- Split View
-
Views
-
Cite
Cite
Ka Shing Cheung, Esther W Chan, Angel Y S Wong, Lijia Chen, Wai Kay Seto, Ian C K Wong, Wai K Leung, Metformin Use and Gastric Cancer Risk in Diabetic Patients After Helicobacter pylori Eradication, JNCI: Journal of the National Cancer Institute, Volume 111, Issue 5, May 2019, Pages 484–489, https://doi.org/10.1093/jnci/djy144
- Share Icon Share
Abstract
Although prior studies showed metformin could reduce gastric cancer (GC) risk in patients with diabetes mellitus, they failed to adjust for Helicobacter pylori infection and glycemic control. We aimed to investigate whether metformin reduced GC risk in H. pylori-eradicated diabetic patients and its association with glycemic control.
This was a territory-wide cohort study using hospital registry database, recruiting all diabetic patients who were prescribed clarithromycin-based triple therapy for H. pylori infection from 2003 to 2012. Subjects were observed from H. pylori therapy prescription until GC diagnosis, death, or end of study (December 2015). Exclusion criteria included GC diagnosed within first year of H. pylori therapy, prior history of GC or gastrectomy, and failure of H. pylori eradication. The hazard ratio (HR) of GC with metformin (defined as at least 180-day use) was estimated by Cox model with propensity score adjustment for covariates (age, sex, comorbidities, medications [including insulin], and time-weighted average hemoglobin A1c [HbA1c]). All statistical tests were two-sided.
During a median follow-up of 7.1 years (IQR = 4.7–9.8), 37 (0.51%) of 7266 diabetic patients developed GC at a median age of 76.4 years (IQR = 64.8–81.5 years). Metformin use was associated with a reduced GC risk (adjusted HR = 0.49, 95% CI = 0.24 to 0.98). There was a trend towards a lower GC risk with increasing duration (Ptrend = .01) and dose of metformin (Ptrend = .02). HbA1c level was not an independent risk factor for GC.
Metformin use was associated with a lower GC risk among H. pylori-eradicated diabetic patients in a duration- and dose-response manner, which was independent of HbA1c level.
Gastric cancer (GC) is the fifth most common cancer and the third leading cause of cancer-related mortality worldwide (1). Helicobacter pylori infection is the major risk factor for GC (2). However, H. pylori eradication could reduce GC risk by only approximately 40% (3–5). Apart from H. pylori, diabetes mellitus (DM) has been reported to increase GC risk by approximately 20% (6). Because DM is a very prevalent medical condition with more than 12% of the adult population being affected (7), the burden of GC cases attributed to DM could be substantial.
Metformin, a biguanide, is frequently used to treat DM. Apart from its effect on glycemic control, metformin has anticancer effects associated with insulin sensitization, reducing hyperinsulinemia and insulin-growth factor (IGF) production, both of which have been shown to enhance proliferation of cancer cells that express IGF receptors (8). Metformin also activates AMP-activated protein kinase, which suppresses cancer cell growth by inhibiting the mammalian target of rapamycin pathway (9). The chemopreventive role of metformin in GC, however, remains controversial. Whereas no association between metformin use and GC was reported by some studies (10,11), others suggested a protective effect with varying effect estimates (12–16). A recent meta-analysis (17) concluded that metformin decreased GC risk by 24%, but there was a statistically significant heterogeneity among studies. More importantly, other important risk factors of GC including H. pylori infection and DM severity have not been adequately addressed in previous studies, potentially undermining the role of metformin on GC prevention (17). Because H. pylori is the most important risk factor of GC, failure to stratify patients according to H. pylori status will affect the true effect estimate of metformin on GC development. Moreover, GC risk was shown to be higher among individuals with higher hemoglobin A1c (HbA1c) levels (18). The beneficial effect of metformin on reducing GC could be mediated via improving DM control rather than the proposed anticancer mechanisms.
With the use of a large cohort of diabetic patients who had received H. pylori eradication therapy, we aimed to investigate the potential chemopreventive effect of metformin on GC in diabetic patients who had received H. pylori eradication therapy and whether higher HbA1c level was associated with an increased GC risk.
Methods
Data Source
This was a territory-wide cohort study using data retrieved from the Clinical Data Analysis and Reporting System (CDARS) of the Hong Kong Hospital Authority. The Hospital Authority is the governing body of all public hospitals and clinics in Hong Kong, with a population of 7.3 million and a 90% coverage of all primary, secondary, and tertiary care during the study period (2003–2015) (19). Patient’s data including demographics, diagnoses, drug dispensing records, procedures and laboratory results, hospitalization, attendance of outpatient clinics and emergency departments, and death are all accessible in the CDARS (20–25). Diagnoses are coded in accordance with the International Classification of Diseases, Ninth Revision (ICD-9). Prior studies using this electronic registry database showed high coding accuracy with positive and negative predictive values of less than 90 (21,26). Individuals’ information is anonymized with a unique reference key to protect patient’s confidentiality. The study was approved by the Institutional Review Board of the University of Hong Kong and the West Cluster of Hospital Authority, Hong Kong (reference no: UW 16–545).
