INTRODUCTION

Microscopic colitis (MC) is characterized by persistent watery diarrhea caused by the inflammation of the colon, resulting in the accumulation of lymphocytes in the colonic epithelium and lamina propria with normal or near-normal colonoscopy findings¹. Two subtypes are known: collagenous colitis (CC) and lymphocytic colitis (LC). For a diagnosis of LC, the histological finding of a diffuse increase of intraepithelial lymphocytes (IELs) (>20 IELs per 100 epithelial cells) in the surface epithelium should be present. The CC variant has features similar to LC, with the additional finding of thickened subepithelial amorphous hyaline eosinophilic bands².

In a Danish study conducted in 2018, the incidence of MC had increased from 4.6% to 24.7 per 100,000 person-years from 2002 to 2011³. Although the pathophysiology is not clearly understood, multiple risk factors have been described, including medications like NSAIDs, proton pump inhibitors (PPIs), specifically lansoprazole, statins, selective serotonin reuptake inhibitors, and other drugs, smoking, celiac disease, inflammatory bowel disease^1,4,5. This can be attributed to a likely increase in the permeability of colonic mucosa in MC⁶.

A study conducted in the Danish population showed a significant positive association between PPIs and CC and SSRIs and LC. They also observed a substantial difference in the associations between the 2 MC categories⁷. According to studies and case reports, there was also a strong association between MC and Lansoprazole use⁴. Our study aimed to evaluate the risk of developing microscopic colitis and its subtypes for patients who are on PPI therapy.

METHODS

DATABASE

Explorys Inc., Cleveland, OH, USA is a validated multicenter and research platform database of more than 360 hospitals from 26 different healthcare systems across the United States consisting of data accumulated from 1999 to September 2022. Explorys was developed and has been prospectively maintained by IBM Corporation, Watson⁸, including electronic health record (EHR) from greater than 60 million unique patients and providing a broad regional distribution of the United States representing approximately 15% of the population. It was utilized to construct a retrospective cohort analysis. A Systematized Nomenclature of Medicine-Clinical Terms (SNOMED-CT) hierarchy⁹, was used to select diagnoses, findings, and procedures. Prescription drug orders are mapped into SNOMED and RxNorm¹⁰. Institutional Review Board (IRB) was not required as source data are de-identified. To protect patient confidentiality, Explorys rounds population counts to the nearest 10 and treats all counts between zero and 10 as equivalent. The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). Access to the database is granted to participating healthcare systems. Use of the Explorys platform has been validated in multiple fields including gastroenterology^11,12.

Patient selection

Adults over 18 years of age were included in the study. Patients with autoimmune diseases were excluded. A subgroup of patients with a diagnosis of “microscopic colitis, lymphocytic colitis and collagenous colitis” was later selected and used in the analysis. The control group was identified as adult patients who did not have a diagnosis of microscopic colitis, lymphocytic colitis and collagenous colitis.

Statistical analysis

Patients who developed microscopic colitis, lymphocytic colitis and collagenous colitis were compared to those who did not. A multivariate regression analysis was performed to account for potential confounders including female gender, smoking history, and the use of proton pump inhibitor, nonsteroidal anti-in­flammatory drugs, and selective serotonin receptor inhibitors. A two-sided P value <0.05 was considered as statistically significant, and all statistical analyses were performed using R version 4.0.2 (R Foundation for Statistical Computing, Vienna, Austria, 2008).

RESULTS

Descriptive epidemiology

78,256,749 individuals aged over 18 years were screened in the database and 69,315,150 were selected after excluding autoimmune diseases. The baseline characteristics of our cohort are displayed in Table 1.

Female gender (55.22%, 83.75%, 81.08%) and Smoking (23.88%, 20%, 21.62%) were respectively more common in patients with microscopic colitis, lymphocytic colitis and collagenous colitis compared to the one without the diseases. PPI (58.2%, 58.75%, 56.75%), SSRI (44.77%, 46.25%, 41.89%) and NSAID (85.07%, 87.5%, 82.43%) were higher in the case group category as well.

