To the Editor: Hemminki et al., 1 on the basis of the uniquely large Swedish Family-Cancer Database, reported an excess of tobacco-related neoplasms, invasive cervix, other female genital tract, anal and bladder cancers in women diagnosed with in situ and invasive cervical cancer. To provide further information on the issue, we up-dated to 1998 our report of incidence of invasive cervical cancer following carcinoma in situ of the cervix, 2 and included a companion analysis of second primary cancers following invasive cervical cancers. Briefly, data for the present report were abstracted from the Vaud Cancer Registry file, which includes incident cases of malignant neoplasms in the canton whose population, according to the 1990 census, was about 600 000 inhabitants. 2 Since 1974, a registration scheme, applying the same standardized rules as for incident malignancies, has been implemented for carcinoma in situ (CIS) of the uterine cervix. 2 After exclusion of all synchronous and other cancers (N = 11 after in situ neoplasm, ie 3 cancers of the breast, 2 of the cervix, 4 of the uterine corpus, 1 of the ovary, and 1 of unspecified genital organs; N = 6 after invasive cervical neoplasms, ie 1 cancer of the breast, 3 of the uterine corpus, and 2 of the ovary), the present updated series comprises a total of 2,681 histologically (at least through a biopsy) confirmed CIS, and 893 invasive cervical neoplasms. The age range was 18–92 years (median age 34 years) for CIS, and 21–98 (median age 60 years) for invasive cervical neoplasms. These cases were actively followed up to the end of 1998 (total number of 33,756 and 6,451 person-years of follow-up for CIS and invasive neoplasms, respectively), for the occurrence of cancer, migration, or death. Calculation of expected numbers was based on site-, age- and calendar year-specific incidence rates, multiplied by the observed number of person-years at risk (standardized incidence ratio, SIR), and the corresponding 95% confidence interval (CI), was based on the exact Poisson distribution. 3 Table 1 gives the number of invasive cancers in selected sites or groups of sites following in situ and invasive cervical cancers, in the overall dataset and in strata of <10 and ≥10 years since original diagnosis of cervical cancer. A total of 150 invasive cancers were observed after an in situ cervical cancer versus 145.8 expected, corresponding to a SIR of 1.03. An excess of cervical cancer was restricted to the 10 years since original diagnosis (SIR = 4.18), and an excess of tobacco-related neoplasms was observed only 10 or more years since original diagnosis (SIR = 1.70). The reduced incidence of breast cancer was similar in the two time periods considered. After invasive cervical cancer a total of 81 cancers were observed versus 62 expected, corresponding to a SIR of 1.31. An excess for other genital sites was larger in the first 10 years since diagnosis of cervical cancer, while an excess of tobacco-related neoplasms was similar in the two time periods considered. The SIR for colorectal cancer after in situ or invasive cervical cancer was 2.18 (95% CI = 0.7–5.1) after 10 or more years following invasive cancer. Table 1: Number of Observed (0) Cases and Standardized Incidence Ratios (SIRs, and Corresponding 95% Confidence Intervals [CI]) for Second Primary Cancer after in Situ and Invasive Cervical Cancers According to Time Since Diagnosis. Vaud, Switzerland, 1974–98This up-dated analysis of the Vaud dataset, including a total number of cases more than twice than previously reported, 2 confirms that the risk of invasive cervical cancer is greatly increased in the 10 years following a diagnosis of in situ cervical cancer, and that of other genital sites and anal cancer is increased following both in situ and invasive cervical cancer, pointing to the role of HPV as common aetiological factor on these neoplasms, 4–6 and stressing the importance of accurate monitoring in the few years following a diagnosis of cervical neoplasm. This study also confirms the existence of an excess of tobacco-related neoplasms following cervical neoplasms, thus supporting the hypothesis of a role of tobacco on cervical carcinogenesis. 7,8 The excess of colorectal and of bladder cancer is also consistent with an effect of radiotherapy on these radiosensitive sites near the cervix, 1,9 particularly 10 or more years since original diagnosis. The data were inadequate to quantify the risk of skin and immune-related (lymphoid) neoplasms following a diagnosis of cervical neoplasm. Acknowledgment The contributions of the Vaud Cancer Registry’s staff are gratefully acknowledged. Fabio Levi Van-Cong Te Lalao Randimbison Carlo La Vecchia

Understanding and perceiving risks is a major issue for any health education and preventive intervention (Zeckhauser and Viscusi, 1990; Hertz-Picciotto, 1995). However, comparing risk assessment and risk perception is subject to substantial error (Trichopoulos, 1996; La Vecchia et al., 1999). To address these issues using a systematic and quantitative approach, we have conducted a survey on belief, perception and behaviour of cancer risk in the general population of 5 European countries: Belgium, France, Italy, Portugal and Spain. The study sample was selected from telephone lists in strata of geographic area; over 95% of European households have a telephone. During September 1998, a total of 65,000 households (13,000/country) were sent an anonymous questionnaire (pre-tested in a pilot phase), requesting a reply by subjects aged 16 or over (20 or over in France). A total of 5,579 valid questionnaires was obtained (3,202 females and 2,377 males). No reminder was sent. Only basic demographic information was available from non-responders. The sample