Study Subjects
All adult patients aged 18 years or above with a baseline diagnosis of DM who were dispensed a course of clarithromycin-based triple therapy for H. pylori between January 1, 2003 and December 31, 2012 were identified. The diagnosis of DM was based on the ICD-9 codes of DM (ICD-9 codes: 249 and 250). The prescription of clarithromycin-based triple therapy included the coprescription of one of the proton pump inhibitors (PPIs) with clarithromycin and either amoxicillin or metronidazole with the correct doses, same prescription start dates, and a treatment duration of 7 to 14 days (27). With the high H. pylori eradication rate (>90%) and low resistance rate to clarithromycin (8%) in Hong Kong (28), clarithromycin-based triple therapy was the first-line H. pylori treatment during the study period (29). A diagnosis of H. pylori infection was made by either biopsy-based tests (rapid urease test and histology) or urea breath test, because other diagnostic modalities were not available in the public hospitals. Exclusion criteria included: gastric lymphoma; GC diagnosed within the first year of H. pylori eradication therapy (as there was a possibility of delayed or missed diagnosis); prior history of GC; prior gastrectomy; and failure of H. pylori eradication. Because there was no ICD-9 code for failure of H. pylori eradication, this could only be inferred from the repeated prescription of clarithromycin-based triple therapy or subsequent prescriptions of either a second-line therapy (either PPI-levofloxacin-amoxycillin or bismuth-based quadruple therapy) or a third-line therapy (rifabutin-based therapy). The patient selection process is illustrated in Supplementary Figure 1 (available online).
Study Outcome
Gastric adenocarcinoma after H. pylori eradication therapy was identified by the ICD-9 coding (Supplementary Table 1, available online). The validation of the GC diagnosis and final H. pylori status of these cancer patients had been validated in our previous studies with 100% accuracy (30). The observation period started from the first date of H. pylori therapy prescription (ie, index date) and censored at GC diagnosis, death, or end of study (December 2015). The earliest date of hospitalization for cancer workup or treatment was regarded as the GC diagnosis date.
Metformin, Insulin, and Other Covariates
We defined metformin use as more than 180-day use [as proposed by Kim et al. (13)] after receiving H. pylori eradication therapy. To study the dose-response relationship of metformin on GC risk, cumulative defined daily dose (cDDD) as per the World Health Organization Collaborating Center for Drug Statistics Methodology (31) was calculated by summing the dispensed DDDs of metformin during the study observation period and categorized into non-metformin use, use of cDDDs below the median, and use of cDDDs equal to or above the median. To study the duration-response relationship, the duration of metformin use was divided into three groups: non-metformin use, less than 3 years, and 3 years or more, as defined previously by Kim et al. (13)
Potential risk factors for GC included the age of receiving H. pylori eradication therapy, sex, smoking and alcohol use, past history of peptic ulcer disease, other comorbidities (atrial fibrillation, ischemic heart disease, congestive heart failure, chronic renal failure, cirrhosis, stroke, dyslipidemia, hypertension, and obesity), concurrent medications uses (aspirin, nonsteroidal antiinflammatory drugs, cyclooxygenase-2 [COX-2] inhibitors (32,33), statins (34), PPIs (30,35), insulin (36,37)), and HbA1c (18) were considered during analysis. Smoking and alcohol statuses were ascertained as reported previously (38). Smoking was identified by the documentation of the smoking status in the CDARS, the ICD-9 code of V15.82, or indirectly by the presence of chronic obstructive pulmonary disease. Alcohol use was signified by the presence of alcohol-related diseases, comprising gastrointestinal, hepatic, psychiatric, and neurological diseases. Obesity was identified by the ICD-9 codes of 278.0 and 278.1 or a body mass index at least 25 kg/m2. Supplementary Table 1 (available online) shows the diagnostic codes of all variables. To adjust for the bias due to irregular interval measurements, time-weighted average HbA1c was used to represent the overall glycemic control during the observation period as described by Cheuk-Fung Yip et al. (39) This was derived as the average HbA1c weighted by the time interval between successive measurements. We categorized the time-weighted average HbA1c into a binary variable by a cut-off value of 7% as individuals with HbA1c at least 7% were at a higher risk of GC than subjects without DM in a previous population-based cohort study (18). For consistency, we used a cut-off of 180 days to define the use of other medications. We also investigated the association between insulin use and GC, because IGFs have been proposed to be involved in GC development (36,37,40).
Statistical Analyses
All statistical analyses were performed using R version 3.2.3 (R Foundation for Statistical Computing) statistical software. Continuous variables were expressed as median with interquartile range (IQR) and categorical variables were presented as number (percentage). Comparison of two groups was analyzed by Mann-Whitney U-test for continuous variables, and χ2 or Fisher’s exact test for categorical variables. The Cox proportional hazards model was used to determine the hazard ratio (HR) of GC with metformin use. We checked whether the Cox proportional-hazard assumption was fulfilled by a “complementary log-log”-scaled Kaplan-Meier plot and Schoenfeld residuals (P > .05, which indicated no interaction of the covariates with time).
Propensity score (PS) adjustment was used to control for selection bias due to different baseline characteristics. The PS was estimated by multivariable logistic regression based on the aforementioned covariates and represents the probability of prescribing metformin to an individual given the covariates. As such, any difference in GC risk would be due to metformin effect only. The PS distributions between metformin and non-metformin groups were compared graphically, showing no statistically significant areas of nonoverlap that would otherwise violate the assumption of PS analysis (Supplementary Figure 2, available online) (41).
The primary analysis was Cox regression with PS adjustment. The PS adjusted absolute difference in GC risk between metformin and non-metformin users was calculated as follows: (adjusted HR – 1) x (crude incidence rate of GC in non-metformin users) (42).
