Use of an α-Glucosidase Inhibitor and the Risk of Colorectal Cancer in Patients With Diabetes: A Nationwide, Population-Based Cohort Study

Colorectal cancer is one of the most common cancers worldwide (1), causing an estimated 694,000 deaths in 2012 or 9.2% of all cancer-related deaths (2). Fecal occult blood screening and improved treatment in recent decades have reduced the annual global mortality of colorectal cancer (3). The incidence of colorectal cancer varies widely by geographic region and by degree of development. Although colorectal cancer is commonly seen in more developed regions, such as northern and western Europe and the U.S., the incidence trend is stable or even low in these regions (2). However, the standardized incidence of colorectal cancer has increased by 34% (33.8 in 2000 to 45.32 in 2010) and has been the third leading cause of cancer-related death in Taiwan for decades (4).

Several epidemiological studies have indicated that certain dietary factors, such as high-fat and low-fiber intake, play important roles in the development of colorectal cancer by promoting an increase in fecal bile acid (5,6). However, carbohydrate malabsorption may protect against colorectal cancer (7). Carbohydrate malabsorption induced by an α-glucosidase inhibitor changes the fecal bile acid pattern and decreases the levels of neutral sterols, suggesting a beneficial effect on the risk of colorectal cancer (8).

The large Cohort of Swedish Men study found an increase of ∼49% in the incidence of colon cancer in subjects with versus those without diabetes after controlling for potential risk factors (9). Various diabetes agents have different effects on cancer incidence. Patients with diabetes receiving metformin, thiazolidinediones (TZDs), nonsteroidal anti-inflammatory drugs (NSAIDs), or low-dose aspirin have a risk reduction in the colorectal cancer incidence (10–13), whereas insulin and sulfonylureas have been shown to promote cancer development (14,15). Acarbose is an antidiabetes agent indicated as adjuvant therapy with other diabetes agents to control hyperglycemia. Acarbose inhibits the α-glucosidase enzyme (16) and delays carbohydrate resorption in the small intestine, resulting in the delivery of more carbohydrates to the colon (17). A study using transformed cells suggested an antineoplastic effect of acarbose treatment (18). In the APC gene knockout animal model, which develops multiple intestinal adenomas, acarbose had a regressive effect on the size of gastrointestinal adenomas but did not significantly decrease the number of colonic neoplasms (19). Previous smaller database studies investigating acarbose use and incident colorectal cancer have provided null results (10,20). We used the Taiwanese National Health Insurance (NHI) research database of a large cohort with diabetes to evaluate the effect of acarbose treatment on the risk of colorectal cancer.

The Taiwanese NHI program provides compulsory universal health insurance, which was implemented on 1 March 1995 and covers all forms of health care services for 99% of the island’s population of 2.3 million. This population-based study of a cohort with diabetes used the NHI research database to obtain patient characteristics, including sex, date of birth, date of hospital admission, date of hospital discharge, outpatient visit dates, and up to five discharge or three outpatient visit diagnoses based on ICD-9 classification. The data files also contained information on patient prescriptions, including dose, duration, and total expenditure. The database was used previously for epidemiologic research; thus, prescription use, diagnosis, and hospitalization information is of high quality (21). Strict confidentiality guidelines were closely followed in accordance with personal electronic data protection regulations, and the National Health Research Institute of Taiwan anonymizes and maintains the reimbursement data as files suitable for research. The ethics review board of the National Taiwan University College of Public Health approved this study.

We identified patients with newly diagnosed diabetes (ICD-9 code 250 and A-code A181) between 1998 and 2010 after excluding coding for diabetes in the first year of the database (1997). Subject enrollment was restricted to patients with diabetes diagnosis codes resulting from at least one hospital admission or more than three outpatient visits within 1 year. This approach yielded a high sensitivity and positive predictive value after the definition of diabetes in the research database was validated (22). Subjects receiving a diabetes diagnosis in the last year of the database were excluded to ensure at least a 1-year follow-up period. Subjects with missing age or sex data or under age 18 years were excluded. We excluded patients who were admitted for any malignant neoplasm (ICD-9 codes 140–208) between 1997 and the date of the initial ambulatory care visit resulting in a diabetes diagnosis in 2010 to avoid misclassifying prevalent cancers as incident cancers.

