). Remaining cancers had near null associations.
Adjustment for residual confounding suggested little impact, except the
RRs for endometrial, kidney, and esophageal cancers were slightly
increased, and the oropharyngeal and liver (European/other) RRs were
decreased.
Conclusions
Our large study
confirms the strong association of smoking with many cancers and
strengthens the evidence for protective associations with thyroid cancer
and melanoma. With large data sets, considering and adjusting for
residual systematic error is as important as quantifying random error.
0 comments :
Post a Comment
Your comments?
Note: Only a member of this blog may post a comment.