ORIGINAL ARTICLE
Drug Exposure and Risk of Microscopic Colitis: A Nationwide Danish Case–Control Study with 5751 Cases Ole K. Bonderup, MD, PhD,* Morten Fenger-Grøn, MSc,†,‡ Tatjana Wigh, MD,§ Lars Pedersen, MSc, PhD,‡ and Gunnar L. Nielsen, MDk,¶
Background: Previous small studies have indicated that commonly prescribed drugs may be associated with microscopic colitis (MC). With an increasing incidence of MC, it is important to explore the association between exposure to proton pump inhibitors (PPIs), nonsteroidal anti-inflammatory drugs, statins, and selective serotonin reuptake inhibitors and MC in a larger setting.
Methods: Case–control study based on nationwide Danish registries. The study included all patients with MC diagnosis during the period 2005 to 2011. One hundred sex- and age-matched controls per case were randomly selected among the Danish population. Prescriptions were recorded in a Prescription Register in the year before the first recorded MC diagnosis. Effect measure is the adjusted odds ratio (OR) of collagenous colitis (CC) and lymphocytic colitis (LC) according to prescriptions of PPIs, nonsteroidal anti-inflammatory drugs, statins, and selective serotonin reuptake inhibitors. Within the control group, we identified a subgroup with MC-free colonic biopsies.
Results: We identified 3474 patients with CC and 2277 with LC and found a positive association between redemption of prescriptions for PPIs and both CC (OR ¼ 7.04; 95% confidence interval, 6.55–7.56) and LC (OR ¼ 3.37; 95% confidence interval, 3.08–3.69). Among patients with MC-free colon biopsies in the control group, the association between PPIs and CC was strongly positive (OR ¼ 3.47; 95% confidence interval, 3.08–3.89). Adding this parameter to the model attenuated all of the associations, but they remained positive for PPIs versus CC and selective serotonin reuptake inhibitors versus LC.
Conclusions: We found positive associations between exposure to all 4 medication classes and MC. Variations in endoscopic frequency by drug category indicate a potential impact of bias. (Inflamm Bowel Dis 2014;20:1702–1707) Key Words: diarrhea, drugs, microscopic colitis, collagenous colitis, lymphocytic colitis
M
icroscopic colitis (MC) is an umbrella term for chronic bowel diseases characterized by chronic diarrhea in patients with a normal or near-normal mucosa at colonoscopy. MC comprises 2 main entities, lymphocytic colitis (LC) and collagenous colitis (CC), which are clinically indistinguishable.1 The 2 subtypes share histopathological features characterized by a marked increase in intraepithelial lymphocytes; CC is additionally characterized by a thickened subepithelial collagen band. Therefore, LC and CC are considered to be 2 related but separate entities.2 Epidemiological studies have shown that MC has been more frequently diagnosed in recent decades, with an incidence rate comparable with other inflammatory bowel diseases.3,4 In a U.S. Received for publication May 18, 2014; Accepted June 12, 2014. From the *Diagnostic Centre, Regional Hospital Silkeborg, Silkeborg and University Research Clinic for Innovative Patient Pathways, Aarhus University, Silkeborg, Denmark; †Research Unit for General Practice, Aarhus University, Aarhus, Denmark; ‡ Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark; § Medical Department, Regional Hospital Randers, Randers, Denmark; kMedical Department, Hospital Himmerland, Denmark; and ¶Department of Clinical Medicine, The Faculty of Medicine, Aalborg University, Aalborg, Denmark. The authors have no conflicts of interest to disclose. Reprints: Ole K. Bonderup, MD, PhD, Diagnostic Centre, Regional Hospital Silkeborg, Falkevej 1, DK-8600 Silkeborg, Denmark (e-mail: [email protected]). Copyright © 2014 Crohn’s & Colitis Foundation of America, Inc. DOI 10.1097/MIB.0000000000000143 Published online 22 August 2014.
1702
| www.ibdjournal.org
study from 2007, the overall annual incidence of MC was reported to be 19.6/100,000 persons.4 Recent reports from 2 regions with long standing interest in this disease have found a more stable incidence of MC in the last decennium.5,6 The etiology of the disease is unknown. The diversion of the fecal stream in patients with CC with a temporary ileostomy leads to the resolution of histological changes, implicating luminal factors in the pathogenesis.7 A variety of such factors have been suspected, and drug consumption has been suggested to act as an environmental risk factor implicated as a causative or triggering agent of MC.8 Increased permeability of the colonic mucosa has been observed in patients with MC.9 Drugs may more easily affect the inflammatory response in the colon mucosa of a patient in whom the mucosal defense mechanism is already impaired. In combination with other luminal factors, this could induce the symptoms of MC. Proton pump inhibitors (PPIs), nonsteroidal anti-inflammatory drugs (NSAIDs), statins, and selective serotonin reuptake inhibitors (SSRIs) have all been proposed to be linked to MC.10 All these drugs have been more frequently prescribed in recent years and could therefore be linked to the increase in incidence of MC. However, most of these presumptive associations are based on case reports or small case–control studies, and in a recent study, no association between drug intake and the risk of MC were found.11 We therefore found it important to explore the potential association between exposure to PPIs, NSAIDs, statins, and Inflamm Bowel Dis Volume 20, Number 10, October 2014
Inflamm Bowel Dis Volume 20, Number 10, October 2014
SSRIs and the risk of MC in a nationwide case–control study based on Danish national registries.
Drug Exposure and Risk of MC
exposure and a higher frequency of MC-free colonic biopsies was assessed in a case–control design within the control population.
Patients with MC MATERIALS AND METHODS Data Sources This registry study was based on a analysis of the following datasets covering the entire Danish population of approximately 5.4 million inhabitants. The Danish Civil Registration System was established in 1968, and since then, all Danes have been assigned a 10-digit personal identifier at birth or on immigration.12 This unique civil registry number, which encodes gender and date of birth, is used in virtually all Danish registries and thus permits unambiguous data linkage. The Danish National Patient Register was established in 1977 and includes civil registry number and in-patient discharge diagnoses. Since 1995, it has also included out-patient diagnoses. All of the diagnoses are coded by medical doctors according to the Danish versions of the International Classification of Diseases (ICD-8 from 1977 to 1993 and ICD-10 since 1994).13 The Danish Pathology Register was established in 1990, and in 1997, it became a legal obligation for all departments of pathology in Denmark to report their pathology data statements to this register. The diagnoses are coded after a Danish version of the Systematized Nomenclature of Medicine (SNOMED).14 The data are used by pathologists in daily diagnostic processes and encompass all pathology data in Denmark. The Danish Prescription Register was initiated in 1994 with electronic recording of all of the prescription drugs sold in Denmark.15 This registry retains key information on prescriptions for refundable drugs dispensed from all of the pharmacies in the country, including the civil registration number of the patient, type of prescribed drug according to the anatomical therapeutic chemical classification system16 and date of dispensing the drug. The National Health Service provides tax-supported health care for all inhabitants, guaranteeing free access to family doctors and hospitals, and refunding a variable proportion (usually 50%–75%) of the costs of drugs prescribed by physicians.17 Using this registry, a complete and time-reliable drug prescription history can be established for each individual.