Sensitivity analysis was conducted to assess robustness of the results, including multivariable analysis, PS adjustment after trimming of individuals in the nonoverlapping parts of PS distribution (43), PS weighting by inverse probability treatment weighting (IPTW) with stabilization (44), and PS matching without replacement in a 1:1 ratio (45). The description of these PS analysis methods was detailed in Supplementary Table 2 (available online). Standardized difference was used to assess the balance of the covariates between the two groups before and after IPTW and PS matching, and a value of less than 0.2 indicated a negligible difference (46). Supplementary Table 3 (available online) shows that most of the covariates were balanced after either IPTW or PS matching. Supplementary Figures 2 and 3 (available online) also show that PS distriubtion between the two groups became largely similar after PS matching.
Similar statistical analyses were performed to determine the association between insulin use and GC. A two-sided P value of less than .05 was used to define statistical significance. The R packages “survival” and “MatchIt” were used for Cox regression analysis and PS matching, respectively.
Results
Patient Characteristics
A total of 7266 diabetic patients (type I: 30, type II: 3394, unclassified: 3842) who had received H. pylori eradication therapy during the study period were included. The median age of receiving clarithromycin-based triple therapy was 65.2 years (IQR: 56.1–74.2), and 52.0% were male. There were 5368 (73.9%) metformin users and the use of other diabetic medications was shown in Table 1.
Characteristic . | All, No. (%) (n = 7266) . | Metformin users, No. (%) (n = 5368) . | Non-metformin users, No. (%) (n = 1898) . |
---|---|---|---|
Median age at triple therapy, y (IQR) | 65.2 (56.1–74.2) | 63.8 (55.6–72.6) | 69.7 (58.2–78.2) |
Male sex | 3779 (52.0) | 2716 (50.6) | 1063 (56.0) |
Median duration of follow-up, y (IQR) | 7.1 (4.7–9.8) | 7.5 (5.2–10.1) | 5.8 (3.5–8.8) |
Time-weighted average HbA1c ≥ 7% | 4191 (57.7) | 3437 (64.0) | 754 (39.7) |
Smoking | 1265 (17.4) | 945 (17.6) | 320 (16.9) |
Alcohol | 116 (1.6) | 61 (1.1) | 55 (2.9) |
History of gastric ulcer | 281 (3.9) | 180 (3.4) | 101 (5.3) |
History of duodenal ulcer | 295 (4.1) | 198 (3.7) | 97 (5.1) |
Hypertension | 4503 (62.0) | 3246 (60.5) | 1257 (66.2) |
Dyslipidemia | 1982 (27.3) | 1483 (27.6) | 499 (26.3) |
Obesity | 1288 (17.7) | 1097 (20.4) | 191 (10.1) |
Ischemic heart disease | 1864 (25.7) | 1318 (24.6) | 546 (28.8) |
Atrial fibrillation | 649 (8.9) | 426 (7.9) | 223 (11.7) |
Congestive heart failure | 1001 (13.8) | 579 (10.8) | 422 (22.2) |
Stroke | 1341 (18.5) | 935 (17.4) | 406 (21.4) |
Chronic renal failure | 770 (10.6) | 346 (6.4) | 424 (22.3) |
Cirrhosis | 274 (3.8) | 148 (2.8) | 126 (6.6) |
Aspirin/NSAIDs/COX-2 inhibitors* | 3457 (47.6) | 2649 (49.3) | 808 (42.6) |
Statins* | 4375 (60.2) | 3562 (66.4) | 813 (42.8) |
Proton pump inhibitors* | 966 (13.3) | 654 (12.2) | 312 (16.4) |
Insulin* | 2075 (28.6) | 1575 (29.3) | 500 (26.3) |
Sulphonylureas* | 4004 (74.6) | 4004 (74.6) | 772 (40.7) |
Acarbose* | 279 (3.8) | 239 (4.5) | 40 (2.1) |
Glitazones* | 288 (4.0) | 254 (4.7) | 34 (1.8) |
Dipeptidyl peptidase IV inhibitors* | 842 (11.6) | 781 (14.5) | 61 (3.2) |
Characteristic . | All, No. (%) (n = 7266) . | Metformin users, No. (%) (n = 5368) . | Non-metformin users, No. (%) (n = 1898) . |
---|---|---|---|
Median age at triple therapy, y (IQR) | 65.2 (56.1–74.2) | 63.8 (55.6–72.6) | 69.7 (58.2–78.2) |
Male sex | 3779 (52.0) | 2716 (50.6) | 1063 (56.0) |
Median duration of follow-up, y (IQR) | 7.1 (4.7–9.8) | 7.5 (5.2–10.1) | 5.8 (3.5–8.8) |
Time-weighted average HbA1c ≥ 7% | 4191 (57.7) | 3437 (64.0) | 754 (39.7) |
Smoking | 1265 (17.4) | 945 (17.6) | 320 (16.9) |
Alcohol | 116 (1.6) | 61 (1.1) | 55 (2.9) |
History of gastric ulcer | 281 (3.9) | 180 (3.4) | 101 (5.3) |
History of duodenal ulcer | 295 (4.1) | 198 (3.7) | 97 (5.1) |
Hypertension | 4503 (62.0) | 3246 (60.5) | 1257 (66.2) |
Dyslipidemia | 1982 (27.3) | 1483 (27.6) | 499 (26.3) |
Obesity | 1288 (17.7) | 1097 (20.4) | 191 (10.1) |
Ischemic heart disease | 1864 (25.7) | 1318 (24.6) | 546 (28.8) |
Atrial fibrillation | 649 (8.9) | 426 (7.9) | 223 (11.7) |
Congestive heart failure | 1001 (13.8) | 579 (10.8) | 422 (22.2) |
Stroke | 1341 (18.5) | 935 (17.4) | 406 (21.4) |
Chronic renal failure | 770 (10.6) | 346 (6.4) | 424 (22.3) |
Cirrhosis | 274 (3.8) | 148 (2.8) | 126 (6.6) |
Aspirin/NSAIDs/COX-2 inhibitors* | 3457 (47.6) | 2649 (49.3) | 808 (42.6) |
Statins* | 4375 (60.2) | 3562 (66.4) | 813 (42.8) |
Proton pump inhibitors* | 966 (13.3) | 654 (12.2) | 312 (16.4) |
Insulin* | 2075 (28.6) | 1575 (29.3) | 500 (26.3) |
Sulphonylureas* | 4004 (74.6) | 4004 (74.6) | 772 (40.7) |
Acarbose* | 279 (3.8) | 239 (4.5) | 40 (2.1) |
Glitazones* | 288 (4.0) | 254 (4.7) | 34 (1.8) |
Dipeptidyl peptidase IV inhibitors* | 842 (11.6) | 781 (14.5) | 61 (3.2) |
Drug use was defined as use for more than 180 days. COX-2 = cyclooxygenase-2; HbA1c = hemoglobin A1c; NSAIDs = nonsteroidal antiinflammatory drugs.
