Cancer is the second leading cause of death globally, causing an estimated 9.6 million deaths in 2018 [1]. Half of the people diagnosed with cancer in England and Wales survive for 10 years or more, but approximately 4 in 10 cases of cancer in the UK could be prevented [2]. Cancer survival has consistently been reported to be lower in the UK than similar European countries [3,4].

Late-stage diagnosis contributes to excess deaths for bowel [5], breast [6], and lung cancer [7] in the UK. Late diagnosis could be related to low awareness of symptoms, leading to delays in seeking medical help. Low cancer symptom awareness has been associated with delays in seeking medical help and poor cancer survival for all cancers combined [8].

Earlier detection can improve patient experience [9], costs to the National Health Service (NHS) [10], and cancer survival, but it relies partly on prompt presentation [11]. Understanding and potentially improving awareness of cancer signs is an important step in reducing the incidence of late-stage cancer and reducing cancer deaths in the UK.

In 2008, Cancer Research UK, in partnership with University College London, King’s College London, and University of Oxford, developed the Cancer Awareness Measure (CAM) [12]. The CAM is a validated survey designed to measure awareness of signs, symptoms, and risk factors for cancer and potential barriers to seeing a doctor.

Cancer Research UK has used the CAM to collect data biannually from 2008 to 2014 from a representative sample of the UK population via the Office for National Statistics (ONS) Opinions and Lifestyle Survey. Questions in the survey are a combination of recall and recognition questions, designed to assess public awareness. Recall questions are open-ended questions, asking participants to list as many cancer warning signs and risk factors that they can think of. These are followed by recognition questions, where participants are given a list of warning signs and risk factors and asked yes/no do they think these are risk factors or warning signs of cancer.

Data from the CAM indicate that the average number of cancer warning signs recognized by representative samples of the UK population has increased from 6.4 (SD 1.9) in 2008 to 6.8 (SD 1.5) in 2014 out of a possible nine warning signs posed in the survey [13]. Recall of risk factors appears to have followed the opposite pattern, with recall decreasing from a mean 2.2 in 2008 to 2.0 in 2014 [13]. Awareness of cancer signs and risk factors has consistently been found to be lower among men [14,15], younger adults [14], and those from lower socioeconomic groups [14,16,17] or ethnic minorities [15,18].

Although CAM data have traditionally been collected via face-to-face interviews conducted by the ONS, the response rates have declined over the years (from 61% in 2008 to 47% in 2017). This study explores the viability of moving data collection online, a move seen in many large market research organizations. In Great Britain, 90% of households have access to the internet, and 73% of people have accessed the internet with a mobile phone [19]. The benefits of online data collection include lower costs [17], higher data quality [20], and a faster rate of return and lower data entry times [21]. Conversely, the limitations may include sampling issues [21] and differences in sampling methodologies [22].

Although the relationships among questionnaire modality, response rates, and accuracy have been described as complex [23], previous research exploring the impact of data collection method is encouraging. Socially desirable behaviors have been reported to be less likely to be disclosed in interviews than online questionnaires [24], and disease prevalence rates are much closer to known rates when using internet studies compared with data collected over the telephone or face-to-face [25].

The primary aim of this study is to identify the extent to which public awareness of cancer and attitudes toward seeking help vary by data collection method (face-to-face vs online data) in adults (aged ≥18 years) in Great Britain. The research objectives are to (1) explore whether online data collection can produce a representative sample of the UK population (differences between samples); (2) compare the awareness of signs, symptoms, and risk factors for cancer, as well as the barriers to seeking help between data collected online and face-to-face (differences in levels of awareness); and (3) explore whether any relationships observed between awareness and demographic variables are consistent across samples (interactions between survey provider and demographic variables).

The Great Britain population statistics were taken from the ONS (midyear population estimates, Households and Individuals Internet Access survey), census data, and NHS Digital (Health Survey for England).

Variables collected in the CAM are outlined in Textbox 1. Details of the development and content of the CAM can be found elsewhere [12].

To reduce bias, open-ended questions about signs, symptoms, and risk factors were asked before closed questions. The number of warning signs endorsed or risk factors recognized were summed to produce total scores. Coding manuals were provided to all market research agencies regarding how to code recalled items to ensure consistency.

