Metformin Use Is Associated with a Lower Risk of Inflammatory Bowel Disease in Patients with Type 2 Diabetes Mellitus

Abstract

Aim

Our aim was to compare the risk of developing inflammatory bowel disease [IBD] between ever users and never users of metformin.

Methods

Patients with newly diagnosed type 2 diabetes mellitus from 1999 to 2005 were enrolled from Taiwan’s National Health Insurance. A total of 340 211 ever users and 24 478 never users who were free from IBD on January 1, 2006 were followed up until December 31, 2011. Hazard ratios were estimated by Cox regression incorporating the inverse probability of treatment weighting using a propensity score.

Results

New-onset IBD was diagnosed in 6466 ever users and 750 never users. The respective incidence rates were 412.0 and 741.3 per 100 000 person-years and the hazard ratio for ever vs never users was 0.55 [95% confidence interval: 0.51–0.60]. A dose–response pattern was observed while comparing the tertiles of cumulative duration of metformin therapy to never users. The respective hazard ratios for the first [<26.0 months], second [26.0–58.3 months] and third [>58.3 months] tertiles were 1.00 [0.93–1.09], 0.57 [0.52–0.62] and 0.24 [0.22–0.26]. While patients treated with oral antidiabetic drugs [OADs] without metformin were treated as a reference group, the hazard ratios for patients treated with OADs with metformin, with insulin without metformin [with/without other OADs] and with insulin and metformin [with/without other OADs] were 0.52 [0.42–0.66], 0.95 [0.76–1.20] and 0.50 [0.40–0.62], respectively.

Conclusion

A reduced risk of IBD is consistently observed in patients with type 2 diabetes mellitus who have been treated with metformin.

Inflammatory bowel disease, Crohn’s disease, ulcerative colitis, metformin, Taiwan

1. Introduction

Inflammatory bowel disease [IBD] is characterized by chronic relapsing inflammation of the gut because of excessive expression of inflammatory mediators. IBD can be classified into two major clinical diagnoses, i.e. Crohn’s disease and ulcerative colitis, but clinical presentations of the two are sometimes difficult to differentiate and examination by colofibroscopy is necessary. The incidence of IBD has been increasing since 1990s in newly industrialized countries in Asia, South America and Africa.^1,2

The pathophysiology of IBD involves interactions between genes and environment, with a false or aberrant recognition of commensals, nutrients and pathogens, leading to activation of the immune system and an inflammatory response that breaks the epithelial barrier in the gut.^2,3 Recent studies in cell cultures and animals have suggested that metformin may reduce proinflammatory cytokines and chemokines in the intestine, resulting in protection against intestinal barrier dysfunction.^4,5 A mouse study conducted in Korea showed that metformin administration for 16 days attenuated IBD severity and reduced inflammation through inhibiting the expression of phosphorylated signal transducer and activator of transcription 3 and interleukin-17.⁶ Therefore, it is reasonable to hypothesize a beneficial effect of metformin on IBD in humans.

To our knowledge, no previous study has investigated whether metformin use might reduce the risk of IBD in humans. Therefore, the purpose of this study was to evaluate such a potential benefit of metformin in patients with type 2 diabetes mellitus by using a nationwide reimbursement database of Taiwan’s National Health Insurance [NHI].

2. Materials and Methods

The NHI is a compulsory healthcare system that was implemented in Taiwan in March 1995. It covers >99% of Taiwan’s population and the Bureau of the NHI has contracts with all hospitals and 93% of all medical settings. The reimbursement database of the NHI keeps computer records of all disease diagnoses, medication prescriptions and clinical procedures. The database can be used for academic research after ethics review and approval by the Research Ethics Committee of the National Health Research Institutes. Because all personal information has been de-identified for the protection of privacy, informed consent was not required according to local regulations. The present study was granted approval number 99274.

Disease diagnoses were coded by the International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM]. Diabetes was coded 250.XX and IBD was coded 555 [regional enteritis] or 556 [ulcerative enterocolitis].