We used the defined daily dose (DDD) to create coherence among the various types of acarbose treatment and to calculate the acarbose dose. The DDD measures a prescribed amount of drug and the assumed average maintenance dose per day. One DDD of acarbose is 0.3 g/user/day. The number of DDDs was calculated by the following equation: total amount of drug / amount of drug in a DDD. The cumulative DDD (cDDD) (i.e., exposure duration) was estimated to compare acarbose use with the risk of colorectal cancer. We categorized acarbose use into three groups for each cohort (>0 to <90, 90 to 364, and ≥365 cDDDs) and compared these groups with an acarbose nonuser group to examine the dose-effect relationship. We defined the index date as that of the first acarbose prescription. Because both the dose and the duration of acarbose vary between and within individuals over time, we further defined a time-dependent covariate of acarbose for each year indicating whether acarbose was dispensed to an individual during the year.

The acarbose user and acarbose nonuser cohorts were directly matched one to one (exposure vs. nonexposure) based on age (10-year increments); sex; age at diabetes diagnosis (by calendar year); and other comorbid diseases, including hypertension, hyperlipidemia, chronic kidney disease, colon polyps, and inflammatory bowel disease, during the same time period. We indicated the index date of first acarbose use by the case subject to correspond with the control subject and as a date representing the beginning of follow-up. In total, 398,592 subjects (or 199,296 pairs) were included in the analysis.

The outcome of interest was the diagnosis of colorectal cancer, which was defined as an ICD-9 code of 153.xx and 154.xx. We deliberately included only patients admitted for colorectal cancer to increase the validity of the diagnosis. We used at least a 1-year lag period after the index date subsequent to the initiation of acarbose to assess for incident colorectal cancer. Follow-up started on the index date and ended with the first of the following censoring events: diagnosis of incident colorectal cancer, death as recorded on admission files, or end of the study.

All potentially confounder-related colorectal cancer between 1 January 1997 and the date of colorectal cancer diagnosis or the end of follow-up were systematically identified. Confounders included treatment for diabetes, hypertension, and hyperlipidemia; aspirin and NSAID use; colon polyps; and inflammatory bowel disease. Demographic characteristics, including age, sex, income, and level of urbanization, were taken into consideration. Urbanization was divided into four categories within the Taiwan NHI research database, with level 1 referring to the most urbanized communities and level 4 referring to the least urbanized communities. The recommended interval for surveillance colonoscopy in patients with diabetes is 3–5 years (23). Surveillance colonoscopy in the current study was defined as undergoing a colonoscopy within 36 months before colorectal cancer diagnosis or the end of follow-up. To avoid influencing the diagnosis of colorectal cancer, colonoscopy performed within 6 months before the diagnosis was excluded (24). At least 3 years of care data in the current cohort were required to ensure a similar period for collecting colonoscopy data.

The Taiwan Health Promotion Administration offered free biennial fecal occult blood tests (FOBTs) to people >50 years of age. We set up the FOBT screening design as done for surveillance colonoscopy but used a 2-year time frame. The adapted Diabetes Complications Severity Index (aDCSI) was developed to evaluate the severity of diabetes, which included seven categories of complications with a total score of 0–13 (25,26). We summed all categories of diabetes complications before the index date to calculate the severity index as a baseline aDCSI for each subject. The categorical variable of the aDCSI score was 0 to ≥5.