Characteristic . | All, No. (%) (n = 7266) . | Metformin users, No. (%) (n = 5368) . | Non-metformin users, No. (%) (n = 1898) . |
---|---|---|---|
Median age at triple therapy, y (IQR) | 65.2 (56.1–74.2) | 63.8 (55.6–72.6) | 69.7 (58.2–78.2) |
Male sex | 3779 (52.0) | 2716 (50.6) | 1063 (56.0) |
Median duration of follow-up, y (IQR) | 7.1 (4.7–9.8) | 7.5 (5.2–10.1) | 5.8 (3.5–8.8) |
Time-weighted average HbA1c ≥ 7% | 4191 (57.7) | 3437 (64.0) | 754 (39.7) |
Smoking | 1265 (17.4) | 945 (17.6) | 320 (16.9) |
Alcohol | 116 (1.6) | 61 (1.1) | 55 (2.9) |
History of gastric ulcer | 281 (3.9) | 180 (3.4) | 101 (5.3) |
History of duodenal ulcer | 295 (4.1) | 198 (3.7) | 97 (5.1) |
Hypertension | 4503 (62.0) | 3246 (60.5) | 1257 (66.2) |
Dyslipidemia | 1982 (27.3) | 1483 (27.6) | 499 (26.3) |
Obesity | 1288 (17.7) | 1097 (20.4) | 191 (10.1) |
Ischemic heart disease | 1864 (25.7) | 1318 (24.6) | 546 (28.8) |
Atrial fibrillation | 649 (8.9) | 426 (7.9) | 223 (11.7) |
Congestive heart failure | 1001 (13.8) | 579 (10.8) | 422 (22.2) |
Stroke | 1341 (18.5) | 935 (17.4) | 406 (21.4) |
Chronic renal failure | 770 (10.6) | 346 (6.4) | 424 (22.3) |
Cirrhosis | 274 (3.8) | 148 (2.8) | 126 (6.6) |
Aspirin/NSAIDs/COX-2 inhibitors* | 3457 (47.6) | 2649 (49.3) | 808 (42.6) |
Statins* | 4375 (60.2) | 3562 (66.4) | 813 (42.8) |
Proton pump inhibitors* | 966 (13.3) | 654 (12.2) | 312 (16.4) |
Insulin* | 2075 (28.6) | 1575 (29.3) | 500 (26.3) |
Sulphonylureas* | 4004 (74.6) | 4004 (74.6) | 772 (40.7) |
Acarbose* | 279 (3.8) | 239 (4.5) | 40 (2.1) |
Glitazones* | 288 (4.0) | 254 (4.7) | 34 (1.8) |
Dipeptidyl peptidase IV inhibitors* | 842 (11.6) | 781 (14.5) | 61 (3.2) |
Characteristic . | All, No. (%) (n = 7266) . | Metformin users, No. (%) (n = 5368) . | Non-metformin users, No. (%) (n = 1898) . |
---|---|---|---|
Median age at triple therapy, y (IQR) | 65.2 (56.1–74.2) | 63.8 (55.6–72.6) | 69.7 (58.2–78.2) |
Male sex | 3779 (52.0) | 2716 (50.6) | 1063 (56.0) |
Median duration of follow-up, y (IQR) | 7.1 (4.7–9.8) | 7.5 (5.2–10.1) | 5.8 (3.5–8.8) |
Time-weighted average HbA1c ≥ 7% | 4191 (57.7) | 3437 (64.0) | 754 (39.7) |
Smoking | 1265 (17.4) | 945 (17.6) | 320 (16.9) |
Alcohol | 116 (1.6) | 61 (1.1) | 55 (2.9) |
History of gastric ulcer | 281 (3.9) | 180 (3.4) | 101 (5.3) |
History of duodenal ulcer | 295 (4.1) | 198 (3.7) | 97 (5.1) |
Hypertension | 4503 (62.0) | 3246 (60.5) | 1257 (66.2) |
Dyslipidemia | 1982 (27.3) | 1483 (27.6) | 499 (26.3) |
Obesity | 1288 (17.7) | 1097 (20.4) | 191 (10.1) |
Ischemic heart disease | 1864 (25.7) | 1318 (24.6) | 546 (28.8) |
Atrial fibrillation | 649 (8.9) | 426 (7.9) | 223 (11.7) |
Congestive heart failure | 1001 (13.8) | 579 (10.8) | 422 (22.2) |
Stroke | 1341 (18.5) | 935 (17.4) | 406 (21.4) |
Chronic renal failure | 770 (10.6) | 346 (6.4) | 424 (22.3) |
Cirrhosis | 274 (3.8) | 148 (2.8) | 126 (6.6) |
Aspirin/NSAIDs/COX-2 inhibitors* | 3457 (47.6) | 2649 (49.3) | 808 (42.6) |
Statins* | 4375 (60.2) | 3562 (66.4) | 813 (42.8) |
Proton pump inhibitors* | 966 (13.3) | 654 (12.2) | 312 (16.4) |
Insulin* | 2075 (28.6) | 1575 (29.3) | 500 (26.3) |
Sulphonylureas* | 4004 (74.6) | 4004 (74.6) | 772 (40.7) |
Acarbose* | 279 (3.8) | 239 (4.5) | 40 (2.1) |
Glitazones* | 288 (4.0) | 254 (4.7) | 34 (1.8) |
Dipeptidyl peptidase IV inhibitors* | 842 (11.6) | 781 (14.5) | 61 (3.2) |
Drug use was defined as use for more than 180 days. COX-2 = cyclooxygenase-2; HbA1c = hemoglobin A1c; NSAIDs = nonsteroidal antiinflammatory drugs.