The procedures in enrolling a cohort of ever users and never users of metformin are shown in Figure 1. In total, 423,949 patients newly diagnosed with diabetes mellitus during 1999–2005 in the outpatient clinics and having been prescribed antidiabetic drugs two or more times were identified. Ineligible patients were then excluded: [1] type 1 diabetes mellitus [n = 2400], [2] missing data [n = 746], [3] patients with a diagnosis of IBD before the start of follow-up or within 6 months of diabetes diagnosis [n = 22 577], and [4] patients followed up for <6 months [n = 33 537]. As a result, 340 211 ever users and 24 478 never users of metformin were identified.

Figure 1.

Flowchart showing the procedures in creating a cohort of metformin ever and never users from the reimbursement database of Taiwan’s National Health Insurance. IBD: inflammatory bowel disease.

Potential confounders were categorized into subgroups of demographic data, major comorbidities, diabetes-related complications, antidiabetic drugs, commonly used medications in diabetes patients, and commonly encountered comorbidities and potential risk factors. These potential confounders were selected based on clinical judgement of a potential relationship with exposure and/or outcome or because of a potential effect on the life expectancy of the patients that might lead to a biased estimate of person-years resulting from short follow-up duration. Because antibiotics, steroid and anti-inflammatory drugs may affect IBD risk, disease diagnoses that might require the long-term use of these drugs were considered in particular. Demographic data included age, sex, occupation and living region. Major comorbidities included hypertension [401–405], dyslipidaemia [272.0–272.4] and obesity [278]. Diabetes-related complications included nephropathy [580–589], eye diseases [250.5: diabetes with ophthalmic manifestations, 362.0: diabetic retinopathy, 369: blindness and low vision, 366.41: diabetic cataract, and 365.44: glaucoma associated with systemic syndromes], stroke [430–438], ischaemic heart disease [410–414] and peripheral arterial disease [250.7, 785.4, 443.81 and 440–448]. Anti-diabetic drugs included insulin, sulfonylurea, meglitinide, acarbose, rosiglitazone and pioglitazone. Medications commonly used by diabetes patients included angiotensin-converting enzyme inhibitors/angiotensin receptor blockers, calcium channel blockers, statins, fibrates and aspirin. Commonly encountered comorbidities and potential risk factors included chronic obstructive pulmonary disease [a surrogate for smoking, 490–496], tobacco misuse [305.1, 649.0 and 989.84], alcohol-related diagnoses [291, 303, 535.3, 571.0–571.3 and 980.0], cancer [140–208], heart failure [398.91, 402.11, 402.91, 404.11, 404.13, 404.91, 404.93 and 428], Parkinson’s disease [332], dementia [abridged codes of A210 or A222, or ICD-9-CM codes of 290.0, 290.1, 290.2, 290.4, 294.1, 331.0–331.2 and 331.7–331.9], head injury [959.01], valvular heart disease [394–396, 424 and 746], gingival and periodontal diseases [523], pneumonia [486], osteoporosis [733.00], arthropathies and related disorders [710–719], psoriasis and similar disorders [696], dorsopathies [720–724], liver cirrhosis [571.5], other chronic non-alcoholic liver diseases [571.8], hepatitis B virus infection [070.22, 070.23, 070.32, 070.33 and V02.61], hepatitis C virus infection [070.41, 070.44, 070.51, 070.54 and V02.62], human immunodeficiency virus infection [042, 079.53, V08, V01.79 and 795.71] and organ transplantation [V42].

The accuracy of disease diagnoses in the NHI database has been studied previously.^7,8 For diabetes mellitus using the codes of 250.XX, the sensitivity was 90.9% and the positive predictive value was 90.2%.⁷ Agreements between claim data and medical records are moderate to substantial, with Kappa values ranging from 0.55 to 0.86.⁸ However, the validity of IBD has not been previously evaluated.

The living region was classified as Taipei, Northern, Central, Southern and Kao-Ping/Eastern. Occupation was classified as class I [civil servants, teachers, employees of governmental or private businesses, professionals and technicians], class II [people without a specific employer, self-employed people or seamen], class III [farmers or fishermen] and class IV [low-income families supported by social welfare, or veterans].