In total, 4075 participants completed the CAM. Online participants made up 80% (n=3256) of the sample (agency A: n=1190; agency B: n=2066). The remaining 20% (n=819) of participants completed face-to-face interviews.

The three weighted samples were generally representative of the Great Britain population (Table 1).

The gender split of all three samples largely matched the Great Britain population; however, both online samples were older than the ONS sample and the Great Britain population. Scottish participants were slightly overrepresented by agency A (10.2% vs 8.6% of Great Britain population).

All samples included a higher proportion of white participants than the Great Britain population (Great Britain population: 86%; agency A: 87.5%; agency B, 93%) and reported higher educational attainment. Both online samples had a larger proportion of participants with a partner (agency A: 63%; agency B: 62%) compared with the Great Britain population (50.9%) and were more likely to report being in good health (agency A: 48.5%; agency B: 47.4%; ONS: 42%). Face-to-face participants were less likely to report their health as bad (3.6%; agency A: 6.4%; agency B: 7.1%; Great Britain population: 7%). More than 90% of agency B participants reported using the internet more than once a day compared with 78.5% of face-to-face and 78.8% of agency A participants.

The number of cancer warning signs and risk factors recognized and recalled within each sample are included in Multimedia Appendix 2.

Online participants recognized more risk factors, a mean of 9 of 12 listed compared with 8 for face-to-face participants.

Participants recruited by online agencies were more likely than face-to-face participants to recognize being overweight (agency B: 74%, P<.001, OR 1.4, 95% CI 1.1-1.7; agency A: 73%, P=.004, OR 1.3, 95% CI 1.1-1.7; face-to-face: 67%), having a family history of cancer (agency B: 77%, P=.02, OR 1.3, 95% CI 1.0-1.5; agency A: 77%, P<.001, OR 1.5, 95% CI 1.2-1.9; face-to-face: 69%), eating too much red or processed meat (agency B: 61%, P<.001, OR 1.5, 95% CI 1.3-1.8; agency A: 58%, P<.001, OR 1.5, 95% CI 1.3-1.8; face-to-face: 52%), and infection with human papillomavirus (HPV) as risk factors of cancer (agency A: 41%, P<.001, OR 1.8, 95% CI 1.5-2.2 agency B: 49%, P<.001, OR 2.4, 95% CI 2.0-2.9; face-to-face: 29%). Agency B participants were more likely than face-to-face participants to recognize older age (68% vs 60%, P<.001, OR 1.4, 95% CI 1.2-1.7) but less likely to recognize smoking (95% vs 96%, P=.001, OR 0.4, 95% CI 0.2-0.7) as risk factors of cancer (Table 4).

Online survey participants were significantly more likely to endorse 8 of 14 barriers to seeing a GP than face-to-face participants. The most frequently endorsed barrier for face-to-face and agency B participants was “I find it difficult to get an appointment at a convenient time”; for agency A participants, it was “I don’t like having to talk to the GP receptionist.” Agency B participants were more likely than face-to-face participants to endorse an additional barrier “my doctor is difficult to talk to” (P=.001, OR 1.6, 95% CI 1.2-2.1). Figure 2 shows the percentage of participants that endorsed barriers to going to the doctor.

Recall of bleeding or blood loss, cough, and difficulty swallowing showed significant interactions between sex and survey provider. For participants living in Scotland, those recruited by agency B were significantly less likely to recall bleeding or blood loss as a sign of cancer compared with those recruited by agency A (P=.04).

Fewer females recognized family history as a risk factor of cancer when completing face-to-face interviews than in online surveys (agency A females: P=.006; agency B females: P<.001). Significantly fewer males recognized not doing enough physical exercise as a risk factor of cancer in the agency B survey compared with agency A (P=.02).

This analysis explored the viability of moving from face-to-face to online data collection for the Cancer Research UK’s CAM.

Comparisons of data collected using face-to-face interviews and online surveys revealed minor differences between samples. Both methods provided broadly representative samples of the UK population with slight differences in awareness of signs, symptoms, and risk factors of cancer and frequency of help-seeking barriers reported, leading us to conclude that online data collection for the CAM is possible.

