Non- or full-laxative CT colonography vs. endoscopic tests for colorectal cancer screening: A randomised survey comparing public perceptions and intentions to undergo testing

The research team sent out postal invitations to 13 primary care centres (General Practices) that were identified by regional Primary Care Research Networks in England as being potentially interested in participating in research. Of the seven practices responding to the invitation, three were selected in order to achieve a diverse mix of geographic and demographic characteristics: two were located in Cumbria, a rural area with an affluent, predominantly white British population; one was in an urban area (London) with a heterogeneous socioeconomic and ethnic population.

Following ethical approval, invitees were randomly assigned to be mailed one of four study packs (in June 2013) consisting principally of an information booklet and leaflet (explaining screening and characteristics of one of the four tests), a standardised questionnaire and a Freepost envelope. Invitees were requested to read the booklet and leaflet, complete the questionnaire and return it in the envelope. Completion of the questionnaire constituted implied consent to participate. Invitees who did not wish to participate were requested to return the questionnaire blank and individuals were sent reminder letters if they had not responded within 2 weeks of being contacted. Written materials are reproduced in Appendices 1.1 to 1.4.

Eligible participants were all adults aged 45-54 years. Exclusion criteria were (1) previous diagnosis of bowel cancer; (2) diagnosis of any cancer in the previous 6 months; (3) learning disability; (4) receiving regular colonoscopies; (5) experiencing significant cognitive decline; (6) clinically judged to be unsuitable to participate in the study (e.g. due to a terminal illness, mental health issue or inability to read English). Electronic patient records were screened using database queries and manual review by primary care practitioners. After exclusions, a random sample of eligible participants was selected at each practice in order to obtain an approximately 2:1 ratio of Cumbria and London invitees for a more nationally representative proportion of white British invitees.

Participants were sent information regarding one of four tests: non-laxative CTC, full-laxative CTC, FS and OC. This information was designed using detailed literature from the English Bowel Cancer Screening Programme (BCSP) as a template. The BCSP provides invitees with written information on the current standard screening method in England (guaiac faecal occult blood testing, followed by OC if the test is abnormal) [19, 20] and these booklets provide a useful indication of the material that would be required if any of the four tests in this study were offered as part of a national screening programme. Hence, study information materials were designed to be identical in style and wording to BCSP materials and were consistent across test conditions wherever possible (see Fig. 1 for an example). Key similarities and differences in the descriptions for each test are described in Appendix 2. As an example of a characteristic that differed between test conditions, FS was described as reducing the risk of getting bowel cancer by 23 % and dying from the disease by 31 % [from 21]. Since tests that demonstrate the whole colon confer greater theoretical benefit than a test of the distal colon only, CTC and OC were described as being estimated to reduce the respective risks by at least 23 % and 31 %. Statistics were derived from existing literature [20‐34] and expert radiologist opinion. Drafts of study materials were piloted using a “think-aloud” technique [35] in which lay members of the public read through each leaflet out loud and articulated what they were thinking at various stages. This allowed text that was difficult to understand to be identified and amended.

Participants were asked: “Imagine you have just received an invitation to have the [test name] test in the post. Realistically speaking, would you decide to have the [test name] test?” Response options were “definitely not”, “probably not”, “yes, probably” and “yes, definitely”. The primary outcome was whether participants stated that they would “definitely” have the test (participants responding with this option were categorised as intenders; participants responding with any other option were categorised as non-intenders). This question was adapted from a multicentre randomised trial on efficacy of FS, where there is evidence that responses are a strong predictor of subsequent screening behaviour [22, 36, 37].