Student’s t tests for age and chi square tests for other variables were used to examine the differences in the distribution of these variables between ever and never users of metformin. Additionally, a standardized difference for each variable was calculated according to Austin and Stuart.⁹ A standardized difference value of >10% was used to indicate potential confounding from the variable.

The cumulative duration of metformin therapy in months was calculated and its tertiles were used for dose–response analyses. Incidence density was calculated for never users, ever users, tertiles of cumulative duration of metformin therapy and subgroups of antidiabetic drugs with regard to metformin exposure. Start of follow-up was set to January 1, 2006. The numerator of the incidence was the case number of newly identified IBD during follow-up. The denominator [expressed in person-years] was the follow-up duration between the start of follow-up and the time of a new diagnosis of IBD, or the date of death or the date of the last reimbursement record, whichever occurred first up to December 31, 2011.

Cumulative incidence functions for IBD were plotted for never users and ever users and for never users and tertiles of cumulative duration. Gray’s test was used to test for differences in the cumulative incidence functions.

All characteristics in Table 1 plus the date of entry were used to create propensity scores by logistic regression. Hazard ratios and their 95% confidence intervals were estimated by Cox regression incorporated with the inverse probability of treatment weighting [IPTW] using the propensity scores. This method was proposed by Austin to reduce potential confounding from the differences in characteristics.¹⁰ In the main analyses, hazard ratios were estimated for ever vs never users, for ever users divided into tertiles of cumulative duration of metformin therapy vs never users, and for antidiabetic subgroups divided into oral antidiabetic drugs without metformin [treated as the reference group], oral antidiabetic drugs with metformin, insulin without metformin [with or without other oral antidiabetic drugs] and insulin with metformin [with or without other oral antidiabetic drugs].

Table 1.

Characteristics in never and ever users of metformin

Variable	Never users		Ever users
	n = 24 478		n = 340 211		p-value	Standardized difference
	n	%	n	%
Demographic data
Agea [years]	65.1	13.4	59.5	12.4	<0.01	−48.7
Sex [men]	13 359	54.6	180 616	53.1	<0.01	−3.0
Occupation
I	8740	35.7	131 019	38.5	<0.01
II	4057	16.6	70 927	20.9		11.9
III	5915	24.2	72 842	21.4		−6.6
IV	5766	23.6	65 423	19.2		−11.7
Living region
Taipei	8456	34.6	115 713	34.0	<0.01
Northern	2636	10.8	42 175	12.4		5.5
Central	4193	17.1	59 288	17.4		0.6
Southern	4206	17.2	55 660	16.4		−1.9
Kao-Ping and Eastern	4987	20.4	67 375	19.8		−0.8
Major comorbidities
Hypertension	19 219	78.5	243 703	71.6	<0.01	−17.8
Dyslipidaemia	14 398	58.8	230 444	67.7	<0.01	20.6
Obesity	480	2.0	14 391	4.2	<0.01	13.5
Diabetes-related complications
Nephropathy	7208	29.5	59 530	17.5	<0.01	−33.7
Eye diseases	2182	8.9	49 800	14.6	<0.01	18.9
Stroke	7912	32.3	77 571	22.8	<0.01	−24.7
Ischaemic heart disease	11 191	45.7	128 055	37.6	<0.01	−19.1
Peripheral arterial disease	4462	18.2	58 775	17.3	<0.01	−3.6
Antidiabetic drugs
Insulin	2259	9.2	7121	2.1	<0.01	−36.2
Sulfonylurea	17 194	70.2	217 182	63.8	<0.01	−6.3
Meglitinide	2304	9.4	12 457	3.7	<0.01	−26.4
Acarbose	2811	11.5	16 678	4.9	<0.01	−23.6
Rosiglitazone	691	2.8	14 316	4.2	<0.01	8.6
Pioglitazone	567	2.3	7871	2.3	0.98	1.3
Commonly used medications in diabetes patients
Angiotensin-converting enzyme inhibitors/angiotensin receptor blockers	15 711	64.2	202 578	59.5	<0.01	−11.1
Calcium channel blockers	15 175	62.0	178 505	52.5	<0.01	−21.4
Statins	9461	38.7	149 173	43.9	<0.01	11.9
Fibrates	6502	26.6	107 414	31.6	<0.01	12.0
Aspirin	13 049	53.3	168 037	49.4	<0.01	−9.6
Commonly encountered comorbidities and potential risk factors
Chronic obstructive pulmonary disease	11 617	47.5	142 241	41.8	<0.01	−14.0
Tobacco misuse	316	1.3	6490	1.9	<0.01	5.3
Alcohol-related diagnoses	1338	5.5	17 222	5.1	<0.01	−2.1
Cancer	3689	15.1	34 622	10.2	<0.01	−17.5
Heart failure	5287	21.6	44 587	13.1	<0.01	−26.8
Parkinson’s disease	1029	4.2	6947	2.0	<0.01	−14.9
Dementia	2221	9.1	16 965	5.0	<0.01	−20.0
Head injury	351	1.4	4093	1.2	<0.01	−2.7
Valvular heart disease	2876	11.8	25 658	7.5	<0.01	−17.3
Gingival and periodontal diseases	18 287	74.7	269 264	79.2	<0.01	11.4
Pneumonia	3588	14.7	31 821	9.4	<0.01	−20.8
Osteoporosis	5697	23.3	62 443	18.4	<0.01	−14.3
Arthropathies and related disorders	17 542	71.7	234 047	68.8	<0.01	−7.4
Psoriasis and similar disorders	524	2.1	7738	2.3	0.17	0.9
Dorsopathies	16 990	69.4	237 489	69.8	0.19	0.5
Liver cirrhosis	1604	6.6	12 756	3.8	<0.01	−14.6
Other chronic non-alcoholic liver diseases	1930	7.9	30 408	8.9	<0.01	3.9
Hepatitis B virus infection	517	2.1	5767	1.7	<0.01	−3.9
Hepatitis C virus infection	1203	4.9	12 082	3.6	<0.01	−7.7
Human immunodeficiency virus infection	16	0.1	181	0.1	0.43	−0.8
Organ transplantation	161	0.7	580	0.2	<0.01	−9.9