Design The association between drug exposure and the risk of obtaining a diagnosis of CC or LC was studied in a case–control design with population-based controls who were randomly selected using incidence density sampling. All data were collected prospectively in the databases by time of drug exposure and at time of a diagnosis of MC. Our design is thus a historical cohort study taking advantage of the prospectively collected data allowing data analyses to be performed without the potential drawbacks of a retrospective design in terms of recall and information bias. The impact of bias due to a possible association between drug
Our case population consisted of all of the patients with a recorded diagnosis of either CC (SNOMED code: S62536) or LC (SNOMED code: S62533) in the Danish Pathology Register. In the analyses, the cohort was restricted to the period January 2005 to December 2011, as drug exposure data were only available after January 1, 2004. The date of the first recording of one of these diagnoses was defined as the case index date. We excluded patients having ever obtained one of the following diagnoses in the Danish National Patient Register: inflammatory bowel disease (ICD-10: K50 and K510–515), gastrointestinal cancer (ICD-10: C15–26), celiac disease (ICD-10: K900), radiation enteritis (ICD-10: K520 + K627) pancreatitis (ICD-10: K86), bacterial enteritis within the last 12 months before the date of recorded MC diagnosis (ICD-10: A00–A09), and 12 cases with an invalid civil registry number.
Controls One hundred persons per case were randomly selected as controls from the nationwide Danish Civil Registration System and matched by sex and date of birth within 2 weeks. Controls were assigned an index date identical to that of the corresponding cases. None of the controls could have previously diagnosed MC, and we excluded persons who met the abovementioned exclusion criteria. A control patient could, at a later time, be diagnosed with MC. Among the selected controls, we identified a subgroup characterized by MC-free colonic biopsies in the Danish Pathology Register.
Exposure to PPIs, NSAIDs, Statins, and SSRIs For both cases and controls, drug exposure was defined as the redemption of at least 1 prescription for one of the drugs in question recorded in the Danish Prescription Registry in a 1-year period before the index dates. We used the following anatomical therapeutic chemical codes: PPIs, A02BC; NSAIDs, M01A and M01B; statins, C10AA; and SSRIs, N06AB.
Statistical Analyses The case–control study was analyzed using a conditional logistic regression model, which is computationally similar to a stratified Cox regression. The estimated exposure odds ratio (OR) is thus an unbiased estimated of the relative risk.18 Matched case–control sets were used as the units of stratification, as this ensured the finest possible correction for gender, age, and calendar time. Exposure to each of the 4 specified types of drugs was considered both independently (referred to as the crude analyses) and simultaneously to provide estimates corrected for concomitant use of drugs from 1 or more of the other 3 groups (referred to as the adjusted analyses). For the second part of the study, estimates for the association between drug use and colonic biopsies in the control group were assessed using a conditional logistic regression model www.ibdjournal.org |
1703
Inflamm Bowel Dis Volume 20, Number 10, October 2014
Bonderup et al
stratified on matched sets as well. Assuming that the prevalence of undiagnosed MC among controls was relatively low, we interpreted these as estimates of the association between drug use and biopsy frequency among MC-free controls. We finally calculated the ratio between the OR of drug exposure on MC (case–control study) and the corresponding OR on biopsy frequency among MC-free controls (cross-sectional study) to obtain corrected estimates for the drug versus MC association. Calculation of 95% confidence intervals (CIs) was based on a standard normal approximation on the log scale while employing approximate independence between the observed effects of drug exposure on MC and on biopsy frequency among MC-free controls. We performed no formal correction for multiple testing, as the presented results are highly significant for the relevant discussion to focus on their clinical relevance and possible bias rather than their statistical significance. All of the calculations were performed using the statistical software package Stata 11.0.
Ethical Consideration The conduct of this study was approved by the Danish Data Protection Agency (Jr. no. 2010-41-4937). In Denmark, no approval by the Ethics Committee is needed to conduct registrybased studies, such as this study. Informed consent from participants was therefore not required.
RESULTS Case Population During the study period, we identified 6812 patients with the first-time recorded diagnosis of MC in the Danish Pathology Registry. A total of 958 persons were excluded due to 1 or more of the exclusion criteria previously listed (ICD-10: K50 and K510–515), n ¼ 287; gastrointestinal cancer (ICD-10: C15–26), n ¼ 184; celiac disease (ICD-10: K900), n ¼ 102; radiation enteritis (ICD-10: K520 + K627), n ¼ 61; pancreatitis (ICD-10: K86), n ¼ 54; bacterial enteritis within the last 12 months before the date of recorded MC diagnosis (ICD-10: A00–A09), n ¼ 313; and 12 cases with an invalid civil registry number. Furthermore, 101 patients with unclassified MC were excluded. The final case group thus consisted of a total of 5751 patients, including 3474 (60%) with CC and 2277 (40%) with LC. Demographic details, including age and gender, are shown in Table 1.
Drug Exposure and Risk of MC The percentages of cases and controls in each category, which redeemed at least 1 prescription for 1 of the 4 drugs in the year preceding the index date are shown in Table 1. The observed ORs for the associations between drug exposure in cases and controls are shown in Table 2. We found a strong and highly significant association between the redemption of prescriptions for PPIs and both CC (adjusted OR ¼ 7.04; 95% CI, 6.55– 7.56) and LC (adjusted OR ¼ 3.37; 95% CI, 3.08–3.69), whereas the association between SSRIs and LC (adjusted OR ¼ 2.89; 95%
1704
| www.ibdjournal.org
TABLE 1. Drug Exposure in Patients with Microscopic Colitis and Controls CC Cases No. Female gender, % Mean age at diagnose, yr PPI prescription in the preceding year, % NSAID prescription in the preceding year, % Statin prescription in the preceding year, % SSRI prescription in the preceding year, % None of the above, %
LC Controls
Cases
Controls
3474 75.5 67.4 55.7
14.2
2277 63.5 63.8 36.7
12.9
41.5
24.7
35.0
23.6
30.6
21.7
26.8
20.0
18.2
8.8
23.1
7.8
19.5
50.2
27.8
53.3
Descriptive characteristics of cases and controls and the percentage with redemption of a prescription of 1 of the 4 drugs in the year preceding the date of the MC diagnosis.
CI, 2.61–3.20) was slightly weaker, albeit highly significant. All of the other recorded associations were also statistically significant but with ORs well below 2. A subanalysis of drug exposure 90 days before diagnosis did not change the results.