We used a competing risks approach, which considered a patient population with a 9.2% mortality rate, because patients who died before developing colorectal cancer would remain in the risk set. We estimated the cumulative incidence of colorectal cancer among subjects with and without acarbose treatment by the Kaplan-Meier and competing risks method. Cox proportional hazards models with death as a competing risk event were used to calculate hazard ratios (HRs) with accompanying 95% CIs. A two-tailed P < 0.05 was considered significant. Subjects in whom colorectal cancer occurred within 1 year after acarbose treatment and those lost to follow-up or who died during the study period were censored. Potential confounders were analyzed separately by univariate analysis from which variables with meaningful clinical evidence or a change in the estimate by >10% were included in the Cox regression multivariate analyses. Multivariable analyses were used to evaluate linear trends in risk by treating acarbose use as a continuous variable after assigning a score to each exposure level. The P value was calculated for trends to confirm the dose-response relationship. We performed additional analyses to model acarbose use as a time-dependent variable in the Cox proportional hazards model. All analyses were conducted using SAS 9.4 statistical software (SAS Institute, Cary, NC).

Many agents have been studied for their potential chemopreventive effects in colorectal cancer. We examined the sensitivity of the results by stratifying (with or without) potential effect modifiers, including metformin, TZD, insulin, statin, aspirin, and NSAID use. These sensitivity analyses evaluated differences and consistencies between acarbose and the risk of colorectal cancer. The consistency of the effects of acarbose on colorectal cancer was analyzed in the subgroups based on age, sex, diabetes duration, and occurrence of surveillance colonoscopy.

In total, 1,343,484 persons with diabetes were included in the study cohort; 240,798 (17.9%) in this group had used acarbose. The median age at the time of the initial diabetes diagnosis was 54.1 years, and the median diabetes duration was 8.9 years. The mean treatment dose was 83.0 cDDD acarbose (Table 1). Patients treated with acarbose did not receive more frequent colonoscopies than those not treated with acarbose (5.64% vs. 5.76%, respectively, P = 0.093). FOBTs were more frequent in patients who used acarbose than in those who did not (12.37% vs. 10.07%, respectively, P < 0.001). Patients treated with acarbose exhibited a higher baseline aDCSI score than those not treated with acarbose (0.48 vs. 0.31, respectively, P < 0.001).

There were 1,332 (0.33%) incident colorectal cancer cases in the cohort with diabetes during the follow-up period of 1,487,136 person-years. Univariate analyses revealed that metformin, TZD, NSAID, and statin use and surveillance colonoscopy were negatively associated with the risk of incident colorectal cancer, whereas baseline aDCSI and insulin use were positively associated (Supplementary Table 1). The overall incidence of colorectal cancer was 89.6 cases per 100,000 person-years and that of acarbose users and nonusers was 75.8 and 103.8, respectively. We noted a 27% reduction in the risk of colorectal cancer in patients who had used acarbose compared with those who never did (crude HR 0.73 [95% CI 0.66–0.82], adjusted HR 0.66 [0.59–0.74]). A time-dependent Cox regression of acarbose exposure revealed a crude HR of 0.69 (0.67–0.71) and an adjusted HR of 0.72 (0.70–0.74) for acarbose users versus acarbose nonusers. The risk estimates in the time-dependent Cox regression model remained identical to those shown by the time-fixed model. The adjusted HRs were 0.73 (0.59–0.74), 0.69 (0.59–0.82), and 0.46 (0.37–0.58) for >0 to <90, 90 to 364, and ≥365 cDDD acarbose, respectively (Table 2). There was a significant trend toward risk reduction with increasing doses of acarbose (P for trend < 0.001). The log-rank test revealed a significant observed difference over the entire cumulative incidence curve (P < 0.001) (Fig. 1).

Sensitivity analysis adjustments had little effect on the estimates of the association between acarbose use and the incidence of colorectal cancer in the various models (Table 3). The beneficial effect of acarbose on colorectal cancer was amplified in this study after adjusting for insulin in all ranges of acarbose doses. In the subgroup analyses of age, sex, and diabetes duration, although the risk reductions for colorectal cancer were generally consistent with those of the overall analysis, statistical significance was not achieved at >0 to <90 cDDD acarbose compared with acarbose nonuse. In the subgroup analysis involving the presence of surveillance colonoscopy, the use of acarbose did not show a protective effect on colorectal cancer risk.

