Design
We provided clinical vignettes to PCPs from twenty European countries with markedly different levels of socio-economic development, healthcare organisation and investment in healthcare. The vignettes described patients presenting with symptoms that could indicate cancer. Recruitment started in November 2015 and was completed at the end of 2016.
Study population
The Örenäs Research Group (ÖRG) is a European collaborative of primary care researchers, formed in 2013 to study the factors influencing national variations in the early diagnosis of cancer in primary care. The research was conducted in 25 ÖRG centres in 20 countries across Europe: Bulgaria, Croatia, Denmark, England, Finland, France, Germany, Greece, Israel, Italy, Netherlands, Norway, Poland, Portugal, Romania, Scotland, Slovenia, Spain, Sweden and Switzerland. Medical doctors were eligible for the survey if they were working mainly in primary care. These doctors, here referred to collectively as ‘Primary Care Practitioners’, included General Practitioners (GPs) and other doctors who had other specialist training but worked in the community and could be accessed directly by patients without referral.
Development of the questionnaire
Following a literature review, ÖRG investigators developed a questionnaire designed to elicit PCPs’ diagnostic actions for patients that could have cancer. A questionnaire with five clinical vignettes was piloted by the ÖRG local leads in January 2015 to check validity. One of the vignettes was found to be invalid and was removed. The next version of the questionnaire, in English, was then piloted by 49 PCPs in 16 ÖRG member countries in July 2015. No changes to the vignettes were made following this second pilot.
ÖRG leads arranged for translations of the questionnaire into their local languages where these were not English, a total of 19 translations from the original English. Translation and validation by backtranslation were done in a standardised way (31) and are described elsewhere (32).
Description of the questionnaire
The questionnaire consisted of 47 items and was divided into four sections: (a) demographic questions (five questions about years since graduation, gender, type and rural/urban location of practice and number of doctors working in the practice); (b) referral availability questions (two questions about tests and specialist opinions that were either directly or indirectly available to the respondent); (c) four clinical vignettes and (d) 20 health system factor questions. Each of the vignettes provided information on the patient’s presenting symptoms, previous medical history, medication, clinical findings and other relevant information. Two of the vignettes were designed and validated by the ICBP (33), and used with permission. The vignettes were:
- A 62-year-old male smoker with chronic obstructive pulmonary disease and now a two-week history of a productive cough; positive predictive value (PPV) for lung cancer: 3.6% (34);
- A 53-year-old woman with lower abdominal pain and abdominal distension; PPV for ovarian cancer: 3.1% (35);
- A 35-year-old breastfeeding woman with an abnormal nipple discharge and eczematous changes around the nipple; PPV for breast cancer: 1.2% (36);
- A 22-year-old man with coeliac disease who now has abdominal pain, rectal bleeding and diarrhoea; PPV for colorectal cancer: 3.4% (34).
For each patient, a range of five possible management decisions was given (whether the respondent would prescribe medication, arrange a follow-up appointment, use watchful waiting, organise a diagnostic test, refer the patient), with a ‘yes/no’ option for each. Those that chose to investigate the patient were able to select from a range of possible diagnostic tests. The response of primary interest was a PCPs’ management choice that would be likely to identify a cancer as a cause of the patients’ symptoms, by either opting to request a significant diagnostic test or by referring to a specialist. The tests used in the analysis were: a plain chest X-ray or lung computerised tomography (CT) for the lung vignette; a tumour marker, diagnostic ultrasound or CT for the ovarian vignette; an ultrasound of the breast or mammography for the breast vignette; and diagnostic ultrasound, sigmoidoscopy, colonoscopy or CT colonography for the colorectal vignette. A factor analysis of the results of the survey section on the effect of health system factors is reported separately (37).
Sample size
We aimed for a total sample size of at least 1000 PCPs, with at least 50 responses from each of the participating countries.
Recruitment of participants
Each ÖRG local lead was asked to email an invitation to take part in the survey to the PCPs in their local health district or jurisdiction, and to recruit at least 50 participants. In six countries (Denmark, Norway, Portugal, Romania, Slovenia, Sweden), the invitation was distributed to a national sample. The recruitment email stated that the research aimed to identify which health system factors affect PCPs’ decisions to refer patients for further investigation. The possibility of cancer as being a cause of the vignette symptoms was not mentioned in either the recruitment email or the survey. As low survey response rates are common in primary care (38) and can vary between jurisdictions, any local leads who had difficulty in achieving the required sample sizes were asked to increase the number of responses by using snowballing, a recognised technique for recruiting hard-to-reach populations in health studies (39,40).
Distribution of the questionnaire
The questionnaire was designed using SurveyMonkey (41). Because of the study’s wide geographical coverage, on-line delivery of the questionnaire was used; this methodology has previously been successfully used in research involving cancer care professionals (42).
Statistical analysis
Demographic questions and those relating to vignette diagnostic actions were analysed using descriptive statistics for decision to arrange a diagnostic test, and to refer to a specialist. As it was considered that some PCPs would not investigate because they were referring to a specialist, and conversely some PCPs would not refer to a specialist because they were investigating, we also used a composite measure of a decision to either arrange a diagnostic test and/or refer to a specialist, i.e. the likelihood of taking diagnostic action for cancer. For each individual PCP, mean diagnostic action rates were calculated from the four individual vignette responses. From those, mean diagnostic action rates were calculated for each country. For comparisons between countries, medians and ranges were calculated.
For each country, the mean 1-year and 5-year relative cancer survival rates for the four cancers of interest (lung, ovary, breast and colorectal) were calculated from EUROCARE–5 data (2). These are shown in Table 1. Linear correlations and 2-tailed significance levels were estimated for the proportions of PCPs opting to take diagnostic action and these national 1-year relative cancer survival rates, and also between PCPs’ likelihood of organising a diagnostic test and their likelihood of referral to a specialist. As one-year relative survival rates for cancer can be affected by lead-time bias, we made a sensitivity analysis using 5-year survival.