In a Chinese cohort of 29,584 adults (participants to the Linxian General Population Nutrition Intervention Trial) followed from 1986 to 2001 and including 1,440 cases of gastric cancer, 1,949 of esophageal cancer and 425 subjects (59 among cases) who reported history of goiter, Abnet et al.1 found goiter associated with an increased risk of gastric noncardia adenocarcinoma. The hazard ratio (HR) was 2.04 (95% confidence interval, CI, 1.01–4.11). The HR were 1.45 (95% CI, 0.91–2.30) for gastric cardia adenocarcinoma and 1.37 (95% CI, 0.97–1.94) for esophageal squamous cell carcinoma. Iodine is concentrated by the gastric mucosa, where it may act as an antioxidant, thus, iodine deficiency, and hence goiter, may be associated with an increased risk of gastric cancer.2 Only a few studies investigated this issue. A study from Turkey3 reported that 49% of 61 gastric cancer cases versus 20% of 55 controls had history of goiter, and cases also reported more frequently thyroid autoimmune disease than controls. Another Turkish study of 19 cases found a lower iodine concentration in gastric cancer tissue than in surrounding normal tissue.4 An Iranian study5 showed that gastric cancer patients excreted more iodine than controls. To investigate the relation between thyroid diseases and gastric cancer risk in a developed country, we analyzed data from a hospital-based case-control study conducted between 1985 and 1997 in the Greater Milan area.6 Briefly, this included 769 incident, histologically confirmed cases of stomach cancer (469 men and 300 women; median age 61 years, range 19–79 years). The controls were 2,081 subjects (1,220 men, 861 women, median age 55 years, range 19–79 years) admitted to the same hospital network as cases for acute conditions other than neoplastic or digestive tract diseases and unrelated to long-term diet modification (46% traumas, 20% other orthopaedic conditions, 19% acute surgical, 15% other miscellaneous disorders). Less than 4% of identified cases and controls refused to be interviewed. Cases and controls were interviewed during their hospital stay using a standard questionnaire, including information on sociodemographic characteristics, anthropometric measures and information on smoking, alcohol, frequency of consumption of 29 food items and a problem-oriented medical history, including 16 diseases or conditions. Among these, there were thyroid diseases, which were subsequently classified as goiter, nodules/adenoma, hyperthyroidism and other or unidentified benign thyroid diseases. Odds ratios (OR) and the corresponding 95% CI were estimated using unconditional multiple logistic regression models, including terms for age, sex and education. Table I gives the distribution of cases and controls according to various thyroid conditions. The frequency of all of these ranged between 0.4% and 1.2%. No relation was found between history of various benign thyroid diseases and gastric cancer risk. The OR was 0.70 (95% CI, 0.19–2.65) for history of goiter, 0.61 (95% CI, 0.23–1.64) for thyroid nodules/adenomas, 1.10 (95% CI, 0.47–2.57) for hyperthyroidism, and 0.71 (95% CI, 0.32–1.57) for other or not specified benign thyroid diseases. In this study, we were not able to distinguish between various subsites of gastric cancer, but the incidence of gastric cardia cancers between 1985 and 1992 in Italy was only around 10% of total gastric cancers.7 Thus, our results refer mainly to noncardia gastric cancers. Among the advantages of the study, there is the relatively large dataset, the practically complete participation of cases and controls approached, and the reliability of information on medical history (r = 0.85 for thyroid conditions8). The study power, however, is limited by the small proportion of cases and controls with history of goiter and thyroid conditions. At the time of study conduction, gastric cancer was common in Italy, and was still the third cause of cancer death after lung and colorectal cancer.9 Furthermore iodized salt is not routinely used by the Italian population and mild iodine deficiency is not rare in the country.10 Still, goiter and other thyroid diseases do not appear to be a relevant cause of gastric cancer in Italy. Yours sincerely, The authors thank Ms. I Garimoldi for editorial assistance. Marta Rossi, Eva Negri, Roberto Foschi, Silvia Franceschi, Carlo La Vecchia.

