Our data were obtained via a questionnaire survey [additional file 1], using an instrument designed for self-completion without supervision. This instrument was based on those used in previous local studies of screening knowledge and attitudes [8, 9], augmented with questions suggested by other studies of cancer risk perceptions [5, 10, 11]. After initial construction, the instrument and supporting materials were reviewed for intelligibility by two lay representatives from the Nottingham Primary Care Research Partnership Consumer Panel. The documentation was revised in the light of their observations and suggestions, presented for ethical approval, and revised again accordingly.
As with the previous local studies, we sought the assistance of general practitioners in distributing questionnaires. Two practices, one in Nottingham and one in Mansfield (approximately 12 miles to the north), agreed to send out questionnaires on our behalf. Although both practices were urban, the areas served by each differed in terms of socio-economic deprivation. On the Index of Multiple Deprivation [12], the latter's area was ranked 3,103 amongst the 32,482 enumeration areas of England, whilst the former's was ranked 19,572 (rank 1 = most deprived). The deprivation status of a general practice site generally proves an adequate proxy for deprivation status of the population which it serves [13].
The general practitioners were provided with pre-prepared survey packages, each package containing an invitation letter, a blank questionnaire, a patient information sheet, and a pre-paid response envelope for the return of the completed questionnaire. The practices were asked to mail packages to registered patients aged between 30 and 70 years, but were granted discretion in exclusion. As with the previous studies, patients with a current diagnosis of cancer or with learning disabilities were excluded from the samples, on the grounds of the need to avoid distress and incapacity to complete the instrument, respectively. As the questionnaires were anonymous, we were unable to enhance compliance by contacting non-responders or to identify the non-responders' characteristics. All data were collected in late-2007 and, as far as the results reported here are concerned, all questions were closed-ended (mostly of the tick-box variety).
The opening section of the questionnaire was entitled: "What do you know about cancer?". Subjects were asked, first, to estimate the number of newly-diagnosed cancer cases each year, by selecting one of six numerical values (125,000 to 500,000, in steps of 75,000). As the true number lay between 275,000 and 350,000 [14], values of 125,000 and 200,000 were classified as under-estimates of incidence, and those of 425,000 and 500,000 were classified as over-estimates. Second, subjects were asked to select (in effect, to vote for) "the two most common cancers" in the UK from an alphabetical list of six, namely, bowel (colorectal), breast, cervical, lung, prostate, skin (melanoma). Epidemiologically, the adjective "most common" is ambiguous and might be taken to refer to incidence, to prevalence or to mortality. The imprecision in phrasing was chosen deliberately, however, on the expectation that respondents would be more comfortable with intuitive than with technical phrasing [15]. Whilst the correct answer to this second question is therefore equivocal, the imprecision does not affect our ability to interpret subject response. For some time, the two cancers with the highest incidence have been breast and lung, closely followed by bowel [14]. Lung and bowel are the two most common causes of cancer-related death [16], whilst breast and bowel have the highest prevalences in the population [17]. It follows that, irrespective of interpretation, prostate cancer, melanoma or cervical cancer cannot be considered amongst the "two most common". Indeed, the incidence of cervical cancer is particularly low in England, owing to the effectiveness of the national screening programme.
Subjects were asked to rate each of eight cancer risk factors as a major risk, a minor risk, or as no risk. A "don't know" option was also available. We included increasing age, smoking tobacco and over-indulgence in alcohol use, all of which are generally presumed by health professionals to represent principal risk factors for most forms of cancer. We included being overweight and a lack of exercise, which are known to be risk factors for some of the more common cancers, such as breast and bowel. Sexually-transmitted infection is relevant for a few specific cancers only, although the association between human papillomavirus and cervical cancer is being publicised increasingly. We included two factors which are commonly-cited, although less-well-validated. These were, first, a family history of cancer (genetic risk), even though it "is uncommon for cancer to run in a family... Most of the time, multiple cases of cancer in a family are just a matter of chance" ([18] p.15). Second, we included persistent stress and anxiety, in spite of the association between stress and cancer remaining equivocal [19].
Each subject was offered three true-or-false questions, namely, whether sun-bathing caused skin cancer, whether cancer was more common amongst women than it was amongst men, and whether more people died of heart disease than died of cancer. The first statement, of course, has long been accepted as true by both practitioners and public alike [20, 21] and we expected most, if not all, subjects to respond accordingly. The numbers of males and of females dying from cancer each year are approximately equal, as are the numbers of new cases diagnosed. Age-standardised rates for males, however, have long been significantly higher than those for females [22]. The relative importance of cancer and heart disease is a question of interpretation: "At all ages, there are more deaths from cancer than from ischaemic heart disease. However, no single cancer is a more common cause of death than ischaemic heart disease" ([23] p.16). It should be stressed that the study was less concerned with subjects being objectively wrong or right in their knowledge and more concerned with the existence of differences in response between the sexes.
Three types of cancer screening currently available via the NHS - bowel, breast and prostate - were described briefly. We restricted discussion to these three types, on the grounds that men and women would each be eligible for two types only. We asked the respondents whether they believed that the NHS should continue to provide screening services such as these. They were also asked to indicate whether they had already been screened, or whether or not they would accept a test were one to be offered.
The final section of the questionnaire was modelled closely on the instruments used in previous local studies [8, 9]. It requested socio-demographic information including sex, age, marital status, and age on leaving full-time education. Ethnic origin was to be identified as one of African, Afro-Caribbean, Asian, Chinese, White or "other". Annual household income was represented by a choice of one from four income bands, beginning at 0-£10,000 and thereafter in increments of £10,000 to £30,000 and above. An individual with average household income in and around the Nottingham would have selected the £20-30,000 band [24]. We requested an indication of smoking status (current smoker, ex-smoker, never smoked) and subjects' perception of their weight (over-, under-, about right). Subjects reported illness experiences, i.e. whether they or a close family member had ever suffered from each of six conditions (stomach complaints, piles or haemorrhoids, heart disease, cancer, stroke, depression). We invited subjects to indicate their degree of worry about the prospect of cancer (4-point scale: not worried, minor concerns, quite worried, very worried). To assess personal risk perception, we invited respondents to assess their own chances of getting cancer, relative to people of their own age and sex (5-point scale: much less, less, the same, more, much more). Subjects described their current states of health by completing standard EQ-5D (EuroQoL) health state classifications [25]. Classified states were translated into EQ index scores using to the current UK tariff [26].
We planned two types of analysis. First, responses to the factual questions, such as estimated incidence, identity of most common cancers and risk factors, were examined with respect to male-female differences in proportions. Beyond establishing whether men's and women's judgements were accurate, we were concerned to establish whether opinions differed systematically by sex. Second, three specific beliefs or attitudes about cancer and screening were modelled using logistic regressions, again with a view of establishing the existence, or otherwise, of male-female differences. The dependent variables in the regressions were (i) the belief that the chances of getting cancer were above average, (ii) having substantial concerns about cancer, and (iii) actual or potential participation in bowel cancer screening. We restricted the analysis of participation to bowel screening, as this was the only type for which both sexes were eligible. All of the socio-demographic and knowledge variables were candidates for the models initially, which were re-estimated after excluding variables with insignificant coefficients.