Refer to ‘Materials and Methods’ for the classification of occupation.

^aAge is expressed as mean and SD.

Sensitivity analyses were conducted by estimating hazard ratios for ever users vs never users in more restricted subgroups in the following models, I, censoring patients from the time 4 months have elapsed since the last prescription. II, excluding patients who received other antidiabetic drugs before the first prescription of metformin [this excluded the potential carry-over effect of other antidiabetic drugs]. III, excluding patients followed up for <12 months. IV, excluding patients who had been treated with metformin for <12 months. V, patients enrolled during 1999–2002. VI, patients enrolled during 2003–2005. VII, excluding two consecutive prescriptions of metformin spanning more than 4 months [because the NHI allows drug prescription for a maximum duration of 3 months at each time, these patients represented those with irregular follow-up and having delayed refill of metformin for at least 1 month after a previous prescription of 3 months]. VIII, excluding patients treated with incretin-based therapies during follow-up. [a recent study suggested that dipeptidyl peptidase 4 inhibitors may affect the composition of gut microbiota¹¹ and in Taiwan, the first incretin-based therapy was not reimbursed by the NHI until after 2009. The exclusion of these patients avoided the potential impact of incretin-based therapies during follow-up]. IX, IBD defined by diagnostic codes plus operation codes [the ICD-9-CM procedure codes for IBD-related bowel surgery were adapted from table S8 in the paper by Di Domenicantonio et al.¹² This is to reduce the potential risk of misdiagnosis and misclassification of IBD at the outpatient clinics]. X, age <50 years. XI, age 50–64 years. XII, age ≥65 years.

To further examine whether the results were consistent when cross-sectional analyses were conducted in the database, we performed logistic regression to estimate the multivariate-adjusted odds ratios for prevalent users vs non-users of metformin for three different periods of time, i.e. 1999–2005, 2006–2011 and 1999–2011.

