Setting and population
The study was conducted in the Central Denmark Region, where the first organised breast cancer screening round was performed in 2008–2009. The population in this region is estimated at 1.2 million. All women aged 50–69 years living in this region (N = 149,234) were invited by postal mail to participate in the screening programme. The invited women received a fixed (yet changeable) date, and no reminders were sent to non-participants. In total, 78.9% participated in the first screening round (Jensen et al. 2012). The age-standardised incidence rate of breast cancer was 155 per 100,000 women in 2008 (National Board of Health 2009).
Study design and exclusion criteria
This observational register-based historical cohort study was based on data from women invited to the first organised breast cancer screening programme in the Central Denmark Region. We studied factors associated with non-participation in breast cancer screening among women with a previous cancer diagnosis (excluding breast cancer). Previous cancer was defined as a diagnosis of cancer recorded in the Danish Cancer Register (DCR) prior to the scheduled screening date (excluding non-melanoma skin cancer). The DCR holds information on all cancer cases in Denmark since 1943 (Gjerstorff 2011). We excluded 4646 women with breast cancer (ICD-10: C50) because many in this group were enrolled in a cancer follow-up programme and were specifically asked not to participate in the organised screening programme. A total of 6638 women were registered with a previous cancer between the age of 14 and their scheduled screening date. We excluded women who had died or moved between the invitation date and the booking date (n = 233) and women registered with a GP outside the caption area (n = 91). A total of 144,269 women were finally included in the analyses.
Registers and variables
All contacts to the Danish health care system are registered in national registers, and all Danes are assigned a unique and permanent 10-digit personal Civil Registration Number (CRN) (Pedersen 2011). The CRN was used to link data on screening participation, cancer diagnosis, treatments, comorbidity and SEP.
The outcome was defined as participation in the first organised breast cancer screening round, and eligible women were categorised into “non-participation” and”participation” on the basis of registrations in an administrative database.
The following variables were defined based on data from the Danish National Patient Register (NPR) with information on all hospital contacts (Lynge et al. 2011): Previous cancer patients were categorised as undergoing “current cancer treatment” if registered with chemotherapy or radiotherapy in the NPR within 6 months prior the scheduled screening date. They were also considered to undergo treatment if registered with any operation related to cancer within 3 months prior the scheduled screening date (i.e. the operation was performed based on a “DC” ICD-10 diagnosis code), except for endoscopies or biopsies as the majority of these procedures are used for diagnostic purposes. “Attending a cancer follow-up programme” was defined based on registration in the NPR with a procedure code for cancer follow-up (SKS: DZ08 and all sub-codes) up to 2 years prior to the scheduled screening date among previous cancer patients. Previous cancer patients were considered to be in “palliative care” if registered with previous cancer and listed as dead in the Civil Registration System within 1 year after the scheduled date of screening. A total of 54 women were registered with “undergoing treatment”, “attending cancer follow-up” and “palliative care”. “Time from diagnosis to the scheduled screening date” was categorised into: 0–1 year, > 1–5 years, > 5–10 years, > 10 years.
Cancer types were stratified into gynaecological cancer (ICD-10: C51-C58), colorectal cancer (ICD-10: C18-C20), lung cancer (ICD-10: C34), haematological cancer (ICD-10: C81-C96), malignant melanoma (ICD-10: C43) and other cancer types (all other ICD-10: “DC” codes).
SEP was included using the following variables: age (divided into 50–54 years, 55–59 years, 60–64 years and 65–69 years), ethnicity (divided into Danish decedents and immigrants), marital status (divided into married/cohabitating or living alone) and education [divided into ≤ 10 years, 11–15 years or > 15 years of education in accordance with UNESCO’s classification (UNESCO 2014)]. Charlson Comorbidity Index (CCI) scores (Quan et al. 2011) were obtained from hospital contacts from the NPR of the included diseases (except for cancer) 10 years prior to the scheduled screening date and categorised into the scores 0, 1 and ≥ 2.
Generalised linear models (GLM) with log link and Bernoulli regression models (Barros and Hirakata 2003; Zou 2004) were applied to study possible factors of importance for non-participation in breast cancer screening among previous cancer patients. Prevalence rate ratios (PRR) with 95% confidence intervals (CI) were chosen since the proportion of non-participation (the outcome) was more than 20% (Barros and Hirakata 2003; Zou 2004). To study if the increased likelihood of non-participation was explained by SEP, CCI or the severity of the previous cancer, we conducted analyses including the entire population (n = 144,264). Unadjusted analyses were performed to study if each variable independently was associated with non-participation. Furthermore, adjustment for socio-demographic variables and comorbidity was performed (model 1). Finally, a fully adjusted model was performed (model 2). We found a high degree of multicollinearity between the variables “previous cancer” and “time since diagnosis”. Therefore, we conducted a restricted analysis among previous cancer patients (n = 6638) to explore if “time since diagnosis” was associated with non-participation. Finally, to study non-participation among specific cancer groups, we restricted the analyses to previous cancer patients and excluded women undergoing current treatment and women in palliative care (n = 5777). Robust variance estimates were used to adjust for clustering of patients by general practice in all models (Davis 2001). All statistical analyses were conducted using Stata statistical software, version 14.
No ethical approval was required according to Danish law and the National Committee on Health Research Ethics in the Central Denmark Region as the study was based on registry data (journal no. 181/2011). The project was approved by the Danish Data Protection Agency (journal no.: 2009-41-3471 and journal no.: 1-16-02-109-09).