Drug Exposure and MC-free Biopsies Among Controls Among the controls, we identified 2202 persons with MCfree colonic biopsies in the Danish Pathology Register. The
TABLE 2. Odds Ratios of Microscopic Colitis and Drug Exposure
CC No prescription PPI NSAID Statin SSRI LC No prescription PPI NSAID Statin SSRI
OR Crude
95% CI
OR Adjusteda
95% CI
1 (ref) 8.15 2.17 1.66 2.33
— 7.60–8.73 2.03–2.33 1.54–1.79 2.14–2.55
1 (ref) 7.04 1.61 1.29 1.59
— 6.55–7.56 1.50–1.72 1.20–1.39 1.45–1.74
1 (ref) 4.11 1.75 1.53 3.64
— 3.77–4.49 1.60–1.91 1.39–1.69 3.29–4.02
1 (ref) 3.37 1.43 1.25 2.89
— 3.08–3.69 1.31–1.56 1.13–1.38 2.61–3.20
Crude and adjusted ORs of CC and LC by the prescription of 1 of the 4 drugs within 1 year preceding a MC diagnosis. a Analyses adjusted for the prescription of 1 or more of the other 3 drugs.
Inflamm Bowel Dis Volume 20, Number 10, October 2014
Drug Exposure and Risk of MC
results regarding the analyses of this subgroup within the control cohort are shown in Table 3. In these analyses, we compared drug exposure in controls with MC-free colon biopsies to controls without colon biopsies. Redemption of a PPI prescription (adjusted OR ¼ 3.47; 95% CI, 3.08–3.89) in CC controls, and to a slighter degree for SSRIs (adjusted OR ¼ 1.34; 95% CI, 1.14–1.57), seemed to be significantly associated with a normal colonic biopsy in the Pathology Registry. There was no such association for exposure to NSAIDs or statins.
Analyses Adjusted for Increased Frequency of Endoscopy Incorporating these results from our control group in our final model attenuated most of the associations, but they remained significant in cases of PPIs and CC (adjusted OR ¼ 2.03; 95% CI, 1.77–2.33) and SSRIs and LC (adjusted OR ¼ 1.77; 95% CI, 1.42–2.21) (Table 4).
DISCUSSION In this large nationwide case–control study, we observed significant positive associations between exposure to each of the 4 drugs and a recorded diagnosis of MC. However, adjustment for more frequent endoscopic examinations attenuated all of the associations. Even so, significant positive associations remained between PPIs and CC and SSRIs and LC. The associations between MC and exposure to NSAIDs and statins were only weak. Among the controls, our findings of a positive association between drug exposure and having an MC-free biopsy indicate
TABLE 3. Odds Ratios of Having a Microscopic ColitisFree Biopsy and Drug Exposure
Controls for CC (N ¼ 1332) No prescription PPI NSAID Statin SSRI Controls for LC (N ¼ 870) No prescription PPI NSAID Statin SSRI
OR Crude
95% CI
OR Adjusteda
95% CI
1 (ref) 3.64 1.33 1.22 1.67
— 3.25–4.08 1.18–1.50 1.08–1.38 1.43–1.96
1 (ref) 3.47 1.12 1.06 1.34
— 3.08–3.89 0.99–1.26 0.94–1.21 1.14–1.57
1 (ref) 3.77 1.27 1.21 2.03
— 3.27–4.35 1.09–1.47 1.03–1.42 1.67–2.46
1 (ref) 3.57 1.06 1.03 1.63
— 3.08–4.13 0.91–1.23 0.88–1.21 1.34–1.98
Analyses within the control group: crude and adjusted ORs of having a normal colonic biopsy recorded in the Pathology Registry by the redemption of a prescription of 1 of the 4 drugs. a Analyses adjusted for the prescription of 1 or more of the other 3 drugs.
TABLE 4. Adjusted Odds Ratios of Microscopic Colitis and Drug Exposure
CC No prescription PPI NSAID Statin SSRI LC No prescription PPI NSAID Statin SSRI
OR Adjusteda
95% CI
OR Adjustedb
95% CI
1 (ref) 7.04 1.61 1.29 1.59
— 6.55–7.56 1.50–1.72 1.20–1.39 1.45–1.74
1 (ref) 2.03 1.43 1.22 1.19
— 1.77–2.33 1.25–1.65 1.05–1.41 0.99–1.43
1 (ref) 3.37 1.43 1.25 2.89
— 3.08–3.69 1.31–1.56 1.13–1.38 2.61–3.20
1 (ref) 0.95 1.35 1.21 1.77
— 0.80–1.12 1.13–1.61 1.01–1.46 1.42–2.21
ORs of CC and LC adjusted for both the prescription of 1 of the other 3 drugs and surveillance. a Analyses adjusted for the prescription of 1 or more of the other 3 drugs (from Table 2). b Analyses adjusted for the prescription of 1 of the 4 drugs divided by the adjusted ORs for the association between drug use and MC-free biopsy in the control group (from Table 3).
that studies regarding the association between drug use and MC are prone to bias because of different intensities of endoscopic workup in patients exposed to different drugs. We do not have a firm explanation for this increased biopsy frequency in drug users. But a variety of drugs can cause diarrhea leading to subsequent investigation including colonic biopsies. Moreover, PPIs will often constitute the initial empirical treatment in patients with nonspecific abdominal discomfort who may later on end up with a colonoscopy. Previous publications have reported associations between the use of several drugs and an increased risk of MC.8 However, only small numbers of cases have been included, and therefore, controversies still exist. Several studies have found a positive association between MC and PPIs.10 We found a significant association between exposure to PPIs and a diagnosis of CC, and we also observed a significant difference in this association between the 2 MC categories. This might indicate a possible etiological role for PPIs in CC and could also lend support to the notion of 2 separate MC entities. However, inferring causality based on associations found in an observational design must be performed with great caution. NSAID usage has been reported in 30% to 70% of patients with MC. Accordingly, we found NSAID exposure in 42% of patients with CC and 35% with LC in the year preceding the diagnosis. A causal relationship between NSAIDs and CC and LC has not been established, and we only found a weak association between NSAID exposure and MC. Our findings regarding SSRIs are in accordance with the results in a case–control study with 39 cases by Fernandes-Banares et al.10 They found a very strong association between the consumption of www.ibdjournal.org |
1705
Bonderup et al
SSRIs and MC, but the association was of the same magnitude when they looked at patients with chronic diarrhea. In a small study of 26 cases, Pascua et al compared the risk of MC with both randomly selected controls and patients with chronic diarrhea and found no association between PPIs, SSRIs, and statins and MC.11 There could be differences among different drugs in the same class. In this study, we do not have data to elucidate this further. It is, however, important to recognize that a substantial number of patients are diagnosed with MC without any form of drug exposure. In this study, 20% of patients with CC and 28% with LC had no recorded exposure to any of the 4 drugs in question. The main strengths of our population-based study are its large size, the uniformly organized health care system permitting analyses in a nationwide cohort, the use of contemporary matched controls, and the nearly uniform coverage of patients with a recorded diagnosis of MC throughout the country, which reduced the risk of selection bias. During the observation period, the number of patients with analyzed colonic specimens increased. This may reflect a high awareness of MC in Denmark. The reliability of the recorded diagnoses in the Pathology Registry has not been validated. However, the diagnostic criteria for MC were established in 1989, and in a previous study, interobserver and intraobserver agreement was very reliable for the diagnostic categories of MC.19 Our data are based on a pathology register with no available clinical data. Because the diagnosis of MC relies on a combination of symptoms and histological findings, there is a risk of overestimating the number of cases of MC, as the diarrhea may have ceased spontaneously in some patients. In this study, the assessment of drug exposure was based on a comprehensive nationwide prescription database collected independently