Risk of Gastric Cancer Development and Data Validation
During a median follow-up of 7.1 years (IQR = 4.7–9.8 years) (Table 1) totalling 52 208 person-years, 37 (0.51%) patients were diagnosed to have GC (incidence rate = 7.1 per 10 000 person-years), with 21 (56.8%) in the noncardia region, 12 (32.4%) in the cardia, and unspecified sites in the remaining 4 (10.8%) cases. The median age at GC diagnosis was 76.4 years (IQR = 64.8–81.5 years). The median duration from receiving H. pylori eradication therapy to GC development was 4.4 years (IQR = 3.1–6.8; metformin group: 5.1 [IQR = 3.5–7.6] vs non-metformin group: 3.9 [IQR = 1.8–6.2]; P = .12) (data not shown).
Effect of Metformin, Insulin, and HbA1c on Gastric Cancer Risk
The median duration of metformin use was 5.5 years (IQR = 3.3–8.4), and the median cDDD was 975 (IQR = 436–1837). Among metformin users, 20 (0.37%) developed GC, with an incidence rate of 4.9 per 10 000 person-years. There was a statistically significantly lower GC risk among metformin users (PS adjusted HR = 0.49; 95% CI = 0.24 to 0.98) (Table 2). Sensitivity analysis by different statistical methods showed similar results (Table 2). The PS-adjusted absolute risk difference between metformin and non-metformin use was 7.60 fewer cancer cases (95% CI = 0.30 to 11.33) per 10 000 person-years (data not shown). When patients with type I DM were excluded (n = 7236, GC =36), the PS adjusted HR was 0.47 (95% CI = 0.23 to 0.96) (data not shown).
Analysis . | No. of patients . | No. of GC events . | HR of GC with metformin use (95% CI) . | P† . |
---|---|---|---|---|
Unadjusted analysis | 7266 | 37 | 0.32 (0.17 to 0.61) | <.001 |
Multivariable analysis | 7266 | 37 | 0.46 (0.23 to 0.93) | .03 |
PS adjustment (without trimming) | 7266 | 37 | 0.49 (0.24 to 0.98) | .045 |
PS adjustment after trimming | 7253 | 37 | 0.49 (0.24 to 0.98) | .045 |
PS weighting by inverse probability treatment weighting | 7266 | 37 | 0.47 (0.23 to 0.96) | .03 |
PS matching | 3608 | 24 | 0.36 (0.15 to 0.87) | .02 |
Analysis . | No. of patients . | No. of GC events . | HR of GC with metformin use (95% CI) . | P† . |
---|---|---|---|---|
Unadjusted analysis | 7266 | 37 | 0.32 (0.17 to 0.61) | <.001 |
Multivariable analysis | 7266 | 37 | 0.46 (0.23 to 0.93) | .03 |
PS adjustment (without trimming) | 7266 | 37 | 0.49 (0.24 to 0.98) | .045 |
PS adjustment after trimming | 7253 | 37 | 0.49 (0.24 to 0.98) | .045 |
PS weighting by inverse probability treatment weighting | 7266 | 37 | 0.47 (0.23 to 0.96) | .03 |
PS matching | 3608 | 24 | 0.36 (0.15 to 0.87) | .02 |
Metformin use was defined as use for more than 180 days.
Cox proportional hazards model was used to calculate the P values, and the test was two-sided. CI = confidence interval; HR = hazard ratio; PS = propensity score.