We thank Dr. Grant for his interest in our work.1 We cannot make inference on latitude of exposure, specifically, on latitudes south of about 40°. The subjects diagnosed with skin cancer are most likely, than those without skin cancer, to have been heavily exposed to sun in this Swiss population. Exposure may have occurred in the country with mountainous areas, where sun irradiation is stronger than at sea level,2, 3 or abroad. With reference to smoking, according to the Monograph 83 of the International Agency for Research on Cancer,4 there is evidence suggesting the lack of carcinogenicity of tobacco smoking for female breast cancer. There is clear evidence for the risk of neither prostate cancer nor skin cancer. There is some evidence from prospective cohort and case-control studies that the risk of colorectal cancer is increased among tobacco smokers. However, it is not possible to conclude that the association between tobacco smoking and colorectal cancer is causal (abstracted from reference4, pp. 1183–86). It is, therefore, unlikely that smoking has introduced any major bias, confounding, or modifying effect in the results of our study. The evidence from population-based cohort studies like ours1, 5-7 is open to criticism, but is in any case more valid than that of correlation (ecologic) ones often quoted by Dr. Grant in support of his views. Dr. Soerjomataram and Dr. de Vries suggest to provide additional information to address the issue of surveillance bias and stratified or sub-group analysis in our study.1 Although we caution toward inference based on subgroups, we are glad to provide the information requested, when available. With reference to potential surveillance bias, the number of male skin cancer cases in Vaud and Neuchatel, Switzerland were 15,246 when compared with 13,541 in the Dutch study, and that of population years at risk was 109,176. There is no reason to suppose that inclusion—rather than exclusion—of early prostate cancer introduces selection bias. In the mid 1990's, about 10% of men aged 65 or over may have undergone prostate-specific antigen (PSA) testing in Vaud and Neuchatel,8 and this proportion has risen over more recent years, in the absence, however, of any organised screening programme. The standardised incidence ratio (SIR) of prostate cancer was 1.18 (95% confidence interval (CI): 0.93–1.47) in the first, 1.24 (95% CI: 0.97–1.57) in the second, 1.16 (95% CI: 0.88–1.49) in the third year after diagnosis of skin cancer, and 1.13 (95% CI: 1.02–1.25) in subsequent years. Corresponding values were 1.34, 1.06, 0.89, 1.08 for colorectal, and 1.07, 1.20, 1.09, 1.06 for breast cancer. This weighs against the hypothesis of an early reduction in risk, which levels off with time, due to potentially reduced sunshine exposure after skin cancer diagnosis. With reference to strata of sex, the SIR of colorectal cancer were 1.06 (95% CI: 0.94–1.19) in men, and 1.14 (95% CI: 1.00–1.30) in women. The estimates were not significantly heterogeneous across sexes. We have information on stage and grade of prostate cancer only for recent years; thus numbers of subsequent cancers are inadequate now for any inference. Thus, none of the additional or subgroup analyses indicated by Dr. Soerjomataram and Dr. de Vries modify our conclusion that incidence of colorectal, breast, and prostate cancers is not reduced in patients diagnosed with skin cancer. We agree with Dr. Soerjomataram and Dr. de Vries that collaborative studies would help clarify the issue, and provide acceptable statistical power for subgroup analyses. Dr. Robert Tarone, from the International Epidemiology Institute, Rockville, MD, USA, kindly pointed to us that the expected rates in Table I were smaller than those in Table II.1 The correct Table I is, therefore, enclosed below. The first paragraph of the results section consequently reads as follows: “Table 1 gives the numbers of observed and expected cancers of the prostate, breast, and colorectum following basal and squamous cell skin cancers, skin melanomas, and all skin cancers combined. Overall, 680 prostate cancers were observed vs. 593.4 expected (SIR = 1.15; 95% CI: 1.06–1.24), 440 breast cancers were observed vs. 402.3 expected (SIR = 1.09; 95% CI: 0.99–1.20), and 535 colorectal cancers were observed vs. 488.0 expected (SIR = 1.10; 95% CI: 1.01–1.19). We, also, considered the risk of prostate, breast, and colorectal cancers in separate strata of age at diagnosis, location (head and neck vs. other), and time since skin cancer diagnosis, but found no indication of reduced risk in any of the strata considered, most SIRs for prostate, breast, and colorectal cancers being around or slightly above unity. The SIRs of colorectal cancer were 1.06 (95% CI: 0.94–1.19) for men, based on 292 observed vs. 275.4 expected, and 1.14 (95% CI: 1.00–1.30) for women, based on 243 vs. 212.6 expected.” These changes do not modify the inference and conclusions of our work. We apologize, nonetheless, the readers for the correction, and again thank Dr. Tarone for his careful reading and valuable help. Yours sincerely, Fabio Levi, Lalao Randimbison, Van-Cong Te, Manuela Maspoli Conconi, Carlo La Vecchia

