Antidepressant use and colorectal cancer risk: a Danish population-based case–control study

Background: Earlier research suggests that use of selective serotonin reuptake inhibitors (SSRIs), but not tricyclic antidepressants (TCAs), reduces the risk of colorectal cancer (CRC). Methods: We conducted a population-based case–control study to investigate the association between antidepressant use and CRC risk. Cases were diagnosed with a first primary CRC from 1991 through 2008. We selected 10 population controls matched to cases on sex, birth year, and residence from the Danish Civil Registration System using risk-set sampling. We estimated the odds ratios (ORs) and 95% confidence intervals (CIs) associating antidepressant use with colorectal cancer occurrence, controlling for potential confounders. Results: The study included 9979 cases and 99 790 controls. We found no notable reduction in CRC risk in ever users (⩾2 prescriptions) of TCAs (OR=0.94; 95% CI: 0.84, 1.05), SSRIs (OR=0.97; 95% CI: 0.90, 1.05), or other antidepressants (OR=0.95; 95% CI: 0.83, 1.07). Associations for recent and former use of antidepressants were also near null. Intensity of antidepressant use (number of pills divided by total duration of use), regardless of duration, was not associated with CRC risk. Conclusions: We found no evidence that antidepressant use substantially reduces the risk of colorectal cancer.

Antidepressants are used to treat depression, anxiety, and pain (Alonso et al, 2004;Paulose-Ram et al, 2007). Most western countries report at least a doubling in the prevalence of antidepressant use over the last two decades (McManus et al, 2000;Paulose-Ram et al, 2007). Studies on the safety of antidepressants, including beneficial side effects, are important.
Given the high prevalence of antidepressant use and CRC in western countries (Parkin et al, 2005), any association between them would have significant public health implications. We therefore conducted a large population-based case -control study using prospectively collected prescription data to examine whether antidepressant use is associated with reduced risk of CRC.

PATIENTS AND METHODS
We conducted this population-based case -control study among the residents of Northern Denmark (the former counties of North Jutland and Aarhus), where approximately 20% (1.1 million inhabitants) of the Danish population lives. The Danish National Health Service provides tax-supported health care to all residents of the country and refunds part of patient expenditures for most prescribed drugs, including antidepressants.
All health-related services are linked to individual patients through their civil personal registration (CPR) number, assigned by the Danish Civil Registration System to all Danish residents since 1968. The CPR number encodes gender and date of birth and facilitates accurate linkage between Danish registries, including the Danish National Patient Registry (DNPR) and prescription databases (Gaist et al, 1997;Sørensen et al, 2009).
We used the DNPR to identify all patients aged at least 35 years whose first discharge diagnosis of CRC ; ICD-10: C18.0-C18.9, C19.9, C20, and C20.9) occurred between 1 January 1991 and 31 December 2008 in North Jutland County or between 1 January 1998 and 31 December 2008 in Aarhus County. The DNPR contains detailed individual-level data on all non-psychiatric hospital admissions since 1977 and on all outpatient hospital contacts since 1995. Immediately after discharge from an inpatient hospitalisation or an outpatient clinic, the DNPR records the patient's CPR number, dates of admission and discharge, and up to 20 discharge diagnoses (Andersen et al, 1999). To ensure availability of at least 2 years of prescription data for each case, we included only CRC patients who resided in the study area for at least 2 years before their diagnosis date, yielding a total of 9979 CRC cases. The CRC diagnoses recorded in the DNPR before 1991 for North Jutland and before 1998 for Aarhus County were excluded from the study.
To identify controls, we accessed the Danish Civil Registration System, which maintains electronic records of all changes in vital status, migration, and residential address among Danish residents (Frank, 2000;Pedersen et al, 2006). We selected 10 unique population controls for each case, matched to cases on birth year, gender, and residence (county through 2006 and region after 2006; Sørensen et al, 2009), using risk-set sampling (Rothman et al, 2008). Risk-set sampling requires that controls be alive and at risk of a first hospital admission for CRC at the time the corresponding case was diagnosed. We assigned this date as the index date to both the case and all of the case's matched controls. We identified a total of 99 790 population controls, sampling only among individuals who were residents of the study area for at least 2 years before the index date.
We excluded antidepressant drug use in the year before CRC diagnosis, to reduce the potential effect of undiagnosed disease on drug use. We measured drug intensity in terms of the median number of pills per day prescribed for each patient. We calculated all other exposures (ever/never use and temporality) on the basis of the total number of prescriptions filled during the exposure period. Accordingly, we defined 'ever users' as persons with at least two prescriptions during the entire exposure period and we defined 'never/rare users' as those with fewer than two prescriptions during the entire exposure period. To compare our drug exposure assessment with that used in earlier studies (Coogan et al, 2009;Haukka et al, 2010;Xu et al, 2006), we conducted a sensitivity analysis categorising ever users of antidepressants as persons with at least one antidepressant prescription.
We examined temporality of drug use by dividing 'ever users' into recent and former users. We defined recent users as those who had at least two prescriptions filled in the period 1 to 2 years before the index date and former users as those who had fewer than two prescriptions filled 1 to 2 years before the index date but at least two prescriptions filled during the entire observation period.
We also examined whether intensity of drug use (number of pills divided by the total duration of use) was associated with CRC risk. Duration was defined as the number of days from the date of a first prescription to the date of a last prescription plus the duration of the last prescription. We calculated the duration of the last prescription based on the prescription date and total number of pills dispensed. We divided duration of drug use into short term (1 year to o2 years), medium term (2 to 5 years) and long term (at least 5 years). We multiplied intensity of drug use by 100 to enable expression as a percentage. Categorisation of the intensity of drug use differed by antidepressant type. We grouped intensity of TCA use as low (o100%), medium (101 -170%), and high (4171%); intensity of SSRI use as low (o70%), medium (71-100%), and high (4101%); and intensity of other antidepressant use as low (o100%) and high (4101%). We based the intensity groups on frequency of pills for each type of antidepressant. We further divided TCAs into genotoxic and non-genotoxic categories (van Schaik and Graf, 1991;van Schaik and Graf, 1993).
For potential confounding drugs, we categorised use as never/ rare, recent (1 to 2 years before diagnosis/index date), and former use (at least 2 years before the diagnosis/index date). The definition of never/rare use varied by medication. For NSAIDs, statins, neuroleptics, and hormone replacement therapy, never/ rare use was considered less than two prescriptions throughout the observation period. For aspirin, less than 100 pills were considered never/rare use. We classified selective Cox-2 inhibitors and nonaspirin NSAIDs together (ATC codes: M01A*) in a group called 'NSAIDs,' and put aspirin (ATC codes: B01AC06, N02BA01, and N02BA51) into a separate group. For disulfiram, we considered only never/ever use.

