Comment
Belmonte/Science Photo Library
Pancreatitis and incretin-based drugs: clarity or confusion?
Published Online November 12, 2013 http://dx.doi.org/10.1016/ S2213-8587(13)70186-4 See Articles page 111
92
An alarming report was published in 2011 by Elashoff and colleagues1 which claimed that patients with type 2 diabetes treated with exenatide (a glucagon-like peptide-1 [GLP-1] receptor agonist) or sitagliptin (a dipeptidyl peptidase 4 [DPP-4] inhibitor) had a 10-times increased risk for developing acute pancreatitis. Since then, attempts have been made to determine the true risk in these patients using data from insurance claims databases and analysis of information regarding drug use and diagnoses made after exposure for up to a few years. Most studies have a reported odds ratio for pancreatitis risk of about 1·0 (ie, no increase in risk) in patients treated with either exenatide, liraglutide, or sitagliptin compared with individuals treated with other glucose-lowering drugs.2 However, although the databases analysed contained between 4700 and 50 000 patients exposed to the drug of interest, the confidence intervals were wide: often ranging up to more than 1·5-times increased risk.2,3 Thus, the risk estimates published by Elashoff and colleagues in 2011,1 which were later extended to other GLP-1 receptor agonists (liraglutide) and DPP-4 inhibitors (saxagliptin and linagliptin),4 are widely accepted to be exaggerated because of various types of reporting bias,2 and they might not correctly estimate the potential effect size. An assessment of data from a US insurance claims database reported a 1·5-times increased risk for pancreatitis associated with incretin therapies (eg, exenatide or sitagliptin) based on raw numbers, which increased to two-times after adjustment for confounders such as gallstone disease and alcohol use.5 Another pooled analysis of two claims databases reported a 1·6-times increase in the risk for acute pancreatitis in previous users of exenatide.3 In The Lancet Diabetes & Endocrinology, Carlo Giorda and colleagues6 have undertaken the first populationbased analysis with information gathered from a general population database that comprises about 4·4 million individuals from Piedmont, Italy. More than 280 000 (6%) individuals in this population were older than 41 years, with type 2 diabetes, and had been prescribed glucoselowering drugs. Within this group, a total of 1003 cases of acute pancreatitis leading to hospital admission were identified in 2008–2012. This number is equivalent to an incidence of 0·7 cases per 1000 treatment-years, and is lower than the published pancreatitis incidence in
patients with type 2 diabetes.2,7 This discrepancy might be because only patients admitted to hospital were counted, or because of population-specific factors. The order of magnitude (1·4 million patient-years studied) and the availability of a detailed list of diagnoses (including common risk factors for acute pancreatitis needed for adjustments) distinguishes Giorda and colleagues’ report6 from preceding ones. Accordingly, at an odds ratio of 0·98, the upper bound of the confidence interval reported is 1·38, which is lower than that noted with most previous risk estimates, and indicates a more accurate estimate of the risk in question. In view of the conflicting reports, estimation of the true risk of pancreatitis associated with therapies that target incretin is difficult. Discrepancies in reporting might be largely caused by different methodological approaches. Retrospective analyses invariably face the problem of unavoidable biases. In case of the Elashoff and colleagues’ analyses,1 systematic over-reporting was likely to have occurred.2 Analyses from other databases are restricted by incomplete availability of concomitant risk factors and confounders. In Giorda and colleagues’ study,6 such underassessment of confounders was exemplified by the unusually low proportion of patients who misused alcohol (2% in cases and
Belmonte/Science Photo Library
Pancreatitis and incretin-based drugs: clarity or confusion?
Published Online November 12, 2013 http://dx.doi.org/10.1016/ S2213-8587(13)70186-4 See Articles page 111
92
An alarming report was published in 2011 by Elashoff and colleagues1 which claimed that patients with type 2 diabetes treated with exenatide (a glucagon-like peptide-1 [GLP-1] receptor agonist) or sitagliptin (a dipeptidyl peptidase 4 [DPP-4] inhibitor) had a 10-times increased risk for developing acute pancreatitis. Since then, attempts have been made to determine the true risk in these patients using data from insurance claims databases and analysis of information regarding drug use and diagnoses made after exposure for up to a few years. Most studies have a reported odds ratio for pancreatitis risk of about 1·0 (ie, no increase in risk) in patients treated with either exenatide, liraglutide, or sitagliptin compared with individuals treated with other glucose-lowering drugs.2 However, although the databases analysed contained between 4700 and 50 000 patients exposed to the drug of interest, the confidence intervals were wide: often ranging up to more than 1·5-times increased risk.2,3 Thus, the risk estimates published by Elashoff and colleagues in 2011,1 which were later extended to other GLP-1 receptor agonists (liraglutide) and DPP-4 inhibitors (saxagliptin and linagliptin),4 are widely accepted to be exaggerated because of various types of reporting bias,2 and they might not correctly estimate the potential effect size. An assessment of data from a US insurance claims database reported a 1·5-times increased risk for pancreatitis associated with incretin therapies (eg, exenatide or sitagliptin) based on raw numbers, which increased to two-times after adjustment for confounders such as gallstone disease and alcohol use.5 Another pooled analysis of two claims databases reported a 1·6-times increase in the risk for acute pancreatitis in previous users of exenatide.3 In The Lancet Diabetes & Endocrinology, Carlo Giorda and colleagues6 have undertaken the first populationbased analysis with information gathered from a general population database that comprises about 4·4 million individuals from Piedmont, Italy. More than 280 000 (6%) individuals in this population were older than 41 years, with type 2 diabetes, and had been prescribed glucoselowering drugs. Within this group, a total of 1003 cases of acute pancreatitis leading to hospital admission were identified in 2008–2012. This number is equivalent to an incidence of 0·7 cases per 1000 treatment-years, and is lower than the published pancreatitis incidence in
patients with type 2 diabetes.2,7 This discrepancy might be because only patients admitted to hospital were counted, or because of population-specific factors. The order of magnitude (1·4 million patient-years studied) and the availability of a detailed list of diagnoses (including common risk factors for acute pancreatitis needed for adjustments) distinguishes Giorda and colleagues’ report6 from preceding ones. Accordingly, at an odds ratio of 0·98, the upper bound of the confidence interval reported is 1·38, which is lower than that noted with most previous risk estimates, and indicates a more accurate estimate of the risk in question. In view of the conflicting reports, estimation of the true risk of pancreatitis associated with therapies that target incretin is difficult. Discrepancies in reporting might be largely caused by different methodological approaches. Retrospective analyses invariably face the problem of unavoidable biases. In case of the Elashoff and colleagues’ analyses,1 systematic over-reporting was likely to have occurred.2 Analyses from other databases are restricted by incomplete availability of concomitant risk factors and confounders. In Giorda and colleagues’ study,6 such underassessment of confounders was exemplified by the unusually low proportion of patients who misused alcohol (2% in cases and