Risk and predictors of microscopic colitis, lymphocytic colitis and collagenous colitis in patients using a multivariate regression analysis

In order to adjust for confounding variables, a multivariate regression analysis was performed (Table 2). The risk of microscopic colitis, lymphocytic colitis and collagenous colitis was respectively increased in females (OR: 1.49; CI: 1.29-1.74, OR: 1.56; CI: 1.37-1.78, OR: 2.81; CI: 2.38-3.32), with smoking history (OR: 2.34; CI: 1.98-2.75, OR: 1.64; CI: 1.39-1.91, OR: 2.46; CI: 2.09-2.89), with PPI use (OR: 4.47; CI: 3.78-5.30, OR: 4.03; CI: 3.47-4.67, OR: 3.57; CI: 3.05-4.19), with NSAID use (OR: 2.33; CI: 1.91-2.85, OR: 2.13; CI: 1.80-2.53, OR: 3.79; CI: 3.10-4.65), with SSRI use (OR: 2.84; CI: 2.43-3.32, OR: 2.92; CI: 2.54-3.36, OR: 1.97 ; CI: 1.69-2.29) as seen in Table 2.

DISCUSSION

In this study, we examined the association between various medications, namely Selective serotonin reuptake inhibitors (SSRIs), proton-pump inhibitors (PPIs), and nonsteroidal anti-inflammatory drugs (NSAIDs), and the risk of developing microscopic colitis (MC). The study included different subtypes such as Lymphocytic colitis (LC), collagenous colitis (CC), and MC without differentiation. Our findings revealed that the prevalence of MC and its subtypes was higher among females when compared to the control group. Moreover, our results showed that the use of NSAIDs was associated with a higher proportion of individuals having these conditions compared to the control group. Additionally, PPI use was significantly associated with all three types of MC, with odds ratios (ORs) of 4.47 (95% CI: 3.78-5.30) for MC, 4.03 (95% CI: 3.47-4.67) for LC, and 3.57 (95% CI: 3.05-4.19) for CC. While SSRIs and NSAIDs were also positively correlated with colitis, PPIs had the highest odds of developing this condition. Interestingly, the prevalence of LC and CC was similar in patients treated with PPIs, indicating no specific predisposition to either subtype.

There have been multiple investigations exploring the link between medication use and MC, yet the outcomes of these studies are inconsistent. Researchers have proposed that the conflicting results may be attributed, among other factors, to methodological choices, such as study design, selection of control groups, and consideration of confounding variables in statistical analyses^1,2. Although numerous classes of drugs have been suggested to be associated with MC, PPI, SSRI, NSAIDs, statins, beta-blockers, venotonics, and diuretics, the majority of studies that have confirmed these associations are descriptive and non-controlled^3-5. According to the European Microscopic Colitis Group and United European Gastroenterology, the quality and quantity of evidence supporting the notion that medications are responsible for causing MC is considered to be low and the notion of withdrawing a drug that may be linked to the development of MC is very low⁶. Despite the unknown etiology of MC, researchers have hypothesized that the pathophysiological mechanism involves the induction of abnormal activation of the immune system, which leads to inflammation of the colonic mucosa⁷. However, the fact that multiple drugs with diverse mechanisms, bioavailability, and interactions have been implicated in causing MC raises significant questions for many authors. Some experts propose that the higher incidence of MC diagnoses in patients taking these drugs may be due to drug-induced diarrhea, more frequent colonoscopy tests in these individuals, or the fact that they are elderly¹³.

