Data sources
For this population-based case-control study, we used data from the Netherlands Cancer Registry (NCR) linked to the PHARMO GP Database (the NCR-PHARMO GP cohort). This cohort covers a catchment area of approximately 4 million inhabitants (approximately 20-25% of the Dutch population). The GP Database comprises data from electronic patient records registered by GPs including information on diagnoses and symptoms (coded according to the International Classification of Primary Care (ICPC) or entered as free text) and healthcare product/drug prescriptions (coded according to the World Health Organization (WHO) Anatomical Therapeutic Chemical (ATC) Classification System)(7).
The NCR is a population-based registry which is maintained by the Comprehensive Cancer Centre the Netherlands (IKNL) and comprises information on newly diagnosed cancer patients in the Netherlands. The NCR is notified for new patients with cancer by pathology departments, general hospitals, and radiotherapy institutes.
Further detailed information on the linkage and formation of the NCR-PHARMO GP cohort can be found elsewhere(8) (9).
Study population
All patients who were diagnosed with primary colorectal cancer (International Classification of Diseases, Tenth Revisions, Clinical Modification (ICD 10-CM) code C18-C20) between January 1st 2007 and December 31st 2014 were selected. Patients with a previous diagnosis of cancer (except basal cell skin carcinoma) were excluded. The same pertained to patients with less than 12 months of history available in the GP Database (defined as the time between the date of entering the PHARMO Database Network to the date of colorectal cancer diagnosis).
Each colorectal cancer case was matched to four cancer-free controls based on gender, birth year GP practice and start follow-up in the PHARMO Database Network. Matched controls received the same index date as the diagnosis date in their matched case with colorectal cancer and could not be matched more than once. The same exclusion criteria for the cases were applied to the matched controls.
Primary healthcare use
Information on primary healthcare use was derived from the GP Database of the PHARMO Database Network, which includes primary care data retrieved directly from the source, i.e. the electronic medical records of the healthcare providers.
According to the Medical Treatment Contract Act (“Wet op de geneeskundige behandelingsovereenkomst” (WGBO)), care providers (including general practitioners) are obliged to create and maintain a complete patient file for each patient. In the 1980s, the first automated system for general practices was introduced, replacing the handwritten files which were often difficult to read and incomplete. Since 2013, almost all general practices in the Netherlands work with an automated system to record the medical data of patients in the Electronic Patient File (EPD). In this EPD, GP consultations are grouped in episodes, i.e. a series of consultations related to a single reason for encounter (a symptom or a diagnosis). Besides information on the reason for encounter, prescriptions, laboratory results, referral letters and the summary of specialist letters are also registered in the system.
For this study, all GP consultations (face-to-face consultations, GP home visits and phone consultations) were extracted in the year prior index date for all colorectal cancer cases and their matched cancer-free controls. All information on diagnoses and symptoms registered by the GP (coded or entered as free text) was used for analyses. Furthermore, all prescribed medication based accompanied with an ATC codes in the year prior index date was extracted.
Statistical analyses
Descriptive statistics were used to present baseline and tumour characteristics. The mean monthly number of GP consultations were calculated by dividing the monthly number of GP consultations by the number of colorectal cancer cases (or cancer-free controls) in each month. The Wilcoxon rank-sum test was used for comparing the mean monthly number of GP consultations between colorectal cancer cases and cancer-free controls. A p-value of <0.05 was considered as a statistically significant difference.
Furthermore, in each month before the index date - going back 12 months - the number of colorectal cancer (or cancer-free controls) receiving a newly prescribed drug was assessed. Newly prescribed drugs were assessed based on the fourth level of the ATC code (i.e. A02BC) and defined as not receiving the drug in the year prior to that period. In each month, the incidence rate of new drug users was calculated by dividing the number of new drugs users by follow-up among colorectal cancer (or cancer-free controls) and presented per 1,000 persons per month. A Poisson regression analysis was used to examine whether the incidence rates per month significantly varied between colorectal cancer cases and cancer-free controls. A p-value of <0.05 was considered as a statistically significant difference.
A post-hoc analysis was performed to assess the type of newly prescribed medication and reason for GP consultation that triggered the increase in the last 6 months before index date among colorectal cancer cases. The type of newly prescribed drugs was assessed on the fourth level of the ATC code and presented if the absolute difference in receiving a specific drug between colorectal cancer cases and cancer-free controls was more than 5%. The reason for a GP consultation was determined by assessing the ICPC code associated with each consultation and by reviewing the free text of diagnosis and symptoms not accompanied by an ICPC code. Diagnoses and symptoms without an ICPC code (i.e. entered as free text) were reviewed and supplemented with an ICPC code were applicable. The reason for a GP consultation were separately listed based on ICPC codes if the absolute difference in the occurrence of a specific diagnosis or symptom between colorectal cancer cases and cancer-free controls was more than 2%.
As clinical features of colorectal cancer may vary significantly depending on the anatomical site, results were stratified by anatomical colorectal cancer site: proximal colon cancer, distal colon cancer and rectal cancer. Patients with colon cancer with an unspecified site or rectosigmoid cancer were not taken into account.
All data was analysed using SAS programs organized within SAS Enterprise Guide version 7.1 (SAS Institute Inc., Cary, NC, USA) and conducted under Windows using SAS version 9.4.