of our outcome data. We thereby avoided the recall bias that often hampers studies based on interviews, questionnaires, or recordings from primary care.20 We defined drug exposure as having picked up at least 1 prescription for a drug in question. We have no data on the actual rate of drug consumption or the total amount of drug consumed. However, we assume that the likelihood of actual drug consumption is high, as the drugs were collected by the subjects and paid for at the pharmacies. During the study period, low-dose formulations of both PPIs and NSAIDs were available over the counter, but the total amount purchased in these formulations has been reported to be very low compared with prescribed drugs.21 Therefore, the impact of over the counter drugs was most likely minimal. Thus, we consider our estimation of drug exposure to be reliable. The time period between drug exposure and the possible development of MC is unknown. Because there could be a delay between drug exposure and the debut of symptoms, we selected time windows of both 90 and 360 days preceding the diagnosis. Analyses using these different time periods did not change our findings. Medication prescribed more than 12 months before the diagnosis of MC was not analyzed because we found it unlikely to affect our associations. To the best of our knowledge, this is the first nationwide study and by far the largest study in this field allowing for us to
1706
| www.ibdjournal.org
Inflamm Bowel Dis Volume 20, Number 10, October 2014
report risk estimates with very narrow CIs. However, one needs to consider whether an observed statistically significant association between drug therapy and a certain condition represents a clinically relevant effect. Strong statistical associations between drug exposure and MC do not necessarily imply that the drugs can be incriminated as etiologic factors. Our design was observational, and we lacked information on a range of factors that may confound the association between exposure to these 4 drugs and MC (e.g., other drugs, comorbidity not accounted for in the exclusion criteria, tobacco, alcohol, or lifestyle), which either individually or in combination could be unevenly distributed between cases and controls and thus affect our risk estimates. This may lead to a risk of identifying statistically significant associations that may either be clinically irrelevant or ascribed to uncontrolled confounding factors. The reported ORs should therefore be interpreted with caution, and we have arbitrarily restricted our conclusions to associations with an OR above 1.5 in adjusted analyses.22 Our findings of relatively strong associations between exposure to PPIs and SSRIs and MC-free colonic biopsies within the healthy control group indicate a substantial impact of bias due to more frequent endoscopic examinations with biopsies in users of at least these 2 drugs. Surprisingly, no such association was found in patients using NSAIDs, a drug that is generally regarded as a causative agent of abdominal discomfort.23 However, the associations between PPIs and CC and SSRIs and LC remained fairly strong even after adjustment for this bias, whereas the remaining associations were attenuated to a degree at which their clinical relevance is questionable. Our observations confirm previous reports of an association between exposure to PPIs, NSAIDs, statins, and SSRIs and the diagnosis of MC. The potential impact of an increased frequency of endoscopy in drug users should be included in future studies. Our findings of relatively strong associations between drug exposure and normal colon biopsies within the healthy control group indicate a substantial impact of bias because of more intense endoscopic frequency. This potential source of bias should be accounted for in future studies.
REFERENCES 1. Pardi DS, Kelly CP. Microscopic colitis. Gastroenterology. 2011;140: 1155–1165. 2. Jackson BK. Are collagenous colitis and lymphocytic colitis distinct syndromes? Dig Dis. 1995;13:301–311. 3. Olesen M, Eriksson S, Bohr J, et al. Microscopic colitis: a common diarrhoeal disease. An epidemiological study in Orebro, Sweden, 1993–1998. Gut. 2004;53:346–350. 4. Pardi DS, Loftus EV Jr, Smyrk TC, et al. The epidemiology of microscopic colitis: a population based study in Olmsted County, Minnesota. Gut. 2007;56:504–508. 5. Wickbom A, Bohr J, Eriksson S, et al. Stable incidence of collagenous colitis and lymphocytic colitis in Örebro, Sweden, 1999–2008: a continuous epidemiologic study. Inflamm Bowel Dis. 2013;19:2387–2393. 6. Gentile NM, Khanna S, Loftus EV Jr, et al. The epidemiology of microscopic colitis in Olmsted County from 2002 to 2012: a population based study. Clin Gastroenterol Hepatol. 2014;12:838–848. 7. Jarnerot G, Tysk C, Bohr J, et al. Collagenous colitis and fecal stream diversion. Gastroenterology. 1995;109:449–455.
Inflamm Bowel Dis Volume 20, Number 10, October 2014
8. Beaugerie L, Pardi DS. Review article: drug-induced microscopic colitis— proposal for a scoring system and review of the literature. Aliment Pharmacol Ther. 2005;22:277–284. 9. Munch A, Soderholm JD, Wallon C, et al. Dynamics of mucosal permeability and inflammation in collagenous colitis before, during, and after loop ileostomy. Gut. 2005;54:1126–1128. 10. Fernandez-Banares F, Esteve M, Espinos JC, et al. Drug consumption and the risk of microscopic colitis. Am J Gastroenterol. 2007;102:324–330. 11. Pascua MF, Kedia P, Weiner MG, et al. Microscopic colitis and medication use. Clin Med Insights Gastroenterol. 2010;2010:11–19. 12. Pedersen CB. The Danish Civil Registration System. Scand J Public Health. 2011;39:22–25. 13. Lynge E, Sandegaard JL, Rebolj M. The Danish National Patient Register. Scand J Public Health. 2011;39:30–33. 14. Bjerregaard B, Larsen OB. The Danish Pathology Register. Scand J Public Health. 2011;39:72–74. 15. Kildemoes HW, Sorensen HT, Hallas J. The Danish National Prescription Registry. Scand J Public Health. 2011;39:38–41.
Drug Exposure and Risk of MC
16. Capella D. Descriptive tools and analysis. WHO Reg Publ Eur Ser. 1993; 45:55–78. 17. Andersen JS, Olivarius NF, Krasnik A. The Danish National Health Service Register. Scand J Public Health. 2011;39:34–37. 18. Szklo M, Nieto MF. Epidemiology. Beyond the Basics. Gaitersberg, MD: Aspen Publishers Inc., 2000. 19. Limsui D, Pardi DS, Smyrk TC, et al. Observer variability in the histologic diagnosis of microscopic colitis. Inflamm Bowel Dis. 2009; 15:35–38. 20. Shapiro S. Case-control surveillance. In: Strom BL, ed. Pharmacoepidemiology. 2nd ed. Chichester, NH: John Wiley & Sons, 1994:301–322. 21. MEDSTAT.DK [online database]. Available at www.medstat.dk. Last modified June 7, 2014. 22. Grimes DA, Schulz KF. False alarms and pseudo-epidemics: the limitations of observational epidemiology. Obstet Gynecol. 2012;120:920–927. 23. Pirmohamed M, James S, Meakin S, et al. Adverse drug reactions as cause of admission to hospital: prospective analysis of 18 820 patients. BMJ. 2004;329:15–19.
www.ibdjournal.org |
1707
Drug Exposure and Risk of Microscopic Colitis: A Nationwide Danish Case–Control Study with 5751 Cases Ole K. Bonderup, MD, PhD,* Morten Fenger-Grøn, MSc,†,‡ Tatjana Wigh, MD,§ Lars Pedersen, MSc, PhD,‡ and Gunnar L. Nielsen, MDk,¶
Background: Previous small studies have indicated that commonly prescribed drugs may be associated with microscopic colitis (MC). With an increasing incidence of MC, it is important to explore the association between exposure to proton pump inhibitors (PPIs), nonsteroidal anti-inflammatory drugs, statins, and selective serotonin reuptake inhibitors and MC in a larger setting.