Recall of certain cancer signs and risk factors varied by demographic group. Recall of bleeding/blood loss, cough, and difficulty swallowing had significant interactions between sex and survey provider. Overall, recognition of risk factors was higher in the online surveys.

Recognition of risk factors varied by sex, education level, and country. Significantly fewer females recognized family history as a risk factor of cancer in the face-to-face survey compared with the online surveys. Significantly fewer males recognized not doing enough physical exercise as a risk factor of cancer in the online samples compared with the face-to-face sample. The reasons for these variations are unclear but provide avenues for further research and action.

Overall, online participants recruited by agency A were significantly more likely to recall cancer signs and risk factors compared with both agency B and face-to-face participants. This finding implies that agency A participants may be more engaged and knowledgeable than the other survey participants. Educational levels did not differ greatly among the three samples. Agency A participants may have been more engaged than other participants because they had previously taken part in a face-to-face survey, indicating that they may be a particularly motivated group.

Previous research has found that levels of awareness of the HPV virus [26] and cholesterol [23] were higher among online than face-to-face or paper survey respondents. In this study, online participants recognized more risk factors than face-to-face participants, including being overweight, having a family history of cancer, eating too much red or processed meat, and infection with HPV (cholesterol was not assessed). However, only one of the online samples reported higher mean recall of risk factors compared with face-to-face participants. This particular panel, agency A, recruited participants after they had taken part in a paper survey, which may have resulted in a more engaged and knowledgeable sample.

Survey research within student populations has suggested that online participants are more likely to answer “don’t know” than those completing the same survey face-to-face [7]. Other research suggests that nonresponse to open-ended questions can be reduced through online data collection [8]. In this study, face-to-face participants were less likely than online participants to respond to recall questions around signs, symptoms, and risk factors with “don’t know.”

Socially desirable behaviors have been found to be less likely to be disclosed in interviews than online questionnaires [27,28]. In this study, online participants were more likely than face-to-face participants to endorse barriers to seeking help. Participants may have found it easier to endorse barriers to visiting the doctor with the context of anonymity afforded by online data collection compared with face-to-face data collection.

Although this study provides insights into the possibility of using online data collection for a large representative sample of the UK, there are limitations that warrant consideration. Regarding recruitment, large differences exist in the size of samples recruited online and face-to-face, highlighting the comparative ease of online recruitment. Previous research indicates that online research may not be as representative as face-to-face interviewing [29], but this is often based on the type of recruitment procedures that precede data collection. In this study, both online samples were recruited through panels; however, there may be differences in the ways that panels are recruited and incentivized, which may have affected the results. To mitigate this, each agency employed procedures to ensure their samples were as representative as possible of the Great Britain population.

For the analysis, it was not possible to calculate unique weighting variables, and we relied on those provided by agencies. The questions within each survey were identical; however, there may have been small differences in the presentation of questions within each sample.

It was necessary to limit the demographic variables studied to control the length of the survey, meaning that unobserved differences may have contributed to the differences observed.

It was not possible to compare the samples collected by each survey agency with the Great Britain population data. The Great Britain population data used were publicly available, although confidence intervals were not provided, and statistically significant comparisons were not possible.

It was not possible to access information about response rates or completion times within each sample. This information may have been useful to explore the differences among samples in more depth.

The relationships between sampling, sample representativeness, survey modality, and subsequent responses are complex. Although sample representativeness varied a little between samples and there are likely unobserved differences, we were encouraged to see that these variations were small overall. This information will be useful in helping us to tailor our recruitment strategy to ensure that we recruit a sample that is as representative as possible of the Great Britain population in future CAM research.

We observed larger differences when looking at responses to the awareness questions themselves, even between the two online samples, which point to the fact that there may be differences in the sampling and running of these panels contributing to these differences.

Nevertheless, the flexibility and potential cost savings of online data collection will enable larger samples and greater variation in content at a lower cost, which will enable the CAM to explore a new and wider range of issues related to the early diagnosis, prevention, and treatment of cancer.