Analysis . | No. of patients . | No. of GC events . | HR of GC with metformin use (95% CI) . | P† . |
---|---|---|---|---|
Unadjusted analysis | 7266 | 37 | 0.32 (0.17 to 0.61) | <.001 |
Multivariable analysis | 7266 | 37 | 0.46 (0.23 to 0.93) | .03 |
PS adjustment (without trimming) | 7266 | 37 | 0.49 (0.24 to 0.98) | .045 |
PS adjustment after trimming | 7253 | 37 | 0.49 (0.24 to 0.98) | .045 |
PS weighting by inverse probability treatment weighting | 7266 | 37 | 0.47 (0.23 to 0.96) | .03 |
PS matching | 3608 | 24 | 0.36 (0.15 to 0.87) | .02 |
Analysis . | No. of patients . | No. of GC events . | HR of GC with metformin use (95% CI) . | P† . |
---|---|---|---|---|
Unadjusted analysis | 7266 | 37 | 0.32 (0.17 to 0.61) | <.001 |
Multivariable analysis | 7266 | 37 | 0.46 (0.23 to 0.93) | .03 |
PS adjustment (without trimming) | 7266 | 37 | 0.49 (0.24 to 0.98) | .045 |
PS adjustment after trimming | 7253 | 37 | 0.49 (0.24 to 0.98) | .045 |
PS weighting by inverse probability treatment weighting | 7266 | 37 | 0.47 (0.23 to 0.96) | .03 |
PS matching | 3608 | 24 | 0.36 (0.15 to 0.87) | .02 |
Metformin use was defined as use for more than 180 days.
Cox proportional hazards model was used to calculate the P values, and the test was two-sided. CI = confidence interval; HR = hazard ratio; PS = propensity score.
A total of 2075 (28.6%) diabetic patients used insulin (Table 1). The median duration of insulin use was 4.7 years (IQR = 1.3–7.2; metformin group: 4.7 [IQR = 1.2–10.1] vs non-metformin group: 4.9 [IQR = 1.4–11.8]). The PS-adjusted HR of GC with insulin use was 0.81 (95% CI = 0.35 to 1.85) (data not shown).
In total, 11 286 HbA1c measurements were taken at baseline and during follow-up with a median of 13 measurements (IQR = 7–22 times) per patient at a median interval of 4.7 months (IQR = 3.4–7.9) (data not shown). For the whole cohort, the median baseline and time-weighted averaged HbA1c were 7.3% (IQR = 6.5–8.7%) and 7.2% (IQR = 6.6–7.9%), respectively. A higher proportion of metformin users had a time-weighted HbA1c level of at least 7% than non-metformin users (64.0% vs 39.7%) (Table 1). The corresponding median level of baseline HbA1c in the metformin and non-metformin groups was 7.4% (IQR = 6.6–8.9%) and 6.8% (IQR = 6.1–8.0%), whereas the time-weighted averaged HbA1c was 7.3% (IQR = 6.8–8.0%) and 6.7% (IQR = 6.2–7.6%), respectively (data not shown). Compared with patients with a time-weighted average HbA1c level of less than 7%, patients with a higher level did not have an increased GC risk on either unadjusted (HR = 0.83; 95% CI = 0.60 to 1.15) or multivariable analysis (HR = 1.60; 95% CI = 0.78 to 3.27) (Table 3).
Variables . | Unadjusted analysis . | Multivariable analysis* . | ||
---|---|---|---|---|
HR (95% CI) . | P† . | HR (95% CI) . | P† . | |
Metformin‡ | 0.32 (0.17 to 0.61) | <.001 | 0.46 (0.23 to 0.93) | .03 |
Insulin‡ | 0.80 (0.38 to 1.70) | .57 | 0.76 (0.33 to 1.75) | .52 |
Time-weighted average HbA1c ≥ 7% | 0.83 (0.60 to 1.15) | .27 | 1.60 (0.78 to 3.27) | .20 |
Variables . | Unadjusted analysis . | Multivariable analysis* . | ||
---|---|---|---|---|
HR (95% CI) . | P† . | HR (95% CI) . | P† . | |
Metformin‡ | 0.32 (0.17 to 0.61) | <.001 | 0.46 (0.23 to 0.93) | .03 |
Insulin‡ | 0.80 (0.38 to 1.70) | .57 | 0.76 (0.33 to 1.75) | .52 |
Time-weighted average HbA1c ≥ 7% | 0.83 (0.60 to 1.15) | .27 | 1.60 (0.78 to 3.27) | .20 |
Adjusted for age at triple therapy, sex, smoking, alcoholism, history of gastric ulcer, history of duodenal ulcer, other comorbidities (hypertension, dyslipidemia, obesity, ischemic heart disease, atrial fibrillation, congestive heart failure, stroke, chronic renal failure, cirrhosis), and concurrent medications (statins, aspirin/nonsteroidal antiinflammatory drugs/cyclooxygenase-2 inhibitors, proton pump inhibitors).
Cox proportional hazards model was used to calculate the P values, and the test was two-sided.
Drug use was defined as use for more than 180 days. CI = confidence interval; HR = hazard ratio; HbA1c = hemoglobin A1c.