Methods: Case–control study based on nationwide Danish registries. The study included all patients with MC diagnosis during the period 2005 to 2011. One hundred sex- and age-matched controls per case were randomly selected among the Danish population. Prescriptions were recorded in a Prescription Register in the year before the first recorded MC diagnosis. Effect measure is the adjusted odds ratio (OR) of collagenous colitis (CC) and lymphocytic colitis (LC) according to prescriptions of PPIs, nonsteroidal anti-inflammatory drugs, statins, and selective serotonin reuptake inhibitors. Within the control group, we identified a subgroup with MC-free colonic biopsies.
Results: We identified 3474 patients with CC and 2277 with LC and found a positive association between redemption of prescriptions for PPIs and both CC (OR ¼ 7.04; 95% confidence interval, 6.55–7.56) and LC (OR ¼ 3.37; 95% confidence interval, 3.08–3.69). Among patients with MC-free colon biopsies in the control group, the association between PPIs and CC was strongly positive (OR ¼ 3.47; 95% confidence interval, 3.08–3.89). Adding this parameter to the model attenuated all of the associations, but they remained positive for PPIs versus CC and selective serotonin reuptake inhibitors versus LC.
Conclusions: We found positive associations between exposure to all 4 medication classes and MC. Variations in endoscopic frequency by drug category indicate a potential impact of bias. (Inflamm Bowel Dis 2014;20:1702–1707) Key Words: diarrhea, drugs, microscopic colitis, collagenous colitis, lymphocytic colitis
M
icroscopic colitis (MC) is an umbrella term for chronic bowel diseases characterized by chronic diarrhea in patients with a normal or near-normal mucosa at colonoscopy. MC comprises 2 main entities, lymphocytic colitis (LC) and collagenous colitis (CC), which are clinically indistinguishable.1 The 2 subtypes share histopathological features characterized by a marked increase in intraepithelial lymphocytes; CC is additionally characterized by a thickened subepithelial collagen band. Therefore, LC and CC are considered to be 2 related but separate entities.2 Epidemiological studies have shown that MC has been more frequently diagnosed in recent decades, with an incidence rate comparable with other inflammatory bowel diseases.3,4 In a U.S. Received for publication May 18, 2014; Accepted June 12, 2014. From the *Diagnostic Centre, Regional Hospital Silkeborg, Silkeborg and University Research Clinic for Innovative Patient Pathways, Aarhus University, Silkeborg, Denmark; †Research Unit for General Practice, Aarhus University, Aarhus, Denmark; ‡ Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark; § Medical Department, Regional Hospital Randers, Randers, Denmark; kMedical Department, Hospital Himmerland, Denmark; and ¶Department of Clinical Medicine, The Faculty of Medicine, Aalborg University, Aalborg, Denmark. The authors have no conflicts of interest to disclose. Reprints: Ole K. Bonderup, MD, PhD, Diagnostic Centre, Regional Hospital Silkeborg, Falkevej 1, DK-8600 Silkeborg, Denmark (e-mail: [email protected]). Copyright © 2014 Crohn’s & Colitis Foundation of America, Inc. DOI 10.1097/MIB.0000000000000143 Published online 22 August 2014.
1702
| www.ibdjournal.org
study from 2007, the overall annual incidence of MC was reported to be 19.6/100,000 persons.4 Recent reports from 2 regions with long standing interest in this disease have found a more stable incidence of MC in the last decennium.5,6 The etiology of the disease is unknown. The diversion of the fecal stream in patients with CC with a temporary ileostomy leads to the resolution of histological changes, implicating luminal factors in the pathogenesis.7 A variety of such factors have been suspected, and drug consumption has been suggested to act as an environmental risk factor implicated as a causative or triggering agent of MC.8 Increased permeability of the colonic mucosa has been observed in patients with MC.9 Drugs may more easily affect the inflammatory response in the colon mucosa of a patient in whom the mucosal defense mechanism is already impaired. In combination with other luminal factors, this could induce the symptoms of MC. Proton pump inhibitors (PPIs), nonsteroidal anti-inflammatory drugs (NSAIDs), statins, and selective serotonin reuptake inhibitors (SSRIs) have all been proposed to be linked to MC.10 All these drugs have been more frequently prescribed in recent years and could therefore be linked to the increase in incidence of MC. However, most of these presumptive associations are based on case reports or small case–control studies, and in a recent study, no association between drug intake and the risk of MC were found.11 We therefore found it important to explore the potential association between exposure to PPIs, NSAIDs, statins, and Inflamm Bowel Dis Volume 20, Number 10, October 2014
Inflamm Bowel Dis Volume 20, Number 10, October 2014
SSRIs and the risk of MC in a nationwide case–control study based on Danish national registries.
Drug Exposure and Risk of MC
exposure and a higher frequency of MC-free colonic biopsies was assessed in a case–control design within the control population.
Patients with MC MATERIALS AND METHODS Data Sources This registry study was based on a analysis of the following datasets covering the entire Danish population of approximately 5.4 million inhabitants. The Danish Civil Registration System was established in 1968, and since then, all Danes have been assigned a 10-digit personal identifier at birth or on immigration.12 This unique civil registry number, which encodes gender and date of birth, is used in virtually all Danish registries and thus permits unambiguous data linkage. The Danish National Patient Register was established in 1977 and includes civil registry number and in-patient discharge diagnoses. Since 1995, it has also included out-patient diagnoses. All of the diagnoses are coded by medical doctors according to the Danish versions of the International Classification of Diseases (ICD-8 from 1977 to 1993 and ICD-10 since 1994).13 The Danish Pathology Register was established in 1990, and in 1997, it became a legal obligation for all departments of pathology in Denmark to report their pathology data statements to this register. The diagnoses are coded after a Danish version of the Systematized Nomenclature of Medicine (SNOMED).14 The data are used by pathologists in daily diagnostic processes and encompass all pathology data in Denmark. The Danish Prescription Register was initiated in 1994 with electronic recording of all of the prescription drugs sold in Denmark.15 This registry retains key information on prescriptions for refundable drugs dispensed from all of the pharmacies in the country, including the civil registration number of the patient, type of prescribed drug according to the anatomical therapeutic chemical classification system16 and date of dispensing the drug. The National Health Service provides tax-supported health care for all inhabitants, guaranteeing free access to family doctors and hospitals, and refunding a variable proportion (usually 50%–75%) of the costs of drugs prescribed by physicians.17 Using this registry, a complete and time-reliable drug prescription history can be established for each individual.