This study is the first to our knowledge to document a dose-response relationship between the use of acarbose and the risk of colorectal cancer in patients with diabetes after controlling for age, sex, onset of diabetes, and comorbid diseases and adjusting for potential confounders. A meta-analysis showed a relationship between diabetes and increased risk of colorectal cancer (9). The close link between diabetes and cancer may be due to insulin resistance or obesity (27). The link between diabetes agents and cancer has been explored extensively (15). However, inconsistent associations between acarbose and colorectal cancer exist. This study investigated the hypothesis that acarbose use is associated with reductions in colorectal cancer incidence in patients with diabetes.

Patients using acarbose had a 27% reduced risk of developing colorectal cancer compared with those who did not use acarbose. The dose-response effect of acarbose on the risk of colorectal cancer we observed concords with a previous clinical biomarker study demonstrating that each dose (50–200 mg) of acarbose enhances colonic butyrate production, which has colonic differentiating and nutritional effects (28). In the current study, the associations persisted after adjustment for the confounding factors. Additionally, the dose-response effect appeared linear and was independent of age, sex, and diabetes duration in the subgroup analysis.

An initial dose of acarbose 25 mg t.i.d. with a gradual titration to 50 mg t.i.d. is recommended. The average prescribed dose of acarbose in the NHI research database was 0.394 DDD, which is compatible with a large observational Asian study showing that 100–150 mg/day at the last visit yielded mild gastrointestinal disorders but was well tolerated (29). Gastrointestinal symptoms, such as flatulence or diarrhea, are the chief adverse effects of acarbose use, and the effects may interfere with adherence to treatment. However, these effects do not influence total energy intake or diet nutrition composition, and no body weight loss has been observed in subjects treated with acarbose (30). Acarbose treatment improved glycemic control and insulin sensitivity without a change in postprandial insulin increment or body weight loss during a 16-week treatment study (31). Acarbose is used as an adjuvant therapy with other diabetes agents, which implies more advanced diabetic conditions and yields a higher aDCSI score than in those who do not use acarbose. No evidence suggests that the severity of diabetes increases the risk of colorectal cancer, and a longer duration of diabetes is not associated with the risk of colorectal cancer (14). Patients who exhibit a higher baseline aDCSI score show an increased risk of colorectal cancer (P for trend < 0.001). The reduced risk of colorectal cancer persists in patients using acarbose when accounting for the severity of diabetes.

Compared with the current study, other observational studies using a smaller database showed that acarbose is not associated with a lower risk of colorectal cancer in patients with diabetes (10,20). In these previous investigations, the authors retrieved and longitudinally analyzed the data of all beneficiaries from 1996 to the end of 2005, using a population of 1 million in 2005 as the entire population presentation. Selection bias might have occurred because patients who died during the year of population sampling were less likely to appear in the survey. In Tseng (10), control subjects selected from the general population had highly heterogeneous baseline variables and a lower colorectal cancer incidence than did the population with diabetes. These factors led to the lack of a beneficial effect of acarbose on colorectal cancer risk reduction in patients with diabetes compared with the general population. In another study by Chiu et al. (20), subjects were matched by enrollment date using stratified random sampling; however, an immortal time bias might have occurred with an inappropriate accounting of the follow-up time in the study design.

Previous studies suggested that fecal butyrate, which is a short-chain fatty acid, is a key colonocyte nutrient and an important survival factor for colonic epithelial cells (32). Delayed carbohydrate absorption leads to more butyrate production by microbial fermentation after acarbose treatment. A partial explanation for the prevention of colorectal cancer by acarbose may be that butyrate inhibits the growth of transformed cells and stimulates normal colonic mucosal proliferation (18). A slower bowel transit time may also account for the increased incidence of colorectal cancer in patients with diabetes (33). Delaying stool transit may alter the production and concentration of bile acids and contribute to DNA damage (34). Acarbose reduces the colonic transit time, and changes in the fecal concentration of bile acids may explain the protective effect against colorectal cancer development (35). The risk of colorectal cancer increases with increasing endogenous hyperinsulinemia in patients with diabetes (36), whereas the use of acarbose reduces hyperinsulinemia (37).