Variables . | Unadjusted analysis . | Multivariable analysis* . | ||
---|---|---|---|---|
HR (95% CI) . | P† . | HR (95% CI) . | P† . | |
Metformin‡ | 0.32 (0.17 to 0.61) | <.001 | 0.46 (0.23 to 0.93) | .03 |
Insulin‡ | 0.80 (0.38 to 1.70) | .57 | 0.76 (0.33 to 1.75) | .52 |
Time-weighted average HbA1c ≥ 7% | 0.83 (0.60 to 1.15) | .27 | 1.60 (0.78 to 3.27) | .20 |
Variables . | Unadjusted analysis . | Multivariable analysis* . | ||
---|---|---|---|---|
HR (95% CI) . | P† . | HR (95% CI) . | P† . | |
Metformin‡ | 0.32 (0.17 to 0.61) | <.001 | 0.46 (0.23 to 0.93) | .03 |
Insulin‡ | 0.80 (0.38 to 1.70) | .57 | 0.76 (0.33 to 1.75) | .52 |
Time-weighted average HbA1c ≥ 7% | 0.83 (0.60 to 1.15) | .27 | 1.60 (0.78 to 3.27) | .20 |
Adjusted for age at triple therapy, sex, smoking, alcoholism, history of gastric ulcer, history of duodenal ulcer, other comorbidities (hypertension, dyslipidemia, obesity, ischemic heart disease, atrial fibrillation, congestive heart failure, stroke, chronic renal failure, cirrhosis), and concurrent medications (statins, aspirin/nonsteroidal antiinflammatory drugs/cyclooxygenase-2 inhibitors, proton pump inhibitors).
Cox proportional hazards model was used to calculate the P values, and the test was two-sided.
Drug use was defined as use for more than 180 days. CI = confidence interval; HR = hazard ratio; HbA1c = hemoglobin A1c.
Duration and Dose Response of Metformin Use and Gastric Cancer
Table 4 shows the duration and dose response between metformin use and GC. A longer duration of metformin use was associated with a lower GC risk (HR = 0.85; 95% CI = 0.74 to 0.96) for every 1-year increase in use. Compared with non-metformin users, those who used metformin for less than 3 years and 3 years or more had HRs of 0.75 (95% CI = 0.32 to 1.74) and 0.35 (95% CI = 0.16 to 0.80), respectively (Ptrend = .01). For dose effect, the HRs of GC with metformin use for less than 975 cDDD (median cDDD) and 975 and more cDDD, when compared with non-metformin use, were 0.73 (95% CI = 0.35 to 1.53) and 0.33 (95% CI = 0.13 to 0.86), respectively (Ptrend = .02).
Duration and dose . | HR (95% CI) . | P* . | Ptrend† . |
---|---|---|---|
Duration | |||
Per every 1 yr increase in use‡ | 0.85 (0.74 to 0.96) | .01 | |
Metformin use§ | |||
Non-metformin users | 1.00 (Reference) | ||
<3 years | 0.75 (0.32 to 1.74) | .51 | .01 |
≥3 years | 0.35 (0.16 to 0.80) | .01 | |
Dose | |||
Non-metformin users | 1.00 (Reference) | ||
< median, 975 cDDD | 0.73 (0.35 to 1.53) | .40 | .02 |
≥ median, 975 cDDD | 0.33 (0.13 to 0.86) | .02 |
Duration and dose . | HR (95% CI) . | P* . | Ptrend† . |
---|---|---|---|
Duration | |||
Per every 1 yr increase in use‡ | 0.85 (0.74 to 0.96) | .01 | |
Metformin use§ | |||
Non-metformin users | 1.00 (Reference) | ||
<3 years | 0.75 (0.32 to 1.74) | .51 | .01 |
≥3 years | 0.35 (0.16 to 0.80) | .01 | |
Dose | |||
Non-metformin users | 1.00 (Reference) | ||
< median, 975 cDDD | 0.73 (0.35 to 1.53) | .40 | .02 |
≥ median, 975 cDDD | 0.33 (0.13 to 0.86) | .02 |
Cox proportional hazards model was used to calculate the P values, and the test was two-sided.
Cox proportional hazards model was used to calculate the P values, and the test was two-sided.
Continuous variable.
Categorical variable. 95% CI = 95% confidence interval; cDDD = cumulative Defined Daily Dose; HR = hazard ratio.
Duration and dose . | HR (95% CI) . | P* . | Ptrend† . |
---|---|---|---|
Duration | |||
Per every 1 yr increase in use‡ | 0.85 (0.74 to 0.96) | .01 | |
Metformin use§ | |||
Non-metformin users | 1.00 (Reference) | ||
<3 years | 0.75 (0.32 to 1.74) | .51 | .01 |
≥3 years | 0.35 (0.16 to 0.80) | .01 | |
Dose | |||
Non-metformin users | 1.00 (Reference) | ||
< median, 975 cDDD | 0.73 (0.35 to 1.53) | .40 | .02 |
≥ median, 975 cDDD | 0.33 (0.13 to 0.86) | .02 |
Duration and dose . | HR (95% CI) . | P* . | Ptrend† . |
---|---|---|---|
Duration | |||
Per every 1 yr increase in use‡ | 0.85 (0.74 to 0.96) | .01 | |
Metformin use§ | |||
Non-metformin users | 1.00 (Reference) | ||
<3 years | 0.75 (0.32 to 1.74) | .51 | .01 |
≥3 years | 0.35 (0.16 to 0.80) | .01 | |
Dose | |||
Non-metformin users | 1.00 (Reference) | ||
< median, 975 cDDD | 0.73 (0.35 to 1.53) | .40 | .02 |
≥ median, 975 cDDD | 0.33 (0.13 to 0.86) | .02 |
Cox proportional hazards model was used to calculate the P values, and the test was two-sided.
Cox proportional hazards model was used to calculate the P values, and the test was two-sided.
Continuous variable.
Categorical variable. 95% CI = 95% confidence interval; cDDD = cumulative Defined Daily Dose; HR = hazard ratio.
