A pooled analysis of case-control studies of thyroid cancer - I. Methods

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is not well described, we conducted a pooled analysis of all published case-control studies, as well as two identified unpublished studies.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 char acteristics of the 14 studies included in the analysis, as well as the stat istical methods employed. Four studies were conducted in the United States (1 each in Washington State, California, Connecticut and Hawaii), 8 in Euro pe (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 menstr ual 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 con trols (3,699 females and 1,077 males) were included in this study. Of the c ases, 79% were classified as papillary thyroid carcinomas, 14% as follicula r, 2% medullary, 1% anaplastic, 1% other histologies, and 3% histological t ype unknown. Each of the datasets was checked for outliers and consistency. Data were analysed separately by study center, gender, and the two major h istologic types (papillary, follicular). Frequency tables and simple statis tics were computed for each variable under study. Conditional logistic regr ession was used to compute odds ratios. For matched studies, the original m atching was preserved, whereas, for unmatched ones, five-year age groups we re used for matching. Study-specific analyses were computed, and then the d ata from all the studies were pooled conditioning on study. Heterogeneity b etween studies, geographic areas and study designs was assessed, and the mo difying effect of age was also evaluated.