Design The association between drug exposure and the risk of obtaining a diagnosis of CC or LC was studied in a case–control design with population-based controls who were randomly selected using incidence density sampling. All data were collected prospectively in the databases by time of drug exposure and at time of a diagnosis of MC. Our design is thus a historical cohort study taking advantage of the prospectively collected data allowing data analyses to be performed without the potential drawbacks of a retrospective design in terms of recall and information bias. The impact of bias due to a possible association between drug
Our case population consisted of all of the patients with a recorded diagnosis of either CC (SNOMED code: S62536) or LC (SNOMED code: S62533) in the Danish Pathology Register. In the analyses, the cohort was restricted to the period January 2005 to December 2011, as drug exposure data were only available after January 1, 2004. The date of the first recording of one of these diagnoses was defined as the case index date. We excluded patients having ever obtained one of the following diagnoses in the Danish National Patient Register: inflammatory bowel disease (ICD-10: K50 and K510–515), gastrointestinal cancer (ICD-10: C15–26), celiac disease (ICD-10: K900), radiation enteritis (ICD-10: K520 + K627) pancreatitis (ICD-10: K86), bacterial enteritis within the last 12 months before the date of recorded MC diagnosis (ICD-10: A00–A09), and 12 cases with an invalid civil registry number.
Controls One hundred persons per case were randomly selected as controls from the nationwide Danish Civil Registration System and matched by sex and date of birth within 2 weeks. Controls were assigned an index date identical to that of the corresponding cases. None of the controls could have previously diagnosed MC, and we excluded persons who met the abovementioned exclusion criteria. A control patient could, at a later time, be diagnosed with MC. Among the selected controls, we identified a subgroup characterized by MC-free colonic biopsies in the Danish Pathology Register.
Exposure to PPIs, NSAIDs, Statins, and SSRIs For both cases and controls, drug exposure was defined as the redemption of at least 1 prescription for one of the drugs in question recorded in the Danish Prescription Registry in a 1-year period before the index dates. We used the following anatomical therapeutic chemical codes: PPIs, A02BC; NSAIDs, M01A and M01B; statins, C10AA; and SSRIs, N06AB.
Statistical Analyses The case–control study was analyzed using a conditional logistic regression model, which is computationally similar to a stratified Cox regression. The estimated exposure odds ratio (OR) is thus an unbiased estimated of the relative risk.18 Matched case–control sets were used as the units of stratification, as this ensured the finest possible correction for gender, age, and calendar time. Exposure to each of the 4 specified types of drugs was considered both independently (referred to as the crude analyses) and simultaneously to provide estimates corrected for concomitant use of drugs from 1 or more of the other 3 groups (referred to as the adjusted analyses). For the second part of the study, estimates for the association between drug use and colonic biopsies in the control group were assessed using a conditional logistic regression model www.ibdjournal.org |
1703
Inflamm Bowel Dis Volume 20, Number 10, October 2014
Bonderup et al
stratified on matched sets as well. Assuming that the prevalence of undiagnosed MC among controls was relatively low, we interpreted these as estimates of the association between drug use and biopsy frequency among MC-free controls. We finally calculated the ratio between the OR of drug exposure on MC (case–control study) and the corresponding OR on biopsy frequency among MC-free controls (cross-sectional study) to obtain corrected estimates for the drug versus MC association. Calculation of 95% confidence intervals (CIs) was based on a standard normal approximation on the log scale while employing approximate independence between the observed effects of drug exposure on MC and on biopsy frequency among MC-free controls. We performed no formal correction for multiple testing, as the presented results are highly significant for the relevant discussion to focus on their clinical relevance and possible bias rather than their statistical significance. All of the calculations were performed using the statistical software package Stata 11.0.
Ethical Consideration The conduct of this study was approved by the Danish Data Protection Agency (Jr. no. 2010-41-4937). In Denmark, no approval by the Ethics Committee is needed to conduct registrybased studies, such as this study. Informed consent from participants was therefore not required.
RESULTS Case Population During the study period, we identified 6812 patients with the first-time recorded diagnosis of MC in the Danish Pathology Registry. A total of 958 persons were excluded due to 1 or more of the exclusion criteria previously listed (ICD-10: K50 and K510–515), n ¼ 287; gastrointestinal cancer (ICD-10: C15–26), n ¼ 184; celiac disease (ICD-10: K900), n ¼ 102; radiation enteritis (ICD-10: K520 + K627), n ¼ 61; pancreatitis (ICD-10: K86), n ¼ 54; bacterial enteritis within the last 12 months before the date of recorded MC diagnosis (ICD-10: A00–A09), n ¼ 313; and 12 cases with an invalid civil registry number. Furthermore, 101 patients with unclassified MC were excluded. The final case group thus consisted of a total of 5751 patients, including 3474 (60%) with CC and 2277 (40%) with LC. Demographic details, including age and gender, are shown in Table 1.
Drug Exposure and Risk of MC The percentages of cases and controls in each category, which redeemed at least 1 prescription for 1 of the 4 drugs in the year preceding the index date are shown in Table 1. The observed ORs for the associations between drug exposure in cases and controls are shown in Table 2. We found a strong and highly significant association between the redemption of prescriptions for PPIs and both CC (adjusted OR ¼ 7.04; 95% CI, 6.55– 7.56) and LC (adjusted OR ¼ 3.37; 95% CI, 3.08–3.69), whereas the association between SSRIs and LC (adjusted OR ¼ 2.89; 95%
1704
| www.ibdjournal.org
TABLE 1. Drug Exposure in Patients with Microscopic Colitis and Controls CC Cases No. Female gender, % Mean age at diagnose, yr PPI prescription in the preceding year, % NSAID prescription in the preceding year, % Statin prescription in the preceding year, % SSRI prescription in the preceding year, % None of the above, %
LC Controls
Cases
Controls
3474 75.5 67.4 55.7
14.2
2277 63.5 63.8 36.7
12.9
41.5
24.7
35.0
23.6
30.6
21.7
26.8
20.0
18.2
8.8
23.1
7.8
19.5
50.2
27.8
53.3
Descriptive characteristics of cases and controls and the percentage with redemption of a prescription of 1 of the 4 drugs in the year preceding the date of the MC diagnosis.
CI, 2.61–3.20) was slightly weaker, albeit highly significant. All of the other recorded associations were also statistically significant but with ORs well below 2. A subanalysis of drug exposure 90 days before diagnosis did not change the results.
