pharmacoepidemiology and drug safety (2015) Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/pds.3770
ORIGINAL REPORT
Anti-diabetic therapies and the risk of acute pancreatitis: a nationwide retrospective cohort study from Taiwan Hsien-Yen Chang1,2, Chi-Feng Hsieh3, Sonal Singh2,4,5, Wenze Tang6, Yi-Ting Chiang3 and Weng-Foung Huang3* 1
Department of Health Policy & Management, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA Center for Drug Safety and Effectiveness, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA 3 Institute of Health and Welfare Policy, National Yang-Ming University, Taipei, Taiwan 4 Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA 5 Division of General Internal Medicine, Johns Hopkins Medicine, Baltimore, MD, USA 6 Department of Epidemiology and Biostatistics, Milken Institute School of Public Health, The George Washington University, Washington, DC, USA 2
ABSTRACT Aims To examine the relationship between different anti-diabetic therapies (dipeptidyl peptidase-4 (DPP-4), metformin and sulfonylureas) and risk of acute pancreatitis among type 2 diabetic patients in Taiwan, and explore each drug’s dose-response relationship. Materials and Methods We derived a nationwide retrospective cohort of patients with type 2 diabetes in Taiwan. The inclusion criteria are adult diabetic patients with continuous baseline enrollment, new users of the studied drugs, and without missing demographics. There were 4113/101 498/44 772 DPP-4/Metformin/Sulfonylurea users. Adjusted hazards ratios for pancreatitis associated with DPP-4, derived from Cox proportional hazard models with propensity score weighting, were estimated; dose-response analyses were also conducted. Results Dipeptidyl peptidase-4 was statistically significantly associated with a decreased risk of acute pancreatitis compared with sulfonylureas (adjusted HR: 0.36, 95%CI [0.17, 0.75]) but not metformin (adjusted HR: 0.67, 95%CI [0.32, 1.41]); metformin was statistically significantly associated with a lower risk of pancreatitis than sulfonylurea (adjusted HR: 0. 53; 95%CI [0.37, 0.76]). In addition, low-dose metformin was statistically significantly associated with a lower risk of pancreatitis compared with high-dose metformin (HR: 0.65; 95% CI [0.44, 0.97]). Conclusions Our findings suggest that sulfonylureas may potentially be associated with an increased risk of pancreatitis compared with DPP-4 or metformin. Studies with longer follow up, larger sample sizes, and more precise capture of confounders may be needed to determine the risk of pancreatitis associated with incretin based therapies. Copyright © 2015 John Wiley & Sons, Ltd. key
words—anti-diabetic therapies; pharmacoepidemiology
dipeptidyl
peptidase-4,
metformin;
sulfonylurea;
acute
pancreatitis;
cohort
study;
Received 18 December 2014; Revised 17 February 2015; Accepted 19 February 2015
INTRODUCTION Incretin-based therapies, including glucagon-like peptide-1 (GLP-1) mimetic and dipeptidyl peptidase4 inhibitor (DPP-4), are new classes of anti-diabetic drugs. While GLP1-1 agonists improve the secretion of glucose-mediated insulin, DPP-4 stalls the proteolysis of endogenous GLP-1.1 Some earlier rodent models *Correspondence to: W.-F. Huang, Institute of Health and Welfare Policy, National Yang-Ming University, Rm 618, No. 155, Sec.2, Linong Street, Taipei 11221, Taiwan. Email: [email protected]
Copyright © 2015 John Wiley & Sons, Ltd.
have shown acute pancreatitis increased among mice exposed to exenatide and sitagliptin.2,3 On the contrary, recent systematic review and meta-analysis examining incretin treatment concluded no compelling evidence of increased risk of pancreatitis with incretin.4–7 One recent meta-analysis, even with 33 350 subjects and follow-up time from 12 to 234 weeks, is still underpowered as the incidence rate of pancreatitis among diabetic patients could be as low as 4.2/1000 person-years.8 Meanwhile, crude analyses using the adverse event reporting and surveillance system across countries continued to report higher
h.-y. chang et al.