It is recognized that the competing risk of death is a potential factor that might have biased the results of the study. If ever users of metformin had a higher risk of death, this would preclude the occurrence of IBD among ever users and bias the estimate favouring metformin. In Taiwan very complete information for death is given in the NHI healthcare system and patients who died are marked in the database. We therefore additionally estimated the multivariate-adjusted hazard ratio for death for ever vs never users.

Analyses were conducted using SAS statistical software, version 9.4 [SAS Institute]. A value of p < 0.05 was considered statistically significant.

3. Results

The characteristics of never users and ever users are shown in Table 1. The p values compared by Student’s t test [for age] and chi-square test [for other variables] were <0.01 for all variables except pioglitazone, psoriasis and similar disorders, dorsopathies and human immunodeficiency virus infection, which showed non-significant p values (> 0.1). For standardized difference, the following variables showed values of >10%: age, occupation, hypertension, dyslipidaemia, obesity, nephropathy, eye diseases, stroke, ischaemic heart disease, insulin, meglitinide, acarbose, angiotensin-converting enzyme inhibitors/angiotensin receptor blockers, calcium channel blockers, statins, fibrates, chronic obstructive pulmonary disease, cancer, heart failure, Parkinson’s disease, dementia, valvular heart disease, gingival and periodontal diseases, pneumonia, osteoporosis and liver cirrhosis. Therefore, ever and never users of metformin showed significant differences in the distribution of potential confounders. This justified the use of the IPTW method for the estimation of hazard ratios.

The cumulative incidence functions for IBD with regard to metformin exposure are shown in Figure 2. Figure 2A shows the curves for never users and ever users, indicating a significantly lower cumulative incidence in ever users [Gray’s test p < 0.01]. Figure 2B shows the respective curves for never users and users divided according to the tertiles of cumulative duration of metformin therapy. The cumulative incidence of the first tertile of users was very similar to that of never users, but patients in the second and the third tertiles showed lower cumulative incidences [Gray’s test p < 0.01].

Figure 2.

Cumulative incidence functions for inflammatory bowel disease in never users and ever users of metformin [A, Gray’s test p < 0.01] and among never users and tertiles of cumulative duration of metformin therapy [B, Gray’s test p < 0.01].

Table 2 shows the incidence of IBD and the hazard ratios by metformin exposure. A significantly lower risk in ever users was demonstrated by the overall hazard ratios for ever vs never users and a dose–response relationship could be seen in the tertile analyses. For ever vs never users, a significant 45% risk reduction was observed [hazard ratio: 0.55, 95% confidence interval: 0.51–0.60, p < 0.01]. In the tertile analysis, patients who had used metformin for <26 months in the first tertile had a neutral association. A significantly lower risk was observed for patients who had used metformin for a longer duration in the second tertile [26.0–58.3 months] and third tertile [>58.3 months]. In the analysis that divided patients according to the use of different antidiabetic drugs, it was evident that a combination of use of metformin in patients who were treated with other oral antidiabetic drugs or who were treated with insulin significantly reduced the risk of IBD. While comparing the two groups of patients who had not been treated with metformin, the risk of IBD was very similar between those who were treated with other oral antidiabetic drugs [reference group] and those treated with insulin [hazard ratio: 0.95, 95% confidence interval: 0.76–1.20].

Table 2.

Incidence rates of inflammatory bowel disease and hazard ratios by metformin exposure

Metformin use	Incident case number	Cases followed	Person-years	Incidence rate [per 100 000 person-years]	Hazard ratio	95% confidence interval	p-value
Never users	750	24 478	101 168.5	741.3	1.00
Ever users	6466	340 211	1569 529.3	412.0	0.55	[0.51–0.60]	<0.01
Tertiles of cumulative duration of metformin therapy [months]
Never users	750	24 478	101 168.5	741.3	1.00
<26.0	2913	112 089	376 343.5	774.0	1.00	[0.93–1.09]	0.92
26.0–58.3	2318	112 425	538 191.1	430.7	0.57	[0.52–0.62]	<0.01
>58.3	1235	115 697	654 994.7	188.6	0.24	[0.22–0.26]	<0.01
Antidiabetic subgroups
OADs without metformin	79	2806	9843.1	802.6	1.00
OADs with metformin	1144	55 629	263 871.0	433.6	0.52	[0.42–0.66]	<0.01
Insulin without metformin [with or without other OADs]	671	21 672	91 325.4	734.7	0.95	[0.76–1.20]	0.69
Insulin with metformin [with or without other OADs]	5322	28 4582	1305658.3	407.6	0.50	[0.40–0.62]	<0.01

OADs: oral antidiabetic drugs.