Because the current study focused on the associations between acarbose use and risk of colorectal cancer, we did not perform an in-depth analysis of other medications. We did identify significant associations between other medications and colorectal cancer, which are mostly consistent with previous studies. A chemopreventive effect on colonic neoplasms by metformin, TZDs, low-dose aspirin, and NSAIDs has been suggested in epidemiologic trials (10–13); however, insulin therapy is associated with the development of colorectal cancer (14). Although a significant chemopreventive effect of statins against colon neoplasm was found in the current study, meta-analysis results failed to support this effect in patients on statin therapy (38).

Concerns were raised regarding differences in colonoscopy for patients with and without acarbose treatment because the patients treated with acarbose were more likely to have gastrointestinal symptoms, such as abdominal fullness or diarrhea. However, these adverse effects are not typical symptoms of colorectal cancer and do not promote colonoscopies in patients treated with acarbose. FOBTs were related to a significantly reduced incidence of colorectal cancer, which is based on the assumption of more colonoscopy examinations with polyp removal (39). Although patients treated with acarbose received more FOBTs than those not treated with acarbose in the current study, FOBTs did not contribute to more colonoscopic examinations or were related to an increased incidence of colorectal cancer. Colorectal cancer can be detected early or potentially avoided by removing adenomatous polyps through surveillance colonoscopy. The univariate analysis revealed that patients who received surveillance colonoscopy showed a 29% reduction in the risk of colorectal cancer compared with those who did not undergo this procedure. Surveillance colonoscopy might preclude the subsequent development of new adenomas and abolish the chemopreventive effect of acarbose on colorectal cancer.

This study has several strengths, including a nationwide, population-based design in a large and representative sample. We accessed a national research database and analyzed∼200,000 patients with diabetes for colorectal cancer who were treated with acarbose over a mean follow-up period of 3.4 years. Patients who were treated with acarbose were matched with those not treated with acarbose, and we used the baseline aDCSI to model diabetes severity and adjusted for potential examinations aimed at colorectal cancer detection to reduce potential bias and confounding effects (40). The competing risks approach was used to estimate the risk of colorectal cancer and, thereby, to avoid overestimations in the results.

The current study has a few limitations. We did not have data on other potential confounders, such as BMI, smoking, alcohol consumption, and physical activity, which are all associated with colorectal cancer. We also did not have laboratory data to determine the severity of diabetes; however, we identified the baseline aDCSI as a proxy for diabetes severity, which had been well validated with the clinical outcome (25). We did not have information on the colonoscopy indications or biopsy results, but surveillance colonoscopy 6–36 months before colorectal cancer diagnosis was adopted as a variable and entered into the main model for adjustment. We used pharmacy records to represent drug prescriptions rather than dispensing data, resulting in a conservative estimation of the association between acarbose use and colorectal cancer. We could not retrieve data on actual patient adherence to acarbose treatment, and an overestimation of the actual acarbose use was possible because we assumed that all prescribed medications were actually taken as prescribed. In summary, this study confirmed that acarbose use reduces the risk of colorectal cancer and that this effect is dose dependent.

Funding. This work was supported by the Taichung Veterans General Hospital, Taiwan (grant numbers TCVGH-1037203C, TCVGH-1030101C, and TCVGH-1030105D), and the National Science Council, Taiwan (grant number NSC 101-2314-B-075A-006-MY3).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. Y.-H.T. and Y.-T.T. contributed to the study concept, data analysis, interpretation of findings, and drafting and revision of the manuscript. W.-C.C. contributed to the study concept, statistical analysis, interpretation of findings, and review of the manuscript. W.H.-H.S. and P.-C.C. contributed to the study concept, supervision of all aspects of study implementation, interpretation of the findings, and review of the manuscript. P.-C.C. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.