Invitees were asked to rate their agreement with six statements regarding their perceptions of the test itself (e.g. “having the test would be embarrassing”). Response options consisted of a 5-point Likert scale ranging from “strongly disagree” to “strongly agree” and were scored 1-5 with higher scores representing more positive perceptions. Cronbach’s α (a measure of whether a set of items reliably measure the same factor) was 0.72 for items relating to perceptions of the test, indicating that the average correlation between scores on these items was acceptably high. Consequently, scores for these six items were summed for each participant to create an overall test perception score. Invitees were also asked to rate how strongly they agreed or disagreed with three items relating to perceptions of health benefits (e.g. “having the test would reduce my risk of getting bowel cancer”) and four items relating to perceptions of the preparation (e.g. “preparing before the test would be uncomfortable”). These items were analysed separately (instead of being summed to create two overall scores for each participant) due to low Cronbach’s α that could not be improved by omitting variables (α = 0.62 and 0.68, respectively). Finally, a 5-point response scale ranging from “very unlikely” to “very likely” allowed participants to rate their views on two items relating to sensitivity (“the test finding bowel cancer if I have it” and “the test finding polyps if I have any”; α = 0.91 and so responses were summed to create an overall sensitivity perception score for each participant) and one item that assessed views of specificity (“having a follow-up test that shows there was no problem”). The choice of items was influenced by previously used measures of screening test benefits and barriers [38, 39]. The secondary outcomes consisted of overall scores for test tolerability and sensitivity along with scores for individual items on preparation, specificity and health benefits.

Participants were also asked to report whether they had read the information materials (question adapted from [40]) and asked six multiple-choice questions to assess comprehension (e.g. “The test would be done: At a hospital/At home/At a GP practice”) with one point given for each correct answer.

The questionnaire ended with questions on demographics (e.g. ethnicity and markers of socioeconomic status), participants’ perceived risk status compared to others (from [41]) and self-rated health. An estimate of socioeconomic status was derived using a previously used method [42] that combined responses to questions on education, and vehicle and home ownership (one point was given for each having no formal qualifications, no vehicle ownership and living in rented accommodation). Scores ranged from 0-3; higher scores represented greater deprivation.

A meta-analysis that pooled heterogeneous studies of hospital-based screening tests and a trial of non-laxative CTC found actual uptake to vary around 30 % for the four tests [43, 44]. Assuming that intentions to be screened would be 20 % higher to account for the gap between intentions and behaviour [37], approximately 50 % would definitely intend to take the test. Four hundred four participants were estimated to be required per group in order to detect a difference of at least 10 % between two tests (considered clinically meaningful; α = 0.05, β = 0.2) in an analysis of valid responders. A total of 3,100 individuals were invited, anticipating a response rate of approximately 50 %.

Frequencies and percentages were used to illustrate demographic and other background data. All inferential analyses used non-parametric methods (i.e. Kruskal-Wallis or Mann-Whitney U tests to compare ordinal data and Pearson χ² tests to compare proportions) since Kolmogorov-Smirnov tests indicated that relevant data were non-normal (p < 0.05).

Pearson χ² tests were used to compare the proportion of intenders between tests. The first analysis compared valid responders who had completed the study per the protocol (i.e. they did not report being outside the eligible age range for the study, had received the correct study pack, stated that they had read the information in full and had answered the question on intention to be screened). Sensitivity analyses compared intentions between tests after re-categorising intenders as those who responded with either “yes, definitely” or “yes, probably”. Both analyses were then repeated in an expanded analysis of all invitees (i.e. irrespective of whether they had responded or not, whether study packs were undelivered or whether they met other criteria for the main analysis). It was assumed that non-responders to the question on intention would not have the test (or could not intend if the study pack was returned undelivered). Wilson score 95 % confidence intervals were calculated for each test individually and across all tests.

Scores on items relating to perceptions were compared using Kruskal-Wallis tests, followed by planned post-hoc Mann-Whitney U tests to compare scores for which there was strong evidence against the null hypothesis (p < 0.05). Missing data for items relating to perceptions were imputed using multiple imputation [45]. Results did not differ meaningfully between imputed and non-imputed analyses so only non-imputed statistics are shown for post-hoc comparisons. All analyses were carried out using SPSS 21 for Windows (IBM, Armonk, NY).