odds of pancreatic events in patients exposed to incretin-based therapies.9–11 A few observational studies using claims databases tried to control for known confounders, such as diabetic severity and alcohol use.12,13 However, they suffered from the lack of statistical power, outcome validation, or insufficient ascertainment of confounders.4 In addition, most observational studies utilized commercial datasets within the USA, where applying age restriction and loss of follow-up due to enrollment were frequent; some retrospective cohort studies did not require continuous exposure, which might result in misclassification.14,15 Furthermore, most studies do not have clearly defined reference groups (only classified as users versus non-users consisting of different types of other drug users), and no study has looked at dose-response relationship between incretin-based therapies and risk of pancreatitis. We aimed to examine the relationship between antidiabetic therapies (DPP-4, metformin, and sulfonylureas) and risk of acute pancreatitis among type 2 diabetic patients in Taiwan, and explore the doseresponse relationship within three anti-hyperglycemic medication user groups. MATERIALS AND METHODS Design This is a retrospective cohort study. We applied a Cox proportional hazard model with propensity score weighting to evaluate the impact of using DPP-4 (relative to metformin or sulfonylurea) on pancreatitis among Type-II diabetic patients. Data source We requested the National Health Research Institutes, in charge of National Health Insurance Research Database, to provide a 6-year (01 January 2006 to 31 December 2011) longitudinal dataset that included enrollment and claims files associated with a patient cohort of diabetes mellitus. Taiwan’s National Health Insurance (NHI) provides comprehensive medical and pharmaceutical coverage; contracted healthcare providers are required to submit claims electronically to the National Health Insurance Administration for reimbursement. Taiwan’s NHI claims data has been adopted and described in detail in various medical research, including health services16–19 and pharmaceutical research.13,20–22 The enrollment files contained individual subscription information and demographic factors, including sex and date of birth. The claims files contained comprehensive records of inpatient, ambulatory, Copyright © 2015 John Wiley & Sons, Ltd.
pharmacy, dental care and Chinese medicine services, including date of service, ICD-9-CM (International Classification of Diseases) diagnosis and procedure codes, prescription, and days of prescription supply. Individuals’ and providers’ identifiers have been encrypted to protect privacy and confidentiality.20,22 Study population and cohort entry/exit date The diabetic patient cohort was identified from the entire 23 million enrollees under Taiwan’s NHI. Enrollees with any prescription of diabetic medications (Appendix 1 in the Supporting information) or any diabetic diagnosis (ICD-9-CM code: 250.xx) from year 2006 to 2011 were included; it contained 1 447 203 diabetic patients. Our study population included Type-II diabetic patients with the first prescription of three oral antihyperglycemic agents (DPP-4, metformin, or sulfonylurea) that occurred as follows: (1) between 1 Mar 2009 and 31 Dec 2011; and (2) after the first diagnosis of diabetes (n: 307 001). GLP-1 agonist was not included given that only 92 prescriptions have been prescribed during our study period from the diabetic cohort. DPP-4 was first reimbursed by NHI on 1 Mar 2009, and included sitagliptin, vildagliptin, and saxagliptin (Appendix 1). An individual’s first prescription of these medications after 1 Mar 2009 would be the drug exposure for that individual, and this prescription date would be that individual’s entry date into the study cohort. To be eligible for this study, an individual has to be 18 years old or older, have continuous enrollment in one year prior to the entry date (the baseline period), have none of the oral antihyperglycemic agents prescribed in the baseline period (a new-user design), have non-missing days of supply, and have anti-diabetic drugs other than prescribed over the study period. After applying these criteria, the final sample size came down to 150 383 (Figure 1). An individual’s cohort exit date would be the earliest of the following dates. The first is the last date of the same drug (continuous) exposure. The last exposure date is the last day of continuous drug at hand plus 7 days.23 We chose 7 days based on the half life of these drugs to take into account the wash-away period. For example, if there was a 7-day or less of a gap between two consecutive prescriptions of the same drug, we considered the exposure to be continuous. However, if an individual switched within the 7 days of the previous continuous exposure to another drug, the day prior to starting a new drug would be the last exposure date. The second is the date of the loss of enrollment. The third is the ending date of the study, which is 31 Dec Pharmacoepidemiology and Drug Safety, 2015 DOI: 10.1002/pds
ANTI-DIABETIC DRUGS AND PANCREATITIS
Figure 1.
