Study design

We performed a cohort study in the Danish County of Aarhus, which has 640,000 inhabitants, approx. 3,000 new cancer cases per year and approx. 500 GPs. Denmark's publicly funded health-care system provides free access to primary care and outpatient and hospital care (Figure 1). More than 98% of the Danish citizens are registered with a GP [19, 20]. GPs function as gatekeepers to the rest of the health-care system, carrying out initial diagnostic investigations including referral to initial x-ray and endoscopic procedures When indicated, they refer patients to hospitals or outpatient clinics, thus delegating further responsibility to secondary care. Danish GPs are required to keep detailed electronic records.

We included all newly diagnosed cancer patients during the 1-year period from 1 September 2004 to 31 August 2005. Patients younger than 18 years and patients with non-melanoma skin cancers were excluded. Patients were enrolled based on computerised data from the county hospital discharge registry (HDR), which for each hospital admission and outpatient visit records the patient's civil registration number (CRN), dates of admission and discharge, surgical procedure(s) performed and discharge diagnoses classified according to the international classification of diseases (ICD-10). We included all patients listed in the registry during the study period with incident cancer diagnoses, i.e. we excluded those with a recurrent cancer. Patients with previous cancer, but of another type, were regarded as incident cases. We then linked the HDR data to the county's Health Service Registry (HSR) in order to identify each patient's GP. The personal identifier used to link records to individuals across Danish registries is the CRN, which is assigned to all Danish citizens [21].

Ethics approval

According to the Scientific Ethics Committee in the County of Aarhus, the project did not need approval by the Danish Biomedical Research Ethics Committee System. The study was approved by the Danish Data Protection Agency and the Danish National Board of Health.

Data collection

A questionnaire was sent to the GP of each patient identified in the HDR. The GP was asked to confirm the diagnosis and to give a detailed description of the patient's diagnostic pathway using the dates reported in the medical record and the mandatory discharge letters from hospitals and specialists which report the date of first admission, the date of diagnosis and of first treatment. Furthermore, the GP provided the dates of the first patient symptoms (adherent to the actual cancer disease) as reported to the GP by the patient, the first presentation to the GP/practice, the initiation of diagnostic procedures, hospital referral, the first hospital visit, the diagnosis and the treatment start. Non-responders received a reminder after three weeks. The questionnaire was pilot-tested by the scientific staff and lay persons at the Department and Research Unit for General Practice, Aarhus and by 40 GPs from Vejle County, Denmark (purposefully selected by gender, age and practice location) [22].

Outcome measures

In accordance with most other research we used the term delay when calculating the different time intervals from first symptom to treatment start. Delay was calculated from the dates entered in the questionnaires by the GPs. Delay was divided into three delay stages. (1) Patient delay: Time from first perceived patient symptom until first presentation to the GP; (2) GP delay: Period from first presentation until initiation of an investigation of potentially cancer-related symptoms, i.e. biopsy, referral to imaging diagnostic, endoscopy or to other health-care professionals; (3) System delay: Time from the start of the GP-initiated investigation until the start of treatment. System delay was further divided into delay encountered in the primary health care sector (GP retaining responsibility for the patient) and in the secondary health care sector (the patient is no longer the responsibility of the GP) (Figure 2). System delay was defined in this way to be able to analyse to which extent cancer was seen and acted upon as an acute disease. In addition, total delay was defined as the interval from the first perceived patient symptom until treatment start.

We first computed delay for all cancer types and then for the 10 most frequent cancers found among the patients. The delay data are presented as medians and interquartile intervals (IQIs). The Wilcoxon rank-sum test was used to compare delay for those patients whose GPs were involved in the diagnostic investigation and those whose GPs were not. A P-value of 0.05 or less was used to signify statistical significance. Data were analysed using Stata software, version 11 (StataCorp, College Station, Tex).

