Discussion
In this population-based, nested case–control study use of NSAIDs, PPIs, selective SSRIs, low-dose aspirin, ACE inhibitors, or beta-blockers were associated with an increased risk of MC. However, when taking confounding-by-indication, diagnostic delay and diagnostic bias into account (by comparing with subjects who have had a colonoscopy negative for colorectal cancer and MC), only PPIs and NSAIDs significantly increased the risk of MC. This finding was supported by dose analyses showing increasing estimates with higher cumulative doses used.
Different studies were conducted before this study with contradicting results. Regarding NSAIDs, an unmatched case–control study compared the frequency of use of several drugs in patients with chronic watery diarrhea, but not with MC, to that in patients attending an outpatient surgery unit. This study suggested that NSAIDs, SSRIs, and statins were significantly associated with chronic watery diarrhea. Our results are well in line with this, as we showed that only PPIs, NSAIDs, and low-dose aspirin increased the risk of MC when compared with colonoscopy controls. The apparent increased MC risk with use of ACE inhibitors, beta-blockers, SSRIs, and statins in that study is probably an artefact. In the current study, these drugs did not increase the risk of MC when compared with community controls. They increase the likelihood of undergoing a colonoscopy by, for instance, inducing abdominal symptoms, increasing stool frequency or worsening diarrhea. In addition, NSAIDs are indicated for treatment of arthralgias, which often accompany MC. Thus, subjects with arthralgia may be more likely to undergo colonoscopy and be diagnosed with MC. Our results are in line with a recent Danish case–control study. This study is, however, limited by the fact they classified having only one filled prescription in the year before MC diagnosis as being exposed to the drug, which may predispose to misclassification of exposure. No distinction was made between current users and past users, whereas filling more than only one prescription would provide a more robust investigation of the exposure. Although the association between NSAIDs and MC may be explained by the underlying comorbid disease, current use of NSAIDs in our population remained associated with an increased risk of MC even when we adjusted for concomitant exposure, polyarthritis, and rheumatoid arthritis. Thus, it is unlikely that the indication of NSAID could explain the association between NSAIDs and MC as observed in the current study. This observation is supported by previous reports. In addition, in our study NSAIDs still increased the risk of MC up to sixfold when taking diagnostic delay and diagnostic bias into consideration. Furthermore, the association was supported by the dose analysis providing increasing risk estimates with higher cumulative doses of NSAIDs used. To examine whether the risk of MC was different for individual types of NSAIDs, we conducted analyses for individual NSAIDs. Diclofenac was the most commonly used NSAID and accounted for 52–67% of NSAIDs used by cases. As fewer exposed cases could be matched to colonoscopy-negative controls, the study had insufficient power to estimate a risk estimate for each individual NSAID. The results on individual NSAIDs should, thus, be interpreted with caution. Physiologically, the association can be explained by loosening the colonic paracellular permeability, inducing lower gastrointestinal bleeding and affecting the bowel integrity. Subsequently, luminal antigens can more easily enter the lamina propria and elicit an immune and inflammatory reaction. Therefore, we suggest that NSAIDs should be avoided in patients at risk of developing MC. To the best of our knowledge, no clinical decision models are available to weigh the benefits and risks of NSAIDs including MC as an adverse event. A balanced decision should be based on clinical knowledge and the preference of the patient.
PPIs
The recent Danish study also noted the association between PPIs and MC, regardless of having had colonic biopsies. Again, although they applied a second control group, the exposure definition was rather broad with only one filled prescription required to being classified as exposed to the drug. In addition, they relied on the MC diagnosis date rather than the start of symptoms of MC as we could verify in the medical records. A previous case–control study from the Netherlands showed that use of PPIs was associated with an increased risk of MC. However, this study may have suffered from selection bias, as cases were retrieved from a secondary and tertiary hospital, whereas controls were selected from the general population. Another case–control study on 26 cases found contrasting results with no increased risk of MC with PPI use. This study may be limited by its sample size and by lack of adjustment for NSAIDs, which are often concomitantly used with PPIs. As is shown in our study, substantial confounding-by-indication is present as the risk of MC for current use of PPIs in the 1-year risk period decreased from 7.3- to 4.4-fold when using colonoscopy controls instead of community controls. Similar to NSAIDs, we investigated the risk of MC for different individual PPIs. Omeprazole was the predominant PPI used and accounted for 59–67% of PPIs used by cases. There have been several case series published on PPI-induced MC; however, the underlying pathophysiological mechanism remains poorly understood. Experimental studies have shown the influence of PPIs on intestinal permeability, inducing smooth muscle relaxation and inhibiting contractile activity. The effect on the contractile activity system may also affect the actinomyosin cytoskeleton, resulting in conformational changes in the cytoskeleton of epithelial cells and subsequent alterations in the function of tight junctions. PPIs also affect the colonic intestinal flora, thereby increasing the risk of bacterial intestinal infections. Evidence on this topic until now, however, is scarce.