Several studies have challenged the proposed association between medications and MC. Some authors argued that a positive association is usually seen when controls are healthy and when controls have diarrhea studies are failing to show such an association¹⁴. In a recent multicenter retrospective cohort study, Zylberberg et al. (2021) investigated the relationship between medications and MC by examining a pool of 334 patients with MC who were matched to 668 community controls from Columbia University and Mayo Clinic¹⁵. The authors observed an inverse association between MC and diabetes medications (OR: 0.47; 95%CI: 0.27-0.81; P=0.007), PPIs (OR: 0.64; 95%CI: 0.47-0.87; P=0.005), and SSRIs (OR: 1.06; 95%CI: 0.77-1.47; P=0.73). Conversely, NSAIDs were positively associated with an OR of 1.63 (95%CI: 1.12-2.38; P=0.01). The study was robust in its selection of cases, confirmed by positive biopsy diagnosis for MC, and controls, confirmed by negative biopsy diagnosis for MC in patients undergoing colonoscopy for diarrhea. However, possible potential bias resulting from the inclusion of celiac disease patients in the analysis were possible as well as the lack of information regarding the indication for drug use and causes of diarrhea among their overmatched controls. Sandler et al. (2021) investigated, in a case-control study, the association between PPIs, NSAIDs, SSRIs, and MC in a cohort of 110 MC cases and 252 non-MC colonoscopy controls¹. The controls were selected from patients referred for elective colonoscopies due to possible MC or diarrhea as an indication. The results showed that the use of PPIs, NSAIDs, and SSRIs was less frequent in the MC cases compared to controls, with percentages of 31.1%, 39.6%, and 41.5%, respectively, versus 45.6%, 51.6%, and 43.0%, respectively. In the multivariate analysis, all three drugs were inversely associated with MC, with ORs of 0.66, 0.68, and 0.92, respectively. However, the eligibility criteria in the study relied on patient self-reporting, which may affect the reliability of the data collected using the Bristol Stool Form Scale and the medication use was prevalent in the control group, with over 45% of patients being on medication, raising concerns on the selection process for patients undergoing elective colonoscopies. A notable investigation conducted by Pascua et al. (2011) aimed to compare a pool of 26 cases with MC to 259 random controls and 259 diarrhea controls². The study findings demonstrated an inverse association between the use of PPIs and SSRIs in all results. It is worth mentioning that the OR increased from 0.06 and 0.55, respectively, when compared to random controls to 0.28 to 0.87, respectively, when compared to diarrhea controls. However, the statistical power of the study is limited due to the small sample size and non-random selection of control patients. Moreover, the selection of cases was not sufficiently rigorous, given the ambiguity surrounding the definition and duration of diarrhea before colonoscopy. Our study yielded contradictory results compared to the aforementioned studies, but provides a more comprehensive methodology with a large sample size, strengthening its statistical power.

Our findings corroborate with several previous studies that have also investigated the association between medications and MC. The earliest case reports linked the use of NSAIDs with colonic inflammation, specifically CC^16,17. Subsequently, a recent study by Yen et al. (2022) established a statistically significant association between NSAIDs and MC with an OR of 3.04 (95%CI: 1.65-5.69; P<0.001)¹⁸. Remarkably, this study employed a control group with diarrhea, contradicting the belief that only studies with healthy controls have presented a favorable relationship between medications and MC Bonderup et al. (2018) Danish population-based case-control study represents the largest investigation to date exploring the relationship between PPIs, NSAIDs, and CC/LC¹⁹. The study yielded promising findings, demonstrating a strong association of PPIs with both CC (OR: 6.98; 95%CI: 6.45-7.55) and LC (OR: 3.95; 95%CI: 3.60-4.33) across all classes of PPIs, with the strongest association observed for lansoprazole. Additionally, NSAIDs were positively associated with both CC and LC. The timing of medication exposure was also taken into account in the analysis, with current PPI users defined as those with a prescription within 90 days prior to diagnosis exhibiting the highest association (OR: 6.98; 95%CI: 6.45-7.55 for CC and OR: 3.95; 95%CI: 3.60-4.33 for LC), followed by recent users with a prescription between 91 and 365 days prior to diagnosis (OR: 5.16; 95%CI: 4.66-5.72 for CC and OR: 2.31; 95%CI: 2.03-2.64 for LC) and past users with a prescription more than 365 days prior to diagnosis (OR: 1.52; 95%CI: 1.39-1.67 for CC and OR: 1.35; 95%CI: 1.23-1.49 for LC). Furthermore, a dose-response analysis was conducted, which did not reveal any increased risk for CC or LC with increasing PPI dosage. This paper’s noteworthy contributions to the literature include its large sample size, robust allocation for cases and community controls, and analyses that took into account medication duration, dosage, and class. Many other studies have provided evidence of an association between medication use and MC^4,20,21. Of note, a highly cited study reported that patients with MC have a higher medication consumption than the control population²². These findings were further supported by a second comparison control group with watery diarrhea. These findings suggest that increased colonoscopy frequency and drug-induced diarrhea may be responsible for the diagnosis of MC rather than the medications themselves.