Descriptive data

A total of 2,212 questionnaires were completed. A total of 467 out of 543 physicians from 255 general practices participated in the study. On average, they filled in questionnaires on 4.7 patients (range 1-15, median 4). The physician response rate was 83% (Figure 3). No significant differences were found between the participating and non-participating physicians concerning gender, practice organization and years since graduation. GPs were involved in the diagnostic pathway in 1,892 (86%) of the 2,212 cancer cases. For the remaining 320 cancer cases, we could only calculate system delay in the secondary health-care sector because they were admitted directly to the hospital by out-of-hours or accident and emergency services (A&E), often for acute-onset symptoms. Table 1 shows the distribution of the 10 most frequent cancers diagnosed in Denmark [23] compared with the cancer distribution by type in our study.

Delay data

The shortest total delay was seen among patients with ovarian cancer (median 60 days, IQI 45 to 112); the second shortest delay was seen among patients with breast cancer (median 65 days, IQI 39 to 106). The longest total delay was found among patients with prostate (median 130, IQI 89 to 254) and bladder cancer (median 134, IQI 93 to 181) (Table 2). As shown in Figure 4, the 4^th quartile of patients experienced long patient, GP, system and total delay. Non-response analysis revealed no major differences between participating and non-participating physicians in terms of their cancer patients' age, gender or distribution of cancer diagnoses.

Comparison of findings with previous literature

Seen from a health services planning and a research and population perspective, it is relevant to include all cancers to be able to formulate a hypothesis about the need to focus on patient awareness of symptoms and access to diagnostic work-up of serious disease. However, our data on the specific cancer types also emphasise the need for detailed analyses of each specific cancer. We also need more research on the influence of symptom presentation in relation to delay [26]. A previous survey from the same region by Bjerager [13] showed that most delay in diagnosing lung cancer was attributable to system delay, which is consistent with our findings.

Our total delay findings accord with those of Allgar and Neal [17] who investigated patient-reported delay for six cancer types in a large population of UK cancer patients. However, in contrast to our results, they found that the main problem was patient and primary care delay. This discrepancy may be attributed to differences in the English and Danish health care systems in terms of culture, organisation and capacity, and to differences in study design and definitions of delay stages. Our broad definition of system delay encompassed not only time to diagnosis, but the time to start of treatment, as we expected a significant delay between diagnosis and treatment initiation. We need an internationally agreed definition on delay stages (Figure 2); a task that has been initiated in the UK National Initiative on Early Diagnosis of Cancer (NAEDI) and in the Cancer and Primary Care Research International Network (CA-PRI) [2, 27].

Implications of the study

Cancer is a serious disease. The diagnostic evaluation is complex and often conducted in a sequential process, which may explain the long system delay. However, another explanation may be that Danish cancer patients have to wait longer for basic diagnostic investigations. Thus, another main finding of the present study was the long patient delay. We lack international knowledge about variation in patient awareness of symptoms and differences in delay and about the interaction between the health care structure and patient delay [28–30].

A recent paper showed that delay in diagnosis may have severe prognostic consequences and may partly explain the bad cancer outcome results in countries like the UK and Denmark [12]. Recent years have therefore seen the introduction of different initiatives to reduce system delay in the UK. However, a fast-track approach based on a GP's suspicion of cancer may not necessarily be effective as only some 50% of patients present to their GP with alarm symptoms [26, 31]. The effect of seeing cancer as an acute disease and thereby circumventing administrative and capacity-driven bottlenecks in the pathway should be investigated. The fast-track approach was adopted in Denmark in 2008 [1, 27], but it is too early to evaluate if this initiative has reduced system-related delay.

Further research is required to test whether improved access to diagnostic investigations and intra- and inter-sectoral cooperation within the health care sector will have an effect on the diagnosis of cancer.

Further analysis of the long delays characterising the 4^th quartile of patients requires clarification of the specific patient, GP and system characteristics that prolong delay. Our investigations invite the conclusion that a more precise research focus on early diagnosis of cancer is, indeed, warranted [2].