Beta-blockers and ACE Inhibitors
Our observation that beta-blockers and ACE inhibitors were associated with MC compared with community controls, but not compared with colonoscopy controls, could be explained by two reasons. First, the association between beta-blockers and ACE inhibitors is true, but could not be confirmed in comparison with the colonoscopy controls because of smaller sample size. Second, the association is false and likely due to confounding-by-indication; for instance, subjects on drugs may be more closely monitored. There is some evidence on a pathophysiological mechanism by which beta-blockers augment small intestinal transit by increasing the propulsive force associated with small intestinal contractions. To which extent these effects result in increased bowel frequency and stool consistency has not been studied.
Selective SSRIs
We demonstrated that SSRIs increased the risk of MC when compared with community controls, but not when compared with colonoscopy controls. This could be explained by confounding-by-indication, in particular as diarrhea is a known side effect of SSRI use.
Strengths and Limitations
Strengths of the current study include the population-based setting on the basis of the electronic medical records, which allowed to identify all potential MC cases while matching to community controls and colonoscopy-negative controls. By doing so, we mitigated against selection bias as the cases and controls were derived from the same source population. In this context, the OR may be interpreted directly as the relative risk. Second, by applying different risk periods, from 1 year up to 2 years before the index date, we corrected for any diagnostic delay of MC, which may take up to several months. This is demonstrated by the fact that exposure closer to the index date (<3 months) yielded higher risks of MC than longer exposure before (>12 months). This could be due to an actual risk increase, but is more likely affected by lag-time bias. We also chose the date of first symptoms leading to the MC diagnosis as the index date in order to avoid any misclassification of the exposure in the risk periods.
We acknowledge the following limitations, however. First, MC is a specialist-confirmed diagnosis, meaning that in a GP database under-reporting of MC could have occurred. We assumed that all relevant medical information on a microscopic diagnosis is recorded by the GP. However, this may not hold true entirely and we may have missed some MC cases. Nevertheless, we extensively reviewed the medical records of all MC cases included in the current study to ensure that the MC cases included in the study had histological confirmation of the diagnosis. Misclassification of MC is therefore unlikely, but if present it will be nondifferential and will have resulted in more conservative, but unbiased estimates. In addition, by excluding the unspecified MC cases we performed sensitivity analyses that yielded similar results as the initial analyses. When stratifying by type of MC, we found similar results as for the main analysis. However, caution should be taken when interpreting the findings on the risk of collagenous and lymphocytic colitis separately by use of drugs, as the subgroup analyses were based on small numbers. We included two control groups, which both have their limitations. The colonoscopy group had a clinical indication to undergo this procedure, which you can see as a test-negative control group; however, this is similarly so for the cases. Confounding for 'indication' is actually reduced in such control group. Colonoscopies were performed for the complete range of clinical indications, including colorectal cancer screening. However, results from analyses are consistent across the control groups, which strengthen the results and provide consistent evidence that NSAIDs and PPIs are associated with MC. Moreover, in the more recent years clinicians became more aware of MC. The majority of cases was diagnosed after 2006. This has an effect on the sample size, but was unlikely to provide spurious or biased associations as cases and controls were matched on calendar time. Second, information on other potential confounders such as smoking status and alcohol use is likely to be under-reported in the database and was therefore not considered in the analysis. This may have resulted in residual confounding. Third, the date of onset of disease could have been misclassified. Although we tried to mitigate against misspecification of the risk window by using the date of onset of symptoms as the index date, we cannot account for patients' delay between actual start of symptoms and recording of symptoms by the GP. Yet, in order to account for diagnostic delay, we applied two different risk periods in the current study. Fourth, we only observed that NSAIDs and PPIs were associated with MC. However, because of the smaller sample size, particularly in the risk period of 2 years while excluding the last year before the index date, we cannot rule that out because of power issues we did not observe an association for the other drugs.
In conclusion, NSAIDs and PPIs significantly increased the risk of MC, even after taking diagnostic bias and diagnostic delay into account. Use of selective SSRIs, low-dose aspirin, beta-blockers, and ACE inhibitors were associated with increased risk of MC when compared with community controls, but not compared with colonoscopy-negative controls. We suggest that the association between SSRIs, low-dose aspirin, beta-blockers, ACE inhibitors, and MC in community controls could be due to worsening of diarrhea or symptoms in patients with underlying colonic disease requiring colonoscopy rather than increasing the risk of MC itself.