We investigated the risk of lymphomas, hemolymphopoietic (HLP) cancers (including lymphomas), and non-HLP cancers in first-degree relatives of non-Hodgkin's lymphoma (NHL) cases in an Italian case-control study on 225 patients (median age, 59 years) with a new diagnosis of NHL and 504 hospital controls (median age, 63 years), admitted for a wide spectrum of acute, nonneoplastic, nonimmune conditions. We estimated odds ratios (OR) adjusted for sex, age, family size, and other potential confounders. We also built the cohort of all first-degree relatives and computed age and sex adjusted hazard ratios (HR) using proportional hazard models. A history of lymphoma in first-degree relatives was reported by 5 NHL cases and 3 controls [OR, 3.2; 95% confidence interval (95% CI), 0.7-14.4] whereas 14 cases and 11 controls reported a family history of HLP cancers (OR, 3.0; 95% CI, 1.2-7.0). The HR of relatives of NHL cases, compared with relatives of controls, was 4.5 (95% CI, 1.1-18.8) for lymphomas, 3.5 (95% CI, 1.5-7.4) for HLP cancers, 1.6 (95% CI, 1.3-2.0) for all cancers, and 1.0 (95% CI, 0.9-1.1) for all causes of deaths. The HRs were higher for relatives of NHL cases diagnosed before the age of 50 years: 7.1 for HLP cancers, 2.0 for all cancers, and 1.6 for all deaths. A family history of cancer of the liver (OR, 2.1; 95% CI, 1.0-4.2), breast (OR, 2.0; 95% CI, 1.0-3.6), and kidney (OR, 4.6; 95% CI, 1.0-20.9) increased NHL risk. The OR was also elevated for all cancer sites (OR, 1.7 95% CI, 1.2-2.4) and the risk increased with the number of affected relatives also when HLP cancers were excluded.

Istituto di Ricerche Farmacologiche “Mario Negri”, via Eritrea 62, 20157 Milano and Istituto di Statistica Medica e Biometria, Università degli Studi di Milano, via Venezian 1, 20133 Milano, Italy Correspondence to Carlo La Vecchia Tel: +39 02 39014 527; e-mail: [email protected]

Sir, Bouchardy et al (2007) reported that the Geneva Cancer Registry records show a substantial rise in breast cancer incidence in women under the age of 40 years in the period 2002–2004, and called for confirmation of their finding in other population-based cancer registries. We therefore abstracted data from the cancer registries of two other French-speaking Swiss Cantons, Vaud and Neuchâtel (Census populations (December 2000) of about 620 300 and 165 700, respectively; Levi et al, 2003a, 2003b). Table 1 gives annual number of registered breast cancer cases in women aged 25–39 years and the corresponding age-standardised rates (on the European standard population) from the two registries over the period 1995–2005. Over the last 3 years, rates were around 33/100 000, and there was some rise over the 11-year calendar period considered (the per cent annual rise was 2.7 in Vaud, 1.6 in Neuchâtel, both not statistically significant), but there was no substantial nor sudden increase in most recent years. Table 1 Trends in numbers and age-adjusted (European standard population) incidence rates of breast cancer per 100 000 women aged 25–39 years at diagnosis in the Cantons of Vaud and Neuchâtel, and annual percent change over the period ... When older age groups were examined, some upward trends in breast cancer registrations were observed, particularly in young and middle-aged women in Vaud, but there was again no indication of a recent sudden rise. At ages 40–49 years, rates were 118.5 in 1995 and 184.9 in 2005, the rates were 307 and 338.2 at ages 50–69 years, 339.9 and 338.4 at ages 70–79 years, and 358.2 and 366.1 at ages 80 and above. Corresponding rates for Neuchâtel were 126.5 in 1995 and 186.4 in 2005 at age 40–49 years, 392.8 and 319.7 at age 50–69 years, 251.0 and 423.7 at age 70–79 years, and 374.4 and 422.5 at ages 80 years and above. These data from the Vaud and Neuchâtel Cancer Registries confirm therefore that breast cancer incidence registration has been rising over the last decade, particularly in young and middle-aged women. This rise is largely or totally due to increased detection. In fact, an organised mammography screening programme has been in operation in Vaud since 1999 for women aged 50–70 years (Bulliard et al, 2003), and the use of mammography has increased also in younger women. Breast cancer mortality has fallen by 30% in French-speaking Swiss women between 1990 and 2002 (Bulliard et al, 2006), further indicating that any appreciable real rise in incidence is unlikely. Breast cancer incidence below age 40 has been stable in the USA between 1999 and 2003 (Morbidity and Mortality Weekly Report, 2007). The present data do not confirm the existence of a substantial rise in breast cancer incidence in younger women in recent years. The upward trends in Geneva may therefore be due to the play of chance, since they were based only on 50 cases registered in 2003-2004, or the detection of prevalent cases due to increased use of mammography in young women in recent years.

