Analysing Multiple Causes of Death: Which Methods For Which Data? An Application to the Cancer-Related Mortality in France and Italy Translated title: Analyse des causes multiples de décès: quelles méthodes pour quelles données? L’exemple de la mortalité par cancer en France et en Italie

This report summarises the epidemiological evidence on the association between tobacco smoking and cancer, which was reviewed by an international group of scientists convened by IARC. Studies published since the 1986 IARC Monograph on "Tobacco smoking" provide sufficient evidence to establish a causal association between cigarette smoking and cancer of the nasal cavities and paranasal sinuses, nasopharynx, stomach, liver, kidney (renal cell carcinoma) and uterine cervix, and for adenocarcinoma of the oesophagus and myeloid leukaemia. These sites add to the previously established list of cancers causally associated with cigarette smoking, namely cancer of the lung, oral cavity, pharynx, larynx, oesophagus, pancreas, urinary bladder and renal pelvis. Other forms of tobacco smoking, such as cigars, pipes and bidis, also increase risk for cancer, including cancer of the lung and parts of the upper aerodigestive tract. A meta-analysis of over 50 studies on involuntary smoking among never smokers showed a consistent and statistically significant association between exposure to environmental tobacco smoke and lung cancer risk. Smoking is currently responsible for a third of all cancer deaths in many Western countries. It has been estimated that every other smoker will be killed by tobacco.

One measure of the accuracy of cancer mortality statistics is the concordance between cancer defined as the underlying cause of death from death certificates and cancer diagnoses recorded in central, population-based cancer registries. Previous studies of such concordance are outdated. To characterize the accuracy of cancer mortality statistics from the concordance between cancer cause of death and primary cancer site at diagnosis. Central cancer registry records from California, Colorado, and Idaho in the U.S. were linked with state vital statistics data and evaluated by demographic and tumor information across 79 site categories. A retrospective arm (confirmation rate per 100 deaths) compared death certificate data from 2002 to 2004 with cancer registry diagnoses from 1993 to 2004, while a prospective arm (detection rate per 100 deaths) compared cancer registry diagnoses from 1993 to 1995 with death certificate data from 1993 to 2004 by International Statistical Classification of Diseases and Related Health Problems (ICD) version used to code deaths. With n=265,863 deaths where cancer was recorded as the underlying cause based on the death certificate, the overall confirmation rate for ICD-10 was 82.8% (95% confidence interval [CI], 82.6-83.0%), the overall detection rate for ICD-10 was 81.0% (95% CI, 80.4-81.6%), and the overall detection rate for ICD-9 was 85.0% (95% CI, 84.8-85.2%). These rates varied across primary sites, where some rates were <50%, some were 95% or greater, and notable differences between confirmation and detection rates were observed. Important unique information on the quality of cancer mortality data obtained from death certificates is provided. In addition, information is provided for future studies of the concordance of primary cancer site between population-based cancer registry data and data from death certificates, particularly underlying causes of death coded in ICD-10. Copyright © 2010 Elsevier Ltd. All rights reserved.