Statistical analyses
We calculated the frequency and proportion of cases and controls in categories of demographic variables, antidepressants use, potentially confounding drugs, and potentially confounding diseases. In all analyses, we used conditional logistic regression to compute odds ratios (OR) and associated 95% confidence intervals (95% CI) adjusted for the confounders discussed above. We used never or rare use (less than two prescriptions in total) as the reference group.
In our sensitivity analyses, ever users were defined as those with at least one antidepressant prescription. We carried out further analyses to investigate the association of continuous use of antidepressants with risk of CRC. We defined continuous use as individuals who were prescribed at least two antidepressants each year from the year of their first prescription to their last prescription. Finally, we conducted separate analyses for the risk of colon cancer and the risk of rectal cancer. Given the risk set sampling of controls, the ORs provide estimates of the corresponding incidence rate ratios in the underlying population. Analyses were performed using SAS version 9.13 (SAS Institute Inc., Cary, NC, USA).

RESULTS
Characteristics of the 9979 cases and 99 790 population controls are presented in Table 1. A slightly higher proportion of cases than controls had a history of diabetes and alcoholism. A lower proportion of cases than controls were recent users of NSAIDs, statins, or HRT, or were ever users of neuroleptics. Tricyclic antidepressants were used by 3.7% of cases and 4.0% of controls, SSRIs by 8.5% of cases and 8.9% of controls, and other antidepressants by 2.9% of cases and 3.1% of controls (Table 2). Median age of the study population was 72.4 years.
Ever use of TCAs, SSRIs, or other antidepressants was not notably associated with risk of CRC (OR ¼ 0.94 (95% CI: 0.84, 1.05); OR ¼ 0.97, (95% CI: 0.90, 1.05); and OR ¼ 0.95 (95% CI: 0.83, 1.07), respectively; Table 2). Associations for recent and former use were also near the null. Analyses combining duration and intensity of use also generated odds ratios near the null, regardless of antidepressant type (Table 3). As some individuals may have used drugs from all three classes, we repeated the analysis with a single model that included variables representing exposure to each drug group. There was little difference in the odds ratios generated by this model (data not presented). Individuals may also have switched from one antidepressant drug type to another. We therefore carried out additional analyses including individuals who used SSRI only, TCA only and both drug types in a model. There was little change in the odds ratios generated by this model (data not presented).
There was a slightly reduced risk of CRC associated with ever use of non-genotoxic TCAs (OR ¼ 0.88, 95% CI: 0.75, 1.04). Risk of CRC was not associated with ever use of genotoxic TCAs (OR ¼ 0.99, 95% CI: 0.83, 1.18) or use of both types of TCAs (OR ¼ 1.00, 95% CI: 0.75, 1.33). Risk estimates by type of TCA were all centred at the null for intensity, recency, and duration of use.
We found a slightly reduced risk of CRC associated with use of TCAs (OR ¼ 0.88, 95% CI: 0.73, 1.06) and SSRIs (OR ¼ 0.88, 95% CI: 0.75, 1.02) in our sensitivity analyses, with ever users defined as individuals with at least one antidepressant prescription.
We saw no evidence of an association between continuous use of antidepressants and CRC risk (data not presented).