Two meta-analyses have investigated the association between medication use and MC^23,24. The more recent meta-analysis by Jinlu et. Al (2015), which included six case-control studies, revealed a significant risk for MC associated with the use of PPIs (OR: 2.68; 95%CI: 1.73-4.17) and SSRIs (OR: 2.41; 95%CI: 1.64-3.53), as well as a slightly elevated risk with the use of NSAIDs (OR: 1.50; 95%CI: 1.14-1.96). Our study’s findings align with those of the aforementioned meta-analysis and the European guidelines on MC, which identified PPI use as the most strongly associated with MC (OR: 2.95; 95%CI: 1.82-4.80; I2=99%)⁶.

The present study also serves as a critical review of recent findings and limitations of previous investigations. Wherein, we provide guidance for future studies. To achieve this, robust study designs of large scale are needed, incorporating precise pathological diagnostic tools for MC and accounting for potential confounding variables such as inflammatory bowel diseases, infectious and toxin-induced diarrhea, smoking, age, sex, autoimmune disease, frequency and indication of colonoscopies, among others. Prospective data collection studies are imperative to minimize recall and protopathic bias, with careful assessment of drug exposure that accounts for all relevant characteristics including dosage, duration of exposure, and indications, thereby eliminating idiosyncratic reactions and enabling identification of dose-dependent associations. Overmatching should be minimized by careful selection of control groups, whether community-based, colonoscopy-based, or those with diarrhea. When selecting diarrhea controls, close attention should be paid to the causes of diarrhea and patient characteristics. Additionally, future studies should investigate the possibility of resolution of diarrhea and histological normalization following withdrawal of drugs, thus addressing the question of what type of controls are needed. These recommendations will facilitate the progression of our comprehension of the cause-and-effect relationship between medication usage and MC.

There are several limitations to our study that should be taken into consideration. First, the absence of information regarding the dose, duration of exposure, and indications for the various medications used, as well as the lack of the precise diagnostic tools used to diagnose MC, limit the conclusions that can be drawn from our results. Second, our study failed to control for other important confounding variables such as race and alcohol use disorder. Finally, the retrospective, population-based nature of our study hinders our ability to establish a causal relationship between MC and medication use, and increases the potential for recall and protopathic bias, thus necessitating further research with more detailed patient characteristics.

CONCLUSION

Our work represents a significant contribution to the current debate surrounding the relationship between medication use and MC. Specifically, we conducted the largest multicenter study, utilizing a population-based approach to implement a multivariate regression analysis that accounted for potential confounding factors. Medications such as SSRI and NSAIDs had a positive correlation with MC and PPIs had the highest odds of developing MC. Our findings provide valuable insights into the MC and medications relationship that will guide physicians to recognize and eliminate risk factors prior to initiating treatment for this disease.

REFERENCES

Edited by

Data availability

Publication Dates

History