Drug Exposure and MC-free Biopsies Among Controls Among the controls, we identified 2202 persons with MCfree colonic biopsies in the Danish Pathology Register. The
TABLE 2. Odds Ratios of Microscopic Colitis and Drug Exposure
CC No prescription PPI NSAID Statin SSRI LC No prescription PPI NSAID Statin SSRI
OR Crude
95% CI
OR Adjusteda
95% CI
1 (ref) 8.15 2.17 1.66 2.33
— 7.60–8.73 2.03–2.33 1.54–1.79 2.14–2.55
1 (ref) 7.04 1.61 1.29 1.59
— 6.55–7.56 1.50–1.72 1.20–1.39 1.45–1.74
1 (ref) 4.11 1.75 1.53 3.64
— 3.77–4.49 1.60–1.91 1.39–1.69 3.29–4.02
1 (ref) 3.37 1.43 1.25 2.89
— 3.08–3.69 1.31–1.56 1.13–1.38 2.61–3.20
Crude and adjusted ORs of CC and LC by the prescription of 1 of the 4 drugs within 1 year preceding a MC diagnosis. a Analyses adjusted for the prescription of 1 or more of the other 3 drugs.
Inflamm Bowel Dis Volume 20, Number 10, October 2014
Drug Exposure and Risk of MC
results regarding the analyses of this subgroup within the control cohort are shown in Table 3. In these analyses, we compared drug exposure in controls with MC-free colon biopsies to controls without colon biopsies. Redemption of a PPI prescription (adjusted OR ¼ 3.47; 95% CI, 3.08–3.89) in CC controls, and to a slighter degree for SSRIs (adjusted OR ¼ 1.34; 95% CI, 1.14–1.57), seemed to be significantly associated with a normal colonic biopsy in the Pathology Registry. There was no such association for exposure to NSAIDs or statins.
Analyses Adjusted for Increased Frequency of Endoscopy Incorporating these results from our control group in our final model attenuated most of the associations, but they remained significant in cases of PPIs and CC (adjusted OR ¼ 2.03; 95% CI, 1.77–2.33) and SSRIs and LC (adjusted OR ¼ 1.77; 95% CI, 1.42–2.21) (Table 4).
DISCUSSION In this large nationwide case–control study, we observed significant positive associations between exposure to each of the 4 drugs and a recorded diagnosis of MC. However, adjustment for more frequent endoscopic examinations attenuated all of the associations. Even so, significant positive associations remained between PPIs and CC and SSRIs and LC. The associations between MC and exposure to NSAIDs and statins were only weak. Among the controls, our findings of a positive association between drug exposure and having an MC-free biopsy indicate
TABLE 3. Odds Ratios of Having a Microscopic ColitisFree Biopsy and Drug Exposure
Controls for CC (N ¼ 1332) No prescription PPI NSAID Statin SSRI Controls for LC (N ¼ 870) No prescription PPI NSAID Statin SSRI
OR Crude
95% CI
OR Adjusteda
95% CI
1 (ref) 3.64 1.33 1.22 1.67
— 3.25–4.08 1.18–1.50 1.08–1.38 1.43–1.96
1 (ref) 3.47 1.12 1.06 1.34
— 3.08–3.89 0.99–1.26 0.94–1.21 1.14–1.57
1 (ref) 3.77 1.27 1.21 2.03
— 3.27–4.35 1.09–1.47 1.03–1.42 1.67–2.46
1 (ref) 3.57 1.06 1.03 1.63
— 3.08–4.13 0.91–1.23 0.88–1.21 1.34–1.98
Analyses within the control group: crude and adjusted ORs of having a normal colonic biopsy recorded in the Pathology Registry by the redemption of a prescription of 1 of the 4 drugs. a Analyses adjusted for the prescription of 1 or more of the other 3 drugs.
TABLE 4. Adjusted Odds Ratios of Microscopic Colitis and Drug Exposure
CC No prescription PPI NSAID Statin SSRI LC No prescription PPI NSAID Statin SSRI
OR Adjusteda
95% CI
OR Adjustedb
95% CI
1 (ref) 7.04 1.61 1.29 1.59
— 6.55–7.56 1.50–1.72 1.20–1.39 1.45–1.74
1 (ref) 2.03 1.43 1.22 1.19
— 1.77–2.33 1.25–1.65 1.05–1.41 0.99–1.43
1 (ref) 3.37 1.43 1.25 2.89
— 3.08–3.69 1.31–1.56 1.13–1.38 2.61–3.20
1 (ref) 0.95 1.35 1.21 1.77
— 0.80–1.12 1.13–1.61 1.01–1.46 1.42–2.21
ORs of CC and LC adjusted for both the prescription of 1 of the other 3 drugs and surveillance. a Analyses adjusted for the prescription of 1 or more of the other 3 drugs (from Table 2). b Analyses adjusted for the prescription of 1 of the 4 drugs divided by the adjusted ORs for the association between drug use and MC-free biopsy in the control group (from Table 3).
that studies regarding the association between drug use and MC are prone to bias because of different intensities of endoscopic workup in patients exposed to different drugs. We do not have a firm explanation for this increased biopsy frequency in drug users. But a variety of drugs can cause diarrhea leading to subsequent investigation including colonic biopsies. Moreover, PPIs will often constitute the initial empirical treatment in patients with nonspecific abdominal discomfort who may later on end up with a colonoscopy. Previous publications have reported associations between the use of several drugs and an increased risk of MC.8 However, only small numbers of cases have been included, and therefore, controversies still exist. Several studies have found a positive association between MC and PPIs.10 We found a significant association between exposure to PPIs and a diagnosis of CC, and we also observed a significant difference in this association between the 2 MC categories. This might indicate a possible etiological role for PPIs in CC and could also lend support to the notion of 2 separate MC entities. However, inferring causality based on associations found in an observational design must be performed with great caution. NSAID usage has been reported in 30% to 70% of patients with MC. Accordingly, we found NSAID exposure in 42% of patients with CC and 35% with LC in the year preceding the diagnosis. A causal relationship between NSAIDs and CC and LC has not been established, and we only found a weak association between NSAID exposure and MC. Our findings regarding SSRIs are in accordance with the results in a case–control study with 39 cases by Fernandes-Banares et al.10 They found a very strong association between the consumption of www.ibdjournal.org |
1705
Bonderup et al
SSRIs and MC, but the association was of the same magnitude when they looked at patients with chronic diarrhea. In a small study of 26 cases, Pascua et al compared the risk of MC with both randomly selected controls and patients with chronic diarrhea and found no association between PPIs, SSRIs, and statins and MC.11 There could be differences among different drugs in the same class. In this study, we do not have data to elucidate this further. It is, however, important to recognize that a substantial number of patients are diagnosed with MC without any form of drug exposure. In this study, 20% of patients with CC and 28% with LC had no recorded exposure to any of the 4 drugs in question. The main strengths of our population-based study are its large size, the uniformly organized health care system permitting analyses in a nationwide cohort, the use of contemporary matched controls, and the nearly uniform coverage of patients with a recorded diagnosis of MC throughout the country, which reduced the risk of selection bias. During the observation period, the number of patients with analyzed colonic specimens increased. This may reflect a high awareness of MC in Denmark. The reliability of the recorded diagnoses in the Pathology Registry has not been validated. However, the diagnostic criteria for MC were established in 1989, and in a previous study, interobserver and intraobserver agreement was very reliable for the diagnostic categories of MC.19 Our data are based on a pathology register with no available clinical data. Because the diagnosis of MC relies on a