Flow chart of the sample size
2011. The last one is the first date of pancreatitis. An individual was administratively censored if the cohort exit date was not equal to the first pancreatitis date. Outcome and independent variables The main independent variable of interest is the type of drug exposure: DPP-4, metformin, or sulfonylurea. Control variables include demographics (four age groups and sex), binary clinical conditions (whether an individual had hypertriglyceridemia, alcohol use, gallstone, tobacco abuse, obesity, biliary cancer, pancreatic cancer, and neoplasm), three levels of adapted Diabetes Complications Severity Index (aDCSI) scores and five levels of clinical classification software (CCS) counts. The aDCSI is a valid measure of diabetes severity using claims data.24,25 Developed at the Agency for Healthcare Research and Quality, the CCS is a tool for clustering patient diagnoses into a manageable Copyright © 2015 John Wiley & Sons, Ltd.
number of clinically meaningful categories.26 All these variables were derived from the baseline claims data. To estimate the dose-response effect, we calculated the defined daily dose (DDD)27 for each individual over the entire observation period (from the cohort entry to exit date). Individuals were then classified into three roughly equally sized groups based on their DDD’s values within users of the same drug: low, medium, or high. The outcome of interest is acute pancreatitis, defined in a prior research using a validated algorithm.28,29 Appendix 1 in the supporting information described the detailed definition of each variable. Propensity score To control for the differences across drug users, we developed a propensity score to predict whether a person used DPP-4, relative to metformin or sulfonylurea. The variables included as independent variables are control Pharmacoepidemiology and Drug Safety, 2015 DOI: 10.1002/pds
h.-y. chang et al.
variables listed earlier, with the only difference in that the CCS is entered into the propensity score model as binary condition indicators. We used propensity score weighing because we did not want to lose the observations of treated subjects and we wanted an interpretable overall treatment effect; in addition, weighting would allow us to estimate both the average treatment effect of the population (ATE weighting) and that of the treated (ATT weighting). After applying propensity score weighting, the differences in covariates between treatment groups were mostly negligible (standardized difference
ORIGINAL REPORT
Anti-diabetic therapies and the risk of acute pancreatitis: a nationwide retrospective cohort study from Taiwan Hsien-Yen Chang1,2, Chi-Feng Hsieh3, Sonal Singh2,4,5, Wenze Tang6, Yi-Ting Chiang3 and Weng-Foung Huang3* 1
Department of Health Policy & Management, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA Center for Drug Safety and Effectiveness, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA 3 Institute of Health and Welfare Policy, National Yang-Ming University, Taipei, Taiwan 4 Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA 5 Division of General Internal Medicine, Johns Hopkins Medicine, Baltimore, MD, USA 6 Department of Epidemiology and Biostatistics, Milken Institute School of Public Health, The George Washington University, Washington, DC, USA 2
ABSTRACT Aims To examine the relationship between different anti-diabetic therapies (dipeptidyl peptidase-4 (DPP-4), metformin and sulfonylureas) and risk of acute pancreatitis among type 2 diabetic patients in Taiwan, and explore each drug’s dose-response relationship. Materials and Methods We derived a nationwide retrospective cohort of patients with type 2 diabetes in Taiwan. The inclusion criteria are adult diabetic patients with continuous baseline enrollment, new users of the studied drugs, and without missing demographics. There were 4113/101 498/44 772 DPP-4/Metformin/Sulfonylurea users. Adjusted hazards ratios for pancreatitis associated with DPP-4, derived from Cox proportional hazard models with propensity score weighting, were estimated; dose-response analyses were also conducted. Results Dipeptidyl peptidase-4 was statistically significantly associated with a decreased risk of acute pancreatitis compared with sulfonylureas (adjusted HR: 0.36, 95%CI [0.17, 0.75]) but not metformin (adjusted HR: 0.67, 95%CI [0.32, 1.41]); metformin was statistically significantly associated with a lower risk of pancreatitis than sulfonylurea (adjusted HR: 0. 53; 95%CI [0.37, 0.76]). In addition, low-dose metformin was statistically significantly associated with a lower risk of pancreatitis compared with high-dose metformin (HR: 0.65; 95% CI [0.44, 0.97]). Conclusions Our findings suggest that sulfonylureas may potentially be associated with an increased risk of pancreatitis compared with DPP-4 or metformin. Studies with longer follow up, larger sample sizes, and more precise capture of confounders may be needed to determine the risk of pancreatitis associated with incretin based therapies. Copyright © 2015 John Wiley & Sons, Ltd. key