The relationship between liver cirrhosis and hepatocellular carcinoma is recognized, but quantification of risk is still uncertain. Therefore, we analysed data from a case-control study conducted in Italy between 1984 and 1997 on 499 cases of incident, histologically confirmed hepatocellular carcinoma and 1,552 controls in hospital with acute, non-neoplastic disease. Overall, 87 (17.4%) cases vs 10 (0.6%) controls reported clinical history of liver cirrhosis. The corresponding odds ratio (OR) was 27.5 (95% confidence interval (CI), 14.3-15.2) after allowance for sociodemographic factors, and 16.2 (95% CI, 7.9-32.9) after simultaneous allowance for all identified confounding factors, including alcohol consumption and clinical history of hepatitis. The association was of similar magnitude for subjects whose cirrhosis was diagnosed < 55 years (OR = 14.8) or at age 55 or over (OR = 20.0), and the multivariate OR was 33.7 < 5 years after diagnosis of cirrhosis, 37.3 between 5 and 9 years, and 7.6 (95% to 2.7-21.3) > or = 10 years since diagnosis of cirrhosis. The association was stronger in males (OR = 23.4) than in females (OR = 5.9), similar in various age groups, and somewhat stronger in more educated subjects (OR = 53.7), with history of hepatitis (OR = 33.1), reporting heavy alcohol consumption (OR = 24.9) or high body mass index (OR = 58.1), although the interaction term was significant only for sex. In terms of population attributable risk, 17% of hepatocellular carcinomas in this population can be attributed to clinical history of liver cirrhosis.

Dietary deficiency of folate and other micronutrients involved in the one-carbon metabolism (i.e., vitamins B2, B6, B12, and methionine) have been related to several diseases, including cancers, but results on non-Hodgkin lymphoma (NHL) are controversial. A hospital-based case-control study was conducted in Italy, in 1999-2002. Cases were 190 incident, histologically confirmed NHL aged 18-84 years. Controls were 484 subjects admitted to hospitals for acute, non-neoplastic diseases supposed to be unrelated to alcohol consumption or to diet modification. Dietary habits, including alcohol drinking, were assessed by a validated food-frequency questionnaire. Nutrient intakes were computed using the Italian food composition database. Odds ratios (ORs) and corresponding 95% confidence intervals for tertiles of nutrients' intake were computed using the energy-adjusted residual models. No significant association emerged between NHL risk and intakes of folate (OR=0.9), vitamin B2 (OR=0.9), vitamin B6 (OR=0.8), and methionine (OR=0.7). However, a significant inverse association was observed for all the nutrients examined among abstainers and former drinkers, whereas no relations between one-carbon nutrients and NHL risk emerged among current alcohol drinkers. Our findings support the possibility of an antagonist effect of alcohol on the one-carbon metabolism in NHL etiology. However, the lack of an overall effect for one-carbon nutrients and the small sample size suggested caution in interpreting our results.

The definition of environment refers to external physical conditions that may affect human health (1), but the word is used in the medical literature with different connotations, both in English and in other languages. This generates confusion in the medical literature, when the role of environment is discussed in the context of disease etiology. In particular, research on the role of environment in human carcinogenesis suffers from this ambiguity. On the one hand, environment can encompass all non-genetic factors such as diet, lifestyle and infectious agents. In this broad sense, the environment is implicated in the causation of the majority of human cancers, as has been demonstrated since the 1960s (2). Such a broad meaning of the word environment is assumed when referring to gene–environment interactions . On the other hand, environmental factors can include only the (natural or man-made) agents and circumstances encountered by humans in their daily life, upon which they have no or limited personal control. In this sense, environmental factors are restricted to air, water, soil and food pollutants, including physical pollutants such as sources of ionizing radiation. These ambiguities in the terminology and the inconsistencies in the use of the vocabulary by cancer researchers contribute to public confusion regarding the role of the environmental causes of cancer. A distinction relevant to cancer prevention may be made between factors related to personal behaviors, lifestyle (e.g. tobacco smoking and alcohol drinking), involuntary exposures, such as those linked to air, water, soil or food pollutants, and occupation. It would be preferable to abandon the term environment and to use terms such as nongenetic or modifiable determinants of disease (broad sense of environment) and pollutants (narrow sense).