A Critical Analysis of the Clinical Use of Incretin-Based Therapies : Are the GLP-1 therapies safe?

Exenatide, the first GLP-1–based therapy, was launched in the U.S. on April 29, 2005. A single case report of acute pancreatitis appeared in 2006 and was spotted by investment advisors who conducted their own search of the U.S. Food and Drug Administration (FDA) database and reported a potential risk of acute pancreatitis on October 2, 2006. The company made a change to its label on October 8, but the FDA did not issue its first alert until October 2007 (1). This was followed by a series of publications, mostly sponsored by the manufacturers, which reported that pancreatitis is more common in established diabetes than previously appreciated, together with pharmacoepidemiological studies using administrative databases that indicated that pancreatitis is no more common with exenatide than with other therapies for diabetes (2–4).

It is not easy to estimate the prevalence of acute pancreatitis, let alone assign a probable cause, and there are genuine difficulties in ascertaining the prevalence of acute pancreatitis in a population with diabetes. Reverse causation is an important confounder since both acute and chronic pancreatitis may give rise to diabetes. Chronic pancreatitis may present with acute episodes of pancreatic pain. The formal criteria for diagnosis—typical pain, enzyme rises, and changes on computed tomography (CT) examination—may not be satisfied or adequately recorded in administrative databases, and unequivocal CT abnormalities may not be present. The source documentation is often inadequate and pharmacoepidemiologic analyses may reach differing rate estimates because of differing criteria. Last but not least, a plausible mechanism to explain the occurrence of pancreatitis was initially lacking. This is no longer the case.

GLP-1 receptors are abundantly expressed in the pancreatic ducts as well as in the pancreatic islets, and the intense interest in GLP-1–based therapies as a potential stimulus to β-cell regeneration has overshadowed the possibility that exocrine pancreatic cells might be similarly affected. Acinar and duct cells proliferate in response to GLP-1 therapy (5) and cause an increase in pancreatic weight (6,7) (Fig. 1). Such observations attracted little attention prior to 2009 when one of eight HIP rats, a model of type 2 diabetes, developed hemorrhagic pancreatitis following exposure to sitagliptin, and some of the remaining animals showed marked acinar to ductal metaplasia, a potentially premalignant change characteristic of chronic pancreatitis (7). Gier et al. (8) noted that the pancreatic duct gland (PDG) compartment of the pancreas is particularly responsive to the proproliferative actions of GLP-1 and confirmed that GLP-1 simulates proliferative signaling in human pancreatic ductal epithelium. Two short-term studies were subsequently performed at the request of the FDA. These studies were carried out with exenatide and liraglutide in the ZDF rat model of diabetes and were reassuring with respect to possible adverse effects of GLP-1 mimetic therapy on the exocrine pancreas. Notwithstanding, pancreatic enzymes rose in both studies: one of twelve animals treated with exenatide died of massive pancreatic necrosis, and pathological findings in treated animals included acinar to ductal metaplasia and foci of ductal hyperplasia (9,10).

Some of the relevant preclinical studies are summarized in Table 1 (5–13). In aggregate, they offer a plausible mechanism for the occurrence of acute pancreatitis in patients exposed to GLP-1–based therapies since duct proliferation might lead to duct occlusion (particularly in the setting of existing dysplastic lesions), occlusion would generate back pressure, and back pressure would stress acinar cells thereby activating and releasing the digestive enzymes that they contain—a well-established causal mechanism for pancreatitis.

Animal studies do not necessarily reflect the experience in humans, but the identification of a plausible mechanism is an important step toward establishing a potential hazard and indicates a need for more detailed analysis in humans. Observational and pharmacoepidemiologic studies have suggested that acute pancreatitis is more common than expected in the diabetic population and is not increased by exenatide relative to other therapies (2–4). Although space does not permit detailed consideration here, there are some anomalies. For example, Dore et al. (2) examined the frequency of pancreatitis in a claims database comprising 25,700 patients on exenatide (past or present users) as compared with 234,500 patients on other antihyperglycemic therapies. Overall, there were more cases of confirmed pancreatitis in past or present exenatide users as compared with other therapies (40/25,719 vs. 254/234,536 = 1.56/1,000 vs. 1.08/1,000 users). The study found a reduced frequency of pancreatitis in present users of exenatide, but a propensity-adjusted RR (relative risk) of 2.8 (CI 1.6–4.7) for past use. The latter observation was discounted because those being studied were no longer taking exenatide at the time of the episode, but the exclusion would not be valid if exenatide had been stopped because of premonitory symptoms of abdominal pain or if the proposed mechanism persisted in those no longer taking the drug. Garg et al. (14) found no evidence of an increased risk of pancreatitis with exenatide, but concede that “the limitations of this observational claims-based analysis cannot exclude the possibility of an increased risk.” A recent case-control study addressed many of the limitations of previous reports, including inadequate power, and found that current and recent (1 month–2 years) users of GLP-1–based therapies had a twofold risk of acute pancreatitis (adjusted odds ratio 2.24 [95% CI 1.36–3.68] for current use and 2.01 [1.27–3.18] for recent use) (15).

Studies conducted by the manufacturer under the eyes of the regulators may provide reliable information. A recent review identified 11 such reports in studies conducted by Novo Nordisk, the manufacturer of liraglutide. Seven occurred in the LEAD (Liraglutide Effect and Action in Diabetes) studies (16), two in other studies, and two in postmarketing reports. Adverse events from the FDA Serious Adverse Event