words—anti-diabetic therapies; pharmacoepidemiology
dipeptidyl
peptidase-4,
metformin;
sulfonylurea;
acute
pancreatitis;
cohort
study;
Received 18 December 2014; Revised 17 February 2015; Accepted 19 February 2015
INTRODUCTION Incretin-based therapies, including glucagon-like peptide-1 (GLP-1) mimetic and dipeptidyl peptidase4 inhibitor (DPP-4), are new classes of anti-diabetic drugs. While GLP1-1 agonists improve the secretion of glucose-mediated insulin, DPP-4 stalls the proteolysis of endogenous GLP-1.1 Some earlier rodent models *Correspondence to: W.-F. Huang, Institute of Health and Welfare Policy, National Yang-Ming University, Rm 618, No. 155, Sec.2, Linong Street, Taipei 11221, Taiwan. Email: [email protected]
Copyright © 2015 John Wiley & Sons, Ltd.
have shown acute pancreatitis increased among mice exposed to exenatide and sitagliptin.2,3 On the contrary, recent systematic review and meta-analysis examining incretin treatment concluded no compelling evidence of increased risk of pancreatitis with incretin.4–7 One recent meta-analysis, even with 33 350 subjects and follow-up time from 12 to 234 weeks, is still underpowered as the incidence rate of pancreatitis among diabetic patients could be as low as 4.2/1000 person-years.8 Meanwhile, crude analyses using the adverse event reporting and surveillance system across countries continued to report higher
h.-y. chang et al.
odds of pancreatic events in patients exposed to incretin-based therapies.9–11 A few observational studies using claims databases tried to control for known confounders, such as diabetic severity and alcohol use.12,13 However, they suffered from the lack of statistical power, outcome validation, or insufficient ascertainment of confounders.4 In addition, most observational studies utilized commercial datasets within the USA, where applying age restriction and loss of follow-up due to enrollment were frequent; some retrospective cohort studies did not require continuous exposure, which might result in misclassification.14,15 Furthermore, most studies do not have clearly defined reference groups (only classified as users versus non-users consisting of different types of other drug users), and no study has looked at dose-response relationship between incretin-based therapies and risk of pancreatitis. We aimed to examine the relationship between antidiabetic therapies (DPP-4, metformin, and sulfonylureas) and risk of acute pancreatitis among type 2 diabetic patients in Taiwan, and explore the doseresponse relationship within three anti-hyperglycemic medication user groups. MATERIALS AND METHODS Design This is a retrospective cohort study. We applied a Cox proportional hazard model with propensity score weighting to evaluate the impact of using DPP-4 (relative to metformin or sulfonylurea) on pancreatitis among Type-II diabetic patients. Data source We requested the National Health Research Institutes, in charge of National Health Insurance Research Database, to provide a 6-year (01 January 2006 to 31 December 2011) longitudinal dataset that included enrollment and claims files associated with a patient cohort of diabetes mellitus. Taiwan’s National Health Insurance (NHI) provides comprehensive medical and pharmaceutical coverage; contracted healthcare providers are required to submit claims electronically to the National Health Insurance Administration for reimbursement. Taiwan’s NHI claims data has been adopted and described in detail in various medical research, including health services16–19 and pharmaceutical research.13,20–22 The enrollment files contained individual subscription information and demographic factors, including sex and date of birth. The claims files contained comprehensive records of inpatient, ambulatory, Copyright © 2015 John Wiley & Sons, Ltd.
