This pooled analysis comprises data from seven case–control studies from the United States (Table ). These studies represent a subset of studies included in the ILCCO collaboration (http://ilcco.iarc.fr
). The remaining studies participating in the ILCCO consortium either excluded by design rare histological types of lung cancer or enrolled less than 10 BAC cases. The seven studies included one study each from the University of California at Los Angeles (UCLA) [16
], Harvard University (HU) [17
], University of Hawaii (UH) [18
], two studies from Wayne State University (WSU) [19
], one study from Mayo Clinic (Mayo) [22
], and one multicentric study coordinated by the American Health Foundation (AHF) [23
]. Results on tobacco smoking and BAC risk have been reported for part of the AHF study [13
]. For some of the studies, the number of subjects included in the pooled analysis was larger than that reported in previous publication because recruitment has continued. Three studies [17
] was hospital-based, and the other four were community-based case–control studies. In the UCLA study, newly diagnosed cases were identified using the rapid ascertainment system of the Cancer Surveillance Program for Los Angeles County. All BAC cases were diagnosed pathologically and validated through pathologic records. A similar approach was followed in the WSU studies, based on the Metropolitan Detroit Cancer Surveillance System. In the HU study, cases were enrolled from the Massachusetts General Hospital in Boston, MA, as part of an ongoing case–control molecular epidemiology study of lung cancer. Pathology reports generated by two pulmonary pathologists as part of routine tissue diagnosis were obtained from the clinical information system. The histological confirmed diagnosis of BAC from pathological reports was reviewed by a study pathologist or oncologist. In the UH study, pathology records were obtained from SEER registries; BAC diagnoses were not validated. In the Mayo study, the histological diagnoses were validated. In the AHF study, information on histology was abstracted from the records of the pathology departments of the participating hospitals and was not validated. The proportion of BAC over total lung cancers varied from 3 to 9%. Information on stage of BAC cases was not available.
Characteristics of studies included in the pooled analysis of lung bronchioloalveolar carcinoma and tobacco smoking
All studies collected information on lifetime history of tobacco smoking, including age of start smoking, duration, intensity, and time since quitting for the former smokers. All studies collected information on cigarette smoking; in the UCLA, Harvard, and Hawaii studies, cigar and pipe smoking was added to cigarette smoking to calculate cumulative tobacco consumption. For the purposes of the pooled analysis, we generated common variables related to smoking status (never, ever, current, and former smoker, defined as smokers who have quit more than 1 year before diagnosis or interview), daily amount of smoking (1–9, 10–19, 20–29, 30–39, and 40 or more cigarettes/day), duration of smoking (1–9, 10–19, 20–29, 30–39, and 40 or more years), and cumulative smoking (1–9, 10–19, 20–29, 30–39, and 40 or more pack years, one pack year being equivalent to 20 cigarettes/day smoked during 1 year). Former smokers were further categorized according to time since quitting (1–9 years, 10–19, 20 or more years).
Other variables included in the pooled analysis were sex, age at diagnosis or interview (categorized as up to 50, 51–60, 61–70, and more than 70), education (basic or elementary, up to high school, and college or more), study center, and race (Asian, Black, White, Hawaiian, Hispanic, other).
All analyses were performed using the STATA statistical package [25
]. Study-specific OR of lung cancer and their 95% CI for smoking intensity, duration, cumulative smoking, and time since quitting were estimated using unconditional logistic regression. All the estimates were adjusted for age at diagnosis, sex, race and study center.
We explored the heterogeneity between study–specific results using the Q statistic at the significance level of p
< 0.05. Galbraith plot where the ratio of the log odds ratio to the standard error (the Z statistic) for each study is plotted against the reciprocal of the standard error [26
]. Further, the influence of each study on the overall meta-analysis estimate was analyzed by an influence analysis, where the meta-analysis estimates are computed after omitting each study. Pooled OR estimates stratified by sex, age at diagnosis, education level, and study design were also calculated.
Linear trends for the amount, duration, cumulative smoking, and time since quitting were tested according to Greenland and Longnecker [27
], using the ‘glst’
command on STATA, which utilizes the generalized least squares for trend estimation of summarized dose–response data. Population attributable fractions were calculated using ‘aflogit
’ on STATA, which estimates the adjusted measures of population attributable fraction from a logistic regression model [28