The role of general practice in routes to diagnosis of lung cancer in Denmark: a population-based study of general practice involvement, diagnostic activity and diagnostic intervals

The study took place in Denmark in 2010. Danish GPs act as gatekeepers to the rest of the health-care system, except for emergencies; and 99% of all Danish citizens are registered with a general practice which they consult for medical advice.

Patients were identified in the Danish National Patient Registry (NPR) [14]. The NPR is a national population-based database containing admission and discharge dates for all Danish citizens treated at Danish hospitals; and all recorded diagnoses are classified according to the International Classification of Diseases (ICD-10).

Patients were included if they met the following inclusion criteria: 1) registered in the NPR with ICD-10 code C34.0-9 as the primary diagnosis, 2) diagnosed in the study period from 1 May 2010 to 31 August 2010, 3) living in Denmark, 4) ≥ 18 years and 5) listed with a Danish GP. We excluded patients who had previously been registered with any cancer type, except for non-melanoma skin cancer (C44), in the Danish Cancer Registry (DCR) [15]. The sampling of patients has been described in detail elsewhere [16].

A questionnaire was sent to the general practice at which the patient was listed. The questionnaire solicited information about the extent of the GP’s involvement in the lung cancer diagnosis and the key dates in the diagnostic process. Participating GPs were also asked to list the symptoms and signs presented by the patients and to describe how they interpreted these symptoms. The questionnaire was developed in 2009 by a group of researchers at the Research Unit for General Practice, Aarhus University [17]. As no pre-designed questionnaires for this specific purpose were available, ad hoc questions were constructed on the basis of previously used validated items [9,16,18]. In practices with several GPs, the GP who was most familiar with the patient was asked to complete the questionnaire on the basis of the medical records. Non-responders received a reminder after four weeks. No reimbursement was offered for participation.

We used the DCR to verify the diagnosis and to obtain data on tumour stage at diagnosis [15]. The DCR contains information about Danish cancer patients, including the date of diagnosis and tumour characteristics. DCR information on tumour characteristics at diagnosis rests on a multidisciplinary health-care team’s evaluation of either pathological (pTNM) or clinical (cTNM) data [19]. If any of the T, N or M values were missing, we categorised SCLC as ‘limited’ if the tumour was M0 (no metastases present) and as ‘extensive’ if the tumour was M1 (metastases present) regardless of the values (known or unknown) of other components. We categorised NSCLC as ‘advanced’ if the TNM stage included values of T4, N3 or M1 regardless of other components [20] (Table 1). To adjust for confounding by patient characteristics, we obtained data on comorbidity from the NPR; these data were based on ICD-10 codes for previous hospitalisations until the date of diagnosis. The presence of comorbidity was defined according to the Charlson Comorbidity Index (CCI) [21,22] and categorised as ‘low’ (CCI = 0), ‘medium’ (CCI = 1-2) or ‘high’ (CCI ≥ 3). Information about performed X-rays was also obtained from the NPR. Furthermore, we obtained socioeconomic information from the Danish Civil Registration System (CSR) [13] and the Danish Integrated Database for Labour Market Research (IDA).

The patients were divided into groups depending on whether or not the GP responded to the questionnaire. GPs involved in the diagnosis were asked to state whether the patient had been referred through a fast-track route. Moreover, GPs were asked to rate their interpretation of presented symptoms as either 1) Alarm symptoms suggestive of cancer (alarm symptoms), 2) Symptoms suggestive of any serious illness (serious, but unspecific symptoms) or 3) Vague or ill-defined symptoms not directly suggestive of cancer or other serious illness (vague symptoms).

The primary care intervals and the diagnostic intervals were calculated by combining data from the DCR and the questionnaire. The primary care interval was defined as the time from the first presentation of lung cancer symptoms in primary care until referral to secondary care, and the diagnostic interval was defined as the time from first presentation until final diagnosis [24].

Cancer stage at diagnosis was classified according to the TNM system (version 6) and dichotomised into local and advanced disease. A cut-point between stage IIB and IIIA was chosen as previous studies have documented a significant difference in mortality between these two stages [25].

The socioeconomic variables considered in the study were education and marital status. Education included basic school and was dichotomised into ‘≤10 years’ and ‘>10 years’ [26]. Marital status was dichotomised into ‘cohabitating’ or ‘living alone’.

Comorbidity was assessed by using the Charlson Comorbidity Index (CCI); the index date was set as the day before the first contact with the GP (for patients for whom the GP was involved in the diagnosis) or the day before diagnosis (for patients for whom the GP was not involved in the diagnosis).

Patient groups were compared using Wilcoxon’s rank-sum test for ordinal or continuous data, including time intervals, Kruskal-Wallis test for differences between groups or Pearson’s chi-squared test for nominal or dichotomous data. In the analyses for Table 2, patients with unknown characteristics were treated similarly to patients with known characteristics, but patients with unknown characteristics were excluded in the analyses for Table 3.

Backward cumulative curves for the dates of the latest and the second-latest X-ray before diagnosis and associated 95% confidence bands were drawn by applying a standard Kaplan-Maier procedure and normal approximation on a reversed time scale.

We used generalised linear models for the binomial family to calculate the associations between long intervals and gender, age, marital status, education, comorbidity, GP interpretation and use of fast-track referral. Long intervals were defined as the 4th quartile for the full study population. Prevalence ratios were preferred to odds ratios, which would tend to overestimate the associations as the prevalence of the outcome measure was above 20% [27]. Consequently, we chose the logarithm for the link function to facilitate direct estimation of prevalence ratios. Analysis of time intervals was restricted to patients in whose diagnosis the GP was involved. Estimates are presented with 95% confidence intervals (95% CIs) when relevant. Data were analysed using the statistical software Stata 12.0 (StataCorp LP, TX, USA).

The study was approved by the Danish Data Protection Agency (Ref. no.: 2010-41-4694) and the Danish Health and Medicines Authority (J. no.: 7-505-29-1484/1 and Ref. no.: 7-604-04-2/195/EHE). The study did not require approval from the Committee on Health Research Ethics of the Central Denmark Region nor did it require consent from individuals as no biomedical intervention was performed.