pharmacy, dental care and Chinese medicine services, including date of service, ICD-9-CM (International Classification of Diseases) diagnosis and procedure codes, prescription, and days of prescription supply. Individuals’ and providers’ identifiers have been encrypted to protect privacy and confidentiality.20,22 Study population and cohort entry/exit date The diabetic patient cohort was identified from the entire 23 million enrollees under Taiwan’s NHI. Enrollees with any prescription of diabetic medications (Appendix 1 in the Supporting information) or any diabetic diagnosis (ICD-9-CM code: 250.xx) from year 2006 to 2011 were included; it contained 1 447 203 diabetic patients. Our study population included Type-II diabetic patients with the first prescription of three oral antihyperglycemic agents (DPP-4, metformin, or sulfonylurea) that occurred as follows: (1) between 1 Mar 2009 and 31 Dec 2011; and (2) after the first diagnosis of diabetes (n: 307 001). GLP-1 agonist was not included given that only 92 prescriptions have been prescribed during our study period from the diabetic cohort. DPP-4 was first reimbursed by NHI on 1 Mar 2009, and included sitagliptin, vildagliptin, and saxagliptin (Appendix 1). An individual’s first prescription of these medications after 1 Mar 2009 would be the drug exposure for that individual, and this prescription date would be that individual’s entry date into the study cohort. To be eligible for this study, an individual has to be 18 years old or older, have continuous enrollment in one year prior to the entry date (the baseline period), have none of the oral antihyperglycemic agents prescribed in the baseline period (a new-user design), have non-missing days of supply, and have anti-diabetic drugs other than prescribed over the study period. After applying these criteria, the final sample size came down to 150 383 (Figure 1). An individual’s cohort exit date would be the earliest of the following dates. The first is the last date of the same drug (continuous) exposure. The last exposure date is the last day of continuous drug at hand plus 7 days.23 We chose 7 days based on the half life of these drugs to take into account the wash-away period. For example, if there was a 7-day or less of a gap between two consecutive prescriptions of the same drug, we considered the exposure to be continuous. However, if an individual switched within the 7 days of the previous continuous exposure to another drug, the day prior to starting a new drug would be the last exposure date. The second is the date of the loss of enrollment. The third is the ending date of the study, which is 31 Dec Pharmacoepidemiology and Drug Safety, 2015 DOI: 10.1002/pds
ANTI-DIABETIC DRUGS AND PANCREATITIS
Figure 1.
Flow chart of the sample size
2011. The last one is the first date of pancreatitis. An individual was administratively censored if the cohort exit date was not equal to the first pancreatitis date. Outcome and independent variables The main independent variable of interest is the type of drug exposure: DPP-4, metformin, or sulfonylurea. Control variables include demographics (four age groups and sex), binary clinical conditions (whether an individual had hypertriglyceridemia, alcohol use, gallstone, tobacco abuse, obesity, biliary cancer, pancreatic cancer, and neoplasm), three levels of adapted Diabetes Complications Severity Index (aDCSI) scores and five levels of clinical classification software (CCS) counts. The aDCSI is a valid measure of diabetes severity using claims data.24,25 Developed at the Agency for Healthcare Research and Quality, the CCS is a tool for clustering patient diagnoses into a manageable Copyright © 2015 John Wiley & Sons, Ltd.
number of clinically meaningful categories.26 All these variables were derived from the baseline claims data. To estimate the dose-response effect, we calculated the defined daily dose (DDD)27 for each individual over the entire observation period (from the cohort entry to exit date). Individuals were then classified into three roughly equally sized groups based on their DDD’s values within users of the same drug: low, medium, or high. The outcome of interest is acute pancreatitis, defined in a prior research using a validated algorithm.28,29 Appendix 1 in the supporting information described the detailed definition of each variable. Propensity score To control for the differences across drug users, we developed a propensity score to predict whether a person used DPP-4, relative to metformin or sulfonylurea. The variables included as independent variables are control Pharmacoepidemiology and Drug Safety, 2015 DOI: 10.1002/pds
h.-y. chang et al.
variables listed earlier, with the only difference in that the CCS is entered into the propensity score model as binary condition indicators. We used propensity score weighing because we did not want to lose the observations of treated subjects and we wanted an interpretable overall treatment effect; in addition, weighting would allow us to estimate both the average treatment effect of the population (ATE weighting) and that of the treated (ATT weighting). After applying propensity score weighting, the differences in covariates between treatment groups were mostly negligible (standardized difference
