For this nested case–control study, we used data of the Achmea Health Database in the Netherlands, which is a healthcare claim database covering approximately 1.2 million subjects (8 % of the Dutch population). The database contains anonymized data on demographic characteristics, reimbursed diagnostic-related groups (DRGs), and medication. The population insured by the Achmea Health Insurance Company represents the urbanized area of the Netherlands with regard to age, gender, and socioeconomic status .
DRGs were introduced in the Netherlands in 2006 and are based on the International Classification of Disease, 9th revision (ICD-9). They are reimbursed per episode of care provided by secondary care physicians for inpatient and outpatient hospital care services. Data on DRGs were available between January 2006 and December 2011 and contained information on the colorectal cancer diagnosis and date of DRG registration, which usually is the first visit to the physician but can also be a follow-up visit.
Data on reimbursed medication were available between January 2001 and December 2011 and contained information on type of drug (ATC codes), date the drug was filled, number of daily defined doses (DDDs), the prescribed daily dose (PDD), and the prescribing physician, i.e., primary or secondary care. The DDD is the average maintenance dose per day for a drug used for its main indication in adults and is defined by the WHO Collaborating Centre for Drug Statistics Methodology . The PDD is the fraction of DDD per day that is actually prescribed by the treating physician. In the Netherlands, antibiotics can only be obtained with a prescription of a physician and these prescriptions are registered in the Achmea Health Database for subjects insured with this insurance company. However, medication prescriptions during hospitalizations are not registered in the database.
This study was approved by the scientific and privacy committee of the Achmea Health Insurance Company and was performed in accordance with the ethical guidelines of our institute.
The complete database was searched for adult (≥18 years) subjects with a DRG for CRC between January 2006 and December 2011. An incidence CRC case was defined as a subject with at least two DRGs for CRC or one DRG for CRC surgery, in which the first DRG was not registered within the first 1.5 years of follow-up. This 1.5-year clean period was chosen to minimize the risk of including prevalent CRC cases and was based on the recommended follow-up of patients with CRC every 6–12 months until 5 years after initial treatment with curative intent or more frequently in case of palliative treatment (according to the national guidelines at that time [24, 25]). The date of CRC diagnosis was defined as the date of first DRG registration. Each case was matched with regard to sex and date of birth to four randomly selected controls without a DRG for CRC and with at least the same period of follow-up as their matched case. Both cases and controls were required to have at least 6 years of complete follow-up before CRC diagnosis. Cases and controls that at some point during follow-up had a DRG for inflammatory bowel disease were excluded.
Antibiotics included were tetracyclines (ATC codes J01A), amphenicols (ATC codes J01B), penicillins (ATC codes J01C), cephalosporins (ATC codes J01D), sulfonamides and trimethoprim (ATC codes J01E), macrolides (ATC codes J01F), aminoglycosides (ATC codes J01G), quinolones (ATC codes J01M), imidazoles (ATC codes J01XD), nitrofuran derivates (ATC codes J01XE), and others (ATC codes J01XA, J01XB, J01XC, J01XX). The number of days for which antibiotics were prescribed was calculated as prescribed days = DDD/PDD. For prescriptions with an unknown DDD (3.6 %) or PDD (7.7 %), values were imputed with SAS PROC MI procedure, under the missing at random assumption and based on ATC code, primary or secondary care prescribing physician, sex, and age. The use of antibiotics was measured as the number of prescriptions and the prescribed number of days during a 5-year period in the period 1–6 years prior to CRC diagnosis. Subjects were categorized as nonusers and very low (1st–50th percentile), low (51st–75th percentile), intermediate (76th–90th percentile), and high (above 90th percentile) users of antibiotics. For the analyses of anti-anaerobic agents and subtypes of antibiotics, we categorized subjects as nonusers and low (1st–75th percentile), intermediate (76th–90th percentile), and high (above 90th percentile) users.
Covariates included in this study were sex, age (continuous), insulin-dependent diabetes (ATC codes A10A, no/yes), insulin-independent diabetes (ATC codes A10B, no/yes), and the use of proton pump inhibitors (ATC codes A02BC), acetylsalicylic acids (ATC codes B01AC06 and B01AC08), nonsteroidal anti-inflammatory drugs (ATC codes M01A), lipid-lowering agents (ATC codes C10AA, C10BA, and C10BX), estrogens (ATC codes G03AA, G03AB, G03FA, and G03FB), and immunosuppressive drugs (ATC codes L04A). The cumulative number of prescribed days per drug was categorized as none and the 1st–75th, 76th–90th, and above 90th percentile within users.
The use of antibiotics and comedication in cases and matched controls was assessed over a 5-year period between 1 and 6 years prior to CRC diagnosis. Medication use within 1 year prior to CRC diagnosis was not included in the analysis to minimize the risk of reversed causation. Differences in baseline characteristics were compared between cases and controls and expressed in means ± standard deviations (SDs), medians (interquartile range—IQR), and frequencies whenever applicable. Student’s t test and Mann–Whitney U test were used for continuous variables and Pearson Chi-square test for categorical variables.
Univariable and multivariable binary logistic regression analyses, conditioned on the matching factors sex and date of birth, were used to calculate the odds ratio (OR) and 95 % confidence intervals (95 % CIs) for the use of antibiotics on a categorical and continuous scale and the risk of developing CRC. Linear trends over different categories were computed using median levels of antibiotic use within the categories in all subjects. Two models were tested: (1) a univariable conditioned model on the matching factors age and sex and (2) a multivariable model conditioned on these matching factors and adjusted for factors statistically significantly associated with the outcome or the use of antibiotics.
Effect modification between the use of antibiotics and other factors that may affect the gut microbiota or that have previously been found to be associated with CRC risk was tested by adding multiplicative interaction terms to the model and using likelihood ratio tests for interaction. For these analyses, we used the interaction terms of overall antibiotic use (number of prescriptions, categorical) with insulin-independent diabetes (no/yes), insulin-dependent diabetes (no/yes), proton pump inhibitors (no/yes), acetylsalicylic acids (no/yes), nonsteroidal anti-inflammatory drugs (no/yes), statins (no/yes), estrogens (no/yes), and immunosuppressive drugs (no/yes).
Sensitivity analyses were performed to study possible reversed causation by assessing the use of antibiotics between 2–7 years and 3–8 years prior to CRC diagnosis.
Statistical analyses were conducted with SAS 9.2 (SAS Institute Inc, Cary, USA). Two-sided p values <0.05 were considered statistically significant.