combination of symptoms and histological findings, there is a risk of overestimating the number of cases of MC, as the diarrhea may have ceased spontaneously in some patients. In this study, the assessment of drug exposure was based on a comprehensive nationwide prescription database collected independently of our outcome data. We thereby avoided the recall bias that often hampers studies based on interviews, questionnaires, or recordings from primary care.20 We defined drug exposure as having picked up at least 1 prescription for a drug in question. We have no data on the actual rate of drug consumption or the total amount of drug consumed. However, we assume that the likelihood of actual drug consumption is high, as the drugs were collected by the subjects and paid for at the pharmacies. During the study period, low-dose formulations of both PPIs and NSAIDs were available over the counter, but the total amount purchased in these formulations has been reported to be very low compared with prescribed drugs.21 Therefore, the impact of over the counter drugs was most likely minimal. Thus, we consider our estimation of drug exposure to be reliable. The time period between drug exposure and the possible development of MC is unknown. Because there could be a delay between drug exposure and the debut of symptoms, we selected time windows of both 90 and 360 days preceding the diagnosis. Analyses using these different time periods did not change our findings. Medication prescribed more than 12 months before the diagnosis of MC was not analyzed because we found it unlikely to affect our associations. To the best of our knowledge, this is the first nationwide study and by far the largest study in this field allowing for us to
1706
| www.ibdjournal.org
Inflamm Bowel Dis Volume 20, Number 10, October 2014
report risk estimates with very narrow CIs. However, one needs to consider whether an observed statistically significant association between drug therapy and a certain condition represents a clinically relevant effect. Strong statistical associations between drug exposure and MC do not necessarily imply that the drugs can be incriminated as etiologic factors. Our design was observational, and we lacked information on a range of factors that may confound the association between exposure to these 4 drugs and MC (e.g., other drugs, comorbidity not accounted for in the exclusion criteria, tobacco, alcohol, or lifestyle), which either individually or in combination could be unevenly distributed between cases and controls and thus affect our risk estimates. This may lead to a risk of identifying statistically significant associations that may either be clinically irrelevant or ascribed to uncontrolled confounding factors. The reported ORs should therefore be interpreted with caution, and we have arbitrarily restricted our conclusions to associations with an OR above 1.5 in adjusted analyses.22 Our findings of relatively strong associations between exposure to PPIs and SSRIs and MC-free colonic biopsies within the healthy control group indicate a substantial impact of bias due to more frequent endoscopic examinations with biopsies in users of at least these 2 drugs. Surprisingly, no such association was found in patients using NSAIDs, a drug that is generally regarded as a causative agent of abdominal discomfort.23 However, the associations between PPIs and CC and SSRIs and LC remained fairly strong even after adjustment for this bias, whereas the remaining associations were attenuated to a degree at which their clinical relevance is questionable. Our observations confirm previous reports of an association between exposure to PPIs, NSAIDs, statins, and SSRIs and the diagnosis of MC. The potential impact of an increased frequency of endoscopy in drug users should be included in future studies. Our findings of relatively strong associations between drug exposure and normal colon biopsies within the healthy control group indicate a substantial impact of bias because of more intense endoscopic frequency. This potential source of bias should be accounted for in future studies.
REFERENCES 1. Pardi DS, Kelly CP. Microscopic colitis. Gastroenterology. 2011;140: 1155–1165. 2. Jackson BK. Are collagenous colitis and lymphocytic colitis distinct syndromes? Dig Dis. 1995;13:301–311. 3. Olesen M, Eriksson S, Bohr J, et al. Microscopic colitis: a common diarrhoeal disease. An epidemiological study in Orebro, Sweden, 1993–1998. Gut. 2004;53:346–350. 4. Pardi DS, Loftus EV Jr, Smyrk TC, et al. The epidemiology of microscopic colitis: a population based study in Olmsted County, Minnesota. Gut. 2007;56:504–508. 5. Wickbom A, Bohr J, Eriksson S, et al. Stable incidence of collagenous colitis and lymphocytic colitis in Örebro, Sweden, 1999–2008: a continuous epidemiologic study. Inflamm Bowel Dis. 2013;19:2387–2393. 6. Gentile NM, Khanna S, Loftus EV Jr, et al. The epidemiology of microscopic colitis in Olmsted County from 2002 to 2012: a population based study. Clin Gastroenterol Hepatol. 2014;12:838–848. 7. Jarnerot G, Tysk C, Bohr J, et al. Collagenous colitis and fecal stream diversion. Gastroenterology. 1995;109:449–455.
Inflamm Bowel Dis Volume 20, Number 10, October 2014
8. Beaugerie L, Pardi DS. Review article: drug-induced microscopic colitis— proposal for a scoring system and review of the literature. Aliment Pharmacol Ther. 2005;22:277–284. 9. Munch A, Soderholm JD, Wallon C, et al. Dynamics of mucosal permeability and inflammation in collagenous colitis before, during, and after loop ileostomy. Gut. 2005;54:1126–1128. 10. Fernandez-Banares F, Esteve M, Espinos JC, et al. Drug consumption and the risk of microscopic colitis. Am J Gastroenterol. 2007;102:324–330. 11. Pascua MF, Kedia P, Weiner MG, et al. Microscopic colitis and medication use. Clin Med Insights Gastroenterol. 2010;2010:11–19. 12. Pedersen CB. The Danish Civil Registration System. Scand J Public Health. 2011;39:22–25. 13. Lynge E, Sandegaard JL, Rebolj M. The Danish National Patient Register. Scand J Public Health. 2011;39:30–33. 14. Bjerregaard B, Larsen OB. The Danish Pathology Register. Scand J Public Health. 2011;39:72–74. 15. Kildemoes HW, Sorensen HT, Hallas J. The Danish National Prescription Registry. Scand J Public Health. 2011;39:38–41.
Drug Exposure and Risk of MC
16. Capella D. Descriptive tools and analysis. WHO Reg Publ Eur Ser. 1993; 45:55–78. 17. Andersen JS, Olivarius NF, Krasnik A. The Danish National Health Service Register. Scand J Public Health. 2011;39:34–37. 18. Szklo M, Nieto MF. Epidemiology. Beyond the Basics. Gaitersberg, MD: Aspen Publishers Inc., 2000. 19. Limsui D, Pardi DS, Smyrk TC, et al. Observer variability in the histologic diagnosis of microscopic colitis. Inflamm Bowel Dis. 2009; 15:35–38. 20. Shapiro S. Case-control surveillance. In: Strom BL, ed. Pharmacoepidemiology. 2nd ed. Chichester, NH: John Wiley & Sons, 1994:301–322. 21. MEDSTAT.DK [online database]. Available at www.medstat.dk. Last modified June 7, 2014. 22. Grimes DA, Schulz KF. False alarms and pseudo-epidemics: the limitations of observational epidemiology. Obstet Gynecol. 2012;120:920–927. 23. Pirmohamed M, James S, Meakin S, et al. Adverse drug reactions as cause of admission to hospital: prospective analysis of 18 820 patients. BMJ. 2004;329:15–19.
www.ibdjournal.org |
1707