Discussion
In this territory-wide cohort study including more than 7200 diabetic patients who had H. pylori eradicated, we showed that metformin use was associated with a 51% reduction in GC risk, with a clear dose- and duration-gradient association. Because GC risk was found to be 20% higher among diabetic patients (6) and eradication of H. pylori could only reduce GC risk by about 40% (3,4), there is a genuine need to identify novel chemopreventive agents for this specific group of high-risk patients.
The beneficial effects of metformin on GC prevention remain controversial because randomized clinical trials were lacking and previous observational studies yielded conflicting results (10–16). Apart from the heterogeneity on definition of drug exposure, comparators, and study design, failure to adjust for other important risk factors of GC including H. pylori infection, HbA1c level, and concurrent medication use is likely the reason (17). To address these limitations, our study included only diabetic patients who were successfully treated for H. pylori at baseline. This would remove the most important risk factor of H. pylori infection that could affect the effect estimate of metformin on GC risk. By including the time-weighted average HbA1c level into analysis, our study not only considered the baseline HbA1c level but also the dynamics of DM control throughout the follow-up so that a more precise effect estimate could be derived. In our study, a higher time-weighted average HbA1c level of at least 7% was not an independent risk factor for GC after H. pylori eradication. In contrary, a Japanese population-based cohort study showed that a higher HbA1c level acted synergistically with H. pylori infection in increasing GC risk (18). Notably, their study cohort differed from our current cohort of H. pylori-eradicated subjects, and no adjustments were made to the use of concurrent medications that could modulate GC risk, such as aspirin/nonsteroidal anti-inflammatory drugs (32), statins (34), and PPIs (35).
We found that HbA1c levels were actually higher among metformin users, possibly related to the poor glycemic control necessitating metformin use. Yet, GC risk was statistically significantly lower in the metformin group. Hence, our findings support the biological mechanisms underlying metformin in preventing GC, which is not mediated through improving glycemic control. Notably, although IGFs have been linked to GC development (40), insulin treatment was not associated with an increased GC risk in this study. The duration and dose response association observed further strengthened the possible chemopreventive effect of metformin on GC. A longer duration of metformin use was associated with a lower GC risk, which was consistent with the study by Kim et al. showing the chemopreventive effect was only evident after at least 3 years of metformin use (13).
The strength of our study is the use of a comprehensive electronic public healthcare database that included drug prescription and dispensing records, reducing selection and information biases of observational studies (32). The long follow-up duration (a median of 7 years) also enables the assessment of GC with a long lag time. The robustness of the result is verified by the consistency from sensitivity analysis using various PS analysis methods, which adjusted for a wide array of covariates. Notably, metformin users had more cormobidities than non-metformin users (Table 1). Given that all these covariates are potential risk factors of GC, the protective effects of statin against GC could only be underestimated.
Several limitations of this study should be acknowledged. First, some risk factors like family history of GC and diet were not routinely captured in the CDARS. Second, because the success of H. pylori eradication was not documented in the ICD coding, we identified patients with failure of clarithromycin-based triple therapy by the repeated prescription or prescription of second- and third-line therapies. Third, drug compliance could not be ascertained in this electronic database. However, medications are prescribed and dispensed together in the same hospital at a very low price (US$1.5 per item for 16 weeks), which can rule out the issue of nondispensing because of cost issue. Further, noncompliance would only bias the beneficial effect of metformin towards null. Fourth, due to the long lag time and relatively low incidence of GC development after H. pylori eradication, there were only 37 GC cases. Stratified analyses according to the cancer location (noncardia and cardia) could not be performed. Lastly, information on GC staging was not available in the CDARS, precluding the exploration of potential effect of metformin on GC staging.
In conclusion, our territory-wide study showed that among diabetic patients who had H. pylori eradicated, metformin use was associated with a statistically significantly lower GC risk in a duration and dose response manner. This cancer protective effect was independent of glycemic control.
Funding
None.
Notes
Affiliations of authors: Department of Medicine, The University of Hong Kong, Queen Mary Hospital, Hong Kong (KSC, LC, WKS, WKL); Department of Pharmacology and Pharmacy, Centre for Safe Medication Practice and Research, The University of Hong Kong, Hong Kong (EWC, AYSW, ICKW); Faculty of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, London, UK (AYSW); UCL School of Pharmacy, University College London, London, UK (ICKW).
Guarantor of the article: Prof. Wai K Leung.
Specific author contributions: Dr Ka Shing Cheung and Wai Kay Seto were involved with study concept and design, analysis and interpretation of data, drafting of manuscript, and approval of the final version of the manuscript. Dr Esther W Chan, Dr Angel YS Wong, and Ms Lijia Chen were involved with acquisition of data, critical revision of the manuscript for important intellectual content, and approval of the final version of the manuscript. Professors Ian CK Wong and Wai K Leung were involved with the study concept and design, analysis and interpretation of data, drafting of manuscript, critical revision of the manuscript for important intellectual content, study supervision, and approval of the final version of the manuscript.
Potential competing interests: WKL has received honorarium for attending advisory board meetings of AbbVie, Takeda, and Abbott Laboratories.
References
World Health Organization. Cancer Fact Sheets: Stomach Cancer. http://gco.iarc.fr/today/fact-sheets-cancers? cancer=5&type=0&sex=0. Accessed June 3, 2017.
The Hospital Authority. Hospital Authority Statistical Report 2012–2013. http://www.ha.org.hk/haho/ho/stat/HASR1415_2.pdf. Accessed July 12, 2017.
WHO Collaborating Center for Drugs Statistics Methoodlogy. https://www.whocc.no/atc_ddd_index/. Accessed July 12, 2017.