DISCUSSION
Our findings show little evidence of a protective association between antidepressant use and CRC risk, regardless of antidepressant type. The odds ratios were near null for both recent and former use, and varied little by intensity and duration of use.
The validity of our estimates depends on several factors. We identified our study population from continuously updated population-based registries, with complete follow-up and high quality data. For example, cancer diagnoses in the DNPR have high sensitivity and specificity Tetsche et al, 2005). Use of a population-based prescription registry, whose completeness approaches 100% (Gaist et al, 1997), ensured unbiased assessment of exposure before CRC diagnosis and eliminated recall bias. This design also facilitated adjustment for potential confounding drugs, diseases, and a more comprehensive list of potential confounders than examined in some earlier studies. The design also allowed assessment of a dose-response effect for antidepressant exposure in terms of intensity and duration of drug use.
Although our study was immune to recall bias by design, nondifferential misclassification of antidepressant use was still an Low-and high-dose aspirin were grouped together.
Antidepressants and colorectal cancer DP Cronin-Fenton et al important limitation. Patients with depression are less likely to comply with prescription medications than those without depression (DiMatteo et al, 2000). However, our drug exposure assessment was based on redeemed prescriptions, for which patients had to collect the prescription and pay a portion of the drug cost. Our estimates are therefore likely to reflect actual use. The SSRIs were introduced to the Danish market about 1987 (Olfson et al, 2002). The prescription registries were established in 1989 and 1996, so we were unable to identify individuals who may have switched to SSRIs from TCAs before the study period. Although we controlled for many confounders, we were unable to adjust for body mass index, physical activity, or diet. We also had no information on the indication for antidepressant use or the severity of the underlying disease. Such confounding by indication (i.e., severity of depression) would underestimate the impact of antidepressant use on CRC risk. Of note, a large cohort study found an increased risk of CRC among women with the highest levels of depressive symptoms (Kroenke et al, 2005). Our almost null results agree with the Finnish cohort study, which also found no reduction in CRC risk associated with SSRI use (Haukka et al, 2010). The Finnish study, however, did report a weak association between increased risk of CRC and 44 years of TCA use, which we did not observe. The Canadian also observed a slight increase in risk of CRC with long-term use of TCA (16-20 years; Coogan et al, 2009;Xu et al, 2006). We did not have such long-term data. In addition, our results are not so inconsistent with the Canadian and US studies, both of which reported a small reduction in the risk of CRC associated with SSRI use (Coogan et al, 2009;Xu et al, 2006), but their CIs were quite broad and included the null.
Our previous experience with the prescription data indicated quite a high prevalence of persons who fill just one antidepressant prescription, which suggests that these persons may not have completed taking even their first prescription. Therefore, to reduce misclassification from this non-compliance, we defined ever users as persons who had filled at least two prescriptions for antidepressants. In a sensitivity analyses, we defined ever use as redemption of one or more prescriptions instead of two or more prescriptions and observed a slight decrease in associations between CRC occurrence and antidepressant use more comparable with the results of the earlier case -control studies (Coogan et al, 2009;Xu et al, 2006). This sensitivity analysis suggests that the definition of ever use (one or more prescriptions) in the earlier studies contributed to the decreased risk of CRC associated with antidepressant use.
We found no evidence of an association between use of antidepressants and risk of colorectal cancer in this large population-based study.  Abbreviations: CI ¼ confidence interval; NSAIDs ¼ non-steroidal anti-inflammatory drugs; OR ¼ odds ratio. a Adjusted for diabetes, alcoholism, inflammatory bowel disease, use of NSAIDs, use of statins, use of hormone replacement therapy, use of low-dose aspirin and high-dose aspirin, use of neuroleptics, and marital status.