The sensitivity analyses shown in Table 3 consistently supported a lower risk of IBD associated with metformin use. The significant risk reduction of IBD associated with metformin use was similarly demonstrated in all three age subgroups of <50, 50–64 and ≥65 years, although the benefit appears to be attenuated with increasing age.

Table 3.

Sensitivity analyses evaluating incidence rates of inflammatory bowel disease and hazard ratios for ever vs never users of metformin

Model/metformin use	Incident case number	Cases followed	Person-years	Incidence rate [per 100 000 person-years]	Hazard ratio	95% confidence interval	p-value
I. Censoring patients from the time 4 months have elapsed since the last prescription
Never users	750	24 478	101 168.5	741.3	1.00
Ever users	5495	340 211	1357 116.5	404.9	0.55	[0.51–0.59]	<0.01
II. Excluding patients who received other antidiabetic drugs before metformin was initiated
Never users	750	24 478	101 168.5	741.3	1.00
Ever users	2924	158 131	740 632.6	394.8	0.53	[0.49–0.57]	<0.01
III. Excluding patients followed up for <12 months
Never users	561	22 592	99 767.4	562.3	1.00
Ever users	5508	328 656	1560 866.9	352.9	0.62	[0.57–0.67]	<0.01
IV. Excluding metformin use <12 months
Never users	750	24 478	101 168.5	741.3	1.00
Ever users	5026	285 390	1405 715.7	357.5	0.47	[0.44–0.51]	<0.01
V. Patients enrolled during 1999–2002
Never users	308	11 046	44 594.4	690.7	1.00
Ever users	3652	188 537	883 387.2	413.4	0.59	[0.53–0.67]	<0.01
VI. Patients enrolled during 2003–2005
Never users	442	13 432	56 574.0	781.3	1.00
Ever users	2814	151 674	686 142.1	410.1	0.52	[0.47–0.58]	<0.01
VII. Excluding two consecutive prescriptions of metformin spanning more than 4 months
Never users	750	24 478	101 168.5	741.3	1.00
Ever users	1732	104 657	454 376.0	381.2	0.51	[0.47–0.56]	<0.01
VIII. Excluding patients treated with incretin-based therapies during follow-up
Never users	739	23 056	94 832.0	779.3	1.00
Ever users	6047	268 841	1206 421.9	501.2	0.64	[0.59–0.69]	<0.01
IX. Inflammatory bowel disease defined by diagnostic codes plus operation codes
Never users	20	25 642	107 505.7	18.6	1.00
Ever users	121	361 210	1679 130.7	7.2	0.39	[0.24–0.62]	<0.01
X. Age ˂ 50 years
Never users	122	3381	15 127.8	806.5	1.00
Ever users	1382	75 346	366 448.8	377.1	0.47	[0.39–0.56]	<0.01
XI. Age 50–64 years
Never users	290	8169	35 719.5	811.9	1.00
Ever users	2882	149 710	707 159.0	407.6	0.50	[0.44–0.56]	<0.01
XII. Age ≥65 years
Never users	338	12 928	50 321.2	671.7	1.00
Ever users	2202	115 155	495 921.6	444.0	0.66	[0.59–0.74]	<0.01

The multivariate-adjusted odds ratios [95% confidence intervals] estimated by logistic regression for prevalent users vs non-users of metformin for the periods 1999–2005, 2006–2011 and 1999–2011 were 0.62 [0.55–0.70], 0.52 [0.47–0.58] and 0.56 [0.52–0.61], respectively. Therefore, the findings seemed to be robust even if prevalent users of metformin were analysed.

The multivariate-adjusted hazard ratio for death for ever vs never users was 0.67 [95% confidence interval: 0.64–0.69, p < 0.01], suggesting that the competing risk of death was not likely to bias the estimate in favour of metformin in the present study.

4. Discussion

This is the first study that has shown a reduced risk of IBD associated with metformin use in a dose–response pattern. Furthermore, it is evident that a combination of metformin in patients who used other oral antidiabetic drugs or insulin could reduce the risk of IBD in the respective treatment groups.

The mechanisms of a reduced risk of IBD associated with metformin use require further investigation, but some basic research may provide reasonable explanations. Gut immune–microbe interactions and inflammation are important features in the pathogenesis of IBD.¹³ Metformin may alleviate systemic inflammation in patients with IBD by improving insulin resistance.¹⁴ Furthermore, site-specific inflammation in the intestine may also be reduced because metformin has been shown to inhibit the expression of proinflammatory cytokines and chemokines in intestinal smooth muscle cells¹⁵ and epithelial cells^4,5 in in vitro and in vivo studies.

Patients with IBD show compositional changes of gut microbiota with reduced amounts of Akkermansia species.¹⁶ Studies have suggested that administration of Akkermansia muciniphila ameliorated ulcerative colitis induced by dextran sulphate sodium in mice¹⁷ while metformin induced the proliferation of A. muciniphila.^11,18

Metformin modulates the production of post-biotics such as short-chain fatty acids and tryptophan,¹⁶ modulates the action of vitamin D, and increases the production of ketone bodies and glucagon-like peptide [GLP] in the gut. Short-chain fatty acids protect against IBD by providing energy to colonocytes, promoting the function of regulatory T cells, affecting cytokine production with an anti-inflammatory effect and enhancing intestinal barrier function by tightening the tight junctions.^3,16,19 Metformin induces the production of short-chain fatty acids from several microbiota.^19,20 Vitamin D is an immunoregulatory factor and its insufficiency or deficiency is associated with IBD.³ Although vitamin D levels did not change after treatment with metformin for 16 months in patients with type 2 diabetes mellitus,²¹ this does not exclude the potential effect of metformin on the activity of vitamin D. Butyrate, which is increased with metformin treatment,²⁰ upregulates the expression of vitamin D receptors in epithelial cells.³ Fasting or a calorie-restricted diet induce the production of ketone bodies, and β-hydroxybutyrate [one of the ketone bodies] exhibits anti-inflammatory properties in the intestine.³ Metformin use is associated with energy deprivation in the cell and induces the production of ketone bodies.²² Both GLP-1²³ and metformin¹⁵ exert anti-inflammatory effects on intestinal smooth muscle cells isolated from male BALB/c mice. Metformin may inhibit the activity of dipeptidyl peptidase IV²⁴ and upregulates the expression of GLP-1/GLP-2 in the gut.¹⁹

Gut epithelial differentiation and barrier function are regulated by 5′ adenosine monophosphate-activated protein kinase [AMPK].²⁵ Metformin, through AMPK activation, restores the tight junction in intestinal epithelium in colitis induced by dextran sulphate sodium in mice²⁶; it also protects against intestinal barrier dysfunction induced by lipopolysaccharide²⁷ or dextran sodium sulphate.⁵

IBD is characterized by innate immune responses to bacteria with false pattern recognition, resulting in the activation of nuclear factor-κB [NF-κB] and mitogen-activated protein kinase [MAPK] and defects in apoptosis of inflammatory cells.^2,13 Interestingly, metformin inhibits the signalling of NF-κB and MAPK²⁸ and induces apoptosis.²⁹ Metformin modulates autoimmune activity,³⁰ which probably also accounts for its beneficial effect on IBD. Finally, the epithelial to mesenchymal transition [EMT] is a major feature of intestinal fibrosis in IBD¹⁶ and metformin may attenuate EMT.³¹

Taking the evidence together, metformin may reduce IBD risk via multiple mechanisms by targeting insulin resistance, changes in the gut microbiota and metabolism of nutrients, an energy shift to ketone bodies, inhibition of inflammation and modulating immunity.

This study has addressed methodological limitations such as selection bias, prevalent user bias, immortal time bias and confounding by indication. Because the database covers >99% of the population, selection bias could be avoided. Prevalent user bias was prevented by enrolling patients with new-onset diabetes mellitus and new users of metformin.

The follow-up period during which the outcome cannot occur is referred to as the immortal time. Immortal time bias can be introduced when treatment status or follow-up time are not appropriately assigned. We excluded patients with an uncertain diagnosis of diabetes mellitus by enrolling patients with two or more prescriptions of antidiabetic drugs. Because the database keeps all prescription information during the long follow-up period, the treatment status of metformin was unlikely to be misclassified. The immortal time between the diagnosis of diabetes mellitus and the initiation of antidiabetic drugs was not included in the calculation of person-years. To avoid the effect of immortal time during a short duration at the start of follow-up, we excluded patients who had been followed up for <6 months in the main analyses and further excluded patients who had been followed up for <12 months in one of the sensitivity analyses. All analyses consistently supported a lower risk of IBD associated with metformin use. The effect of immortal time during the waiting period between drug prescription and drug dispense at hospital discharge can be observed in some countries, but not in Taiwan because the patients receive all discharge drugs at the hospital at the time of discharge.

We reduced the effect of confounding by indication by using the Cox regression incorporated with IPTW in modelling. This method has been demonstrated to allow the estimation of hazard ratios with minimal bias.¹⁰ However, note that this statement only holds for measured confounders, and not for unmeasured confounders. Therefore, additional studies including other unmeasured potential confounders in statistical analyses are necessary to confirm our findings. Our consistent findings in the sensitivity analyses and in the cross-sectional analyses suggest that the conclusions are robust and solid.

‘A competing risk is an event whose occurrence precludes the occurrence of the primary event of interest’.³² Therefore, death is a potential competing risk that might have biased the results of the study. In data analyses, patients who died were censored at the time of death in fitting the Cox models and this is one way recommended for fitting models in the presence of competing risk.³² Additionally, we followed the recommendation of the use of cumulative incidence functions in the presence of competing risks because estimates of incidence with Kaplan–Meier survival functions may lead to upward bias.³² Furthermore, because metformin ever users had a lower risk of death than never users, the competing risk of death was not likely to bias the estimate in favour of metformin.

Study limitations include a lack of measurement data of confounders such as biochemical, humoral and autoimmune profiles, anthropometric factors such as body mass index and waist circumference, lifestyle, psychological stress, physical activity, dietary pattern, history of constipation and diarrhoea, history of vaccination, cigarette smoking, alcohol intake, family history and genetic parameters. Because the diagnosis of IBD was based on ICD-9-CM codes without laboratory data and endoscopic findings, it remains possible that misclassification of IBD might have occurred. However, the hazard ratios would be underestimated only if the misclassifications were not differential.³³ Underestimation of the hazard ratio for ever vs never users in the main analyses was evident because a hazard ratio that considered a more stringent diagnostic criterion of IBD by diagnostic codes plus operation codes in sensitivity analysis diverged further from unity. Finally, we did not have information regarding gut microbiota for in-depth investigation on their roles on IBD and their interactions with metformin.

5. Conclusions

The findings support a lower risk of IBD associated with metformin use in patients with type 2 diabetes mellitus. Because metformin as a monotherapy does not cause hypoglycaemia and it is inexpensive and safe for long-term use, the prevention and/or treatment of IBD with metformin is worthy of confirmation and more in-depth investigation, not only in patients with diabetes mellitus but also in non-diabetes patients.

Funding

The study was supported by the Ministry of Science and Technology, Taiwan [MOST 107-2221-E-002-129-MY3]. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Conflict of Interest

The author declares no conflicts of interest.

Acknowledgments

The study is based in part on data from the National Health Insurance Research Database provided by the Bureau of National Health Insurance, Department of Health and managed by National Health Research Institutes. The interpretation and conclusions contained herein do not represent those of the Bureau of National Health Insurance, Department of Health or National Health Research Institutes.

Author Contributions

C.H.T. researched the data and wrote the manuscript.