of respondents over-represented women, younger age groups and subjects with higher education. Consequently, direct standardisation was used to correct for these factors. The questionnaire included a general section on perception of risk of death and of selected diseases, such as cardiovascular disease, accidents, infectious diseases and cancer. Information was specifically collected on perception of risk and related behaviour for 16 major recognised or potential risk factors for cancer (La Vecchia et al., 1999). For each factor (e.g., “avoiding smoking”), 3 replies were included in the questionnaire for belief (yes, no, I do not know) and 2 for behaviour adoption (yes, no). Table I gives the percent population perceiving (belief) or adopting (behaviour) selected measures to reduce cancer risk, ranked according to belief on the overall data set. Avoiding smoking ranked first according to belief (96.8%) but third (after reducing alcohol and limiting exposure to sunshine) according to behaviour. The second rank according to belief, before alcohol, was limiting exposure to sunshine (92.2%). The fourth and fifth ranks were limiting exposure to UV radiation and avoiding consumption of pesticide-treated vegetables and fruit (74.4%), and a surprisingly high score was given to extremely low-frequency electromagnetic fields (65.3%). A smaller relevance was given to selected nutritional and dietary factors, such as avoiding overweight (59.1%) or excessive calorie intake (51.6%), which was believed to be as important as avoiding genetically modified foods (51.2%). Other measures with no documented impact of cancer risk also showed appreciable proportions of belief, including avoiding using cellular phones (39.6%), avoiding microwaved foods (34.3%) and using food supplements (29.4%). These were also the factors with generally larger between-country variation (e.g., between 57.8% in Italy and 26.3% in France for cellular phones), in the absence, however, of any single and consistent pattern (La Vecchia et al., 1999). With reference to behaviour adopted, 64.1% of respondents reported limiting alcohol drinking, 61.5% limiting exposure to sunshine and 60.9% avoiding smoking. Most other items ranged between 30% and 45%. Between-country variation was, if anything, larger for several behaviours than for beliefs. The sample was large enough to provide reliable estimates for most factors considered but over-sampled women, younger age groups and more educated subjects. These potential sources of bias were corrected by direct standardisation, but some residual bias is possible. A non-quantifiable bias, moreover, may have been introduced by the low response rate, which is inherent in the study design. In conclusion, tobacco and alcohol, the 2 major determinants of cancer on a population level in Europe, were perceived to be major risk factors (Sutton, 1998; Doll, 1999), though consequent behaviours were adopted by only about 60% of the population. Sunshine exposure is a well-defined but quantitatively smaller cancer risk factor on a population level, but was perceived to be a risk factor by most subjects (English et al., 1997). Other sources of non-ionizing radiation, such as electromagnetic fields, whose role in cancer occurrence, if any, remains largely undefined (Poole and Trichopoulos, 1991; Trichopoulos, 1996; Doll, 1999), appeared to be grossly over-estimated. Most nutritional and dietary factors ranked relatively low as risk factors for cancer, including some of the best established ones (i.e., overweight), whereas the role of food colouring, other food additives or pesticides was largely over-estimated (Ames et al., 1995), thus confirming the uncertainties and difficulties in the process of cancer risk communication and perception (Fischhoff, 1999; Gerrard et al., 1999). Yours sincerely, Carlo La Vecchia* , Marco Anelli , Ettore Zuccato*, Roberto Fanelli*, Guy Sermeus§, Natalia Milazzo

H2-receptor antagonists have been widely used since the late 1970s for the treatment of gastrointestinal ulcers and other benign conditions of the stomach, oesophagus and duodenum. Several case reports suggested that long-term therapy with H2-receptor antagonists, mainly cimetidine and ranitidine, might increase the risk of gastric cancer. After early case reports, at least six analytical epidemiological studies (two cohort and four case-control) were published, including a total of about 1000 cases of gastric cancer. The relative risks (RR) were systematically and substantially elevated in the first year since starting H2-receptor antagonist use, and levelled off in the following years. Some excess risk was still apparent during the first 5 years of drug use, probably due to incorrect diagnosis and treatment of pre-existing neoplastic gastric lesions, but the estimated RR was not above unity for > or = 10 years since starting drug treatment in the two studies including information on long-term use. The findings of analytical epidemiological studies are thus consistent with the absence of a causal association between H2-receptor antagonist use and gastric cancer risk. Data on oesophageal and colorectal cancer do not support a relevant relation between cimetidine use and the risk of these neoplasms. With reference to total cancer mortality, in a Danish cohort study, for males the RR was 1.9 in the first year, and 1.4 in the first 5 years; corresponding values for females were 1.7 and 1.5. In a British cohort study, the RR was 3.4 in the first year, and 1.3 in the years 2-10. The excess risk in the first year was essentially due to gastric cancer. Post-marketing surveillance data for omeprazole and other proton pump inhibitors are much scantier than for H2-receptor antagonists, particularly on long-term use.