A pooled analysis of case-control studies of thyroid cancer¶ I. Methods
Eva Negri,Elaine Ron,Silvia Franceschi,Luigino Dal Maso,Steven D. Mark,Susan Preston-Martin,Anne McTiernan,Laurence Kolonel,Ruth Kleinerman,Charles Land,Fan Jin,Gun Wingren,Maria Rosaria Galanti,Arne Hallquist,Eystein Glattre,Eiliv Lund,Fabio Levi,Dimitrios Linos,Claudia Braga,Carlo La Vecchia
DOI: https://doi.org/10.1023/A:1008851613024
1999-01-01
Cancer Causes & Control
Abstract:Objective: Because the etiology of thyroid cancer is not well described, we conducted a pooled analysis of all published case- control studies, as well as two identified unpublished studies. This paper describes the major characteristics of the 14 studies included in the analysis, as well as the statistical methods employed. Four studies were conducted in the United States (1 each in Washington State, California, Connecticut and Hawaii), 8 in Europe (3 in Sweden, 2 in Norway, 1 in Switzerland, 1 in Italy and 1 in Greece), and 2 in Asia (1 in China and 1 in Japan). Methods: The original datasets were obtained and restructured in a uniform format. Data on socio-demographic characteristics, anthropometric measures, smoking and alcohol consumption, history of benign thyroid diseases and of other selected medical conditions and treatments, family history of cancer and of benign thyroid conditions, occupation, residence in endemic goitre areas, and dietary habits were analyzed. For women, we also analyzed menstrual and reproductive factors and use of female hormones. Radiotherapy and, in Japan, exposure to the A-bombs were considered as potential confounding factors. Results: A total of 2,725 cases (2,247 females and 478 males) and 4,776 controls (3,699 females and 1,077 males) were included in this study. Of the cases, 79% were classified as papillary thyroid carcinomas, 14% as follicular, 2% medullary, 1% anaplastic, 1% other histologies, and 3% histological type unknown. Each of the datasets was checked for outliers and consistency. Data were analysed separately by study center, gender, and the two major histologic types (papillary, follicular). Frequency tables and simple statistics were computed for each variable under study. Conditional logistic regression was used to compute odds ratios. For matched studies, the original matching was preserved, whereas, for unmatched ones, five-year age groups were used for matching. Study-specific analyses were computed, and then the data from all the studies were pooled conditioning on study. Heterogeneity between studies, geographic areas and study designs was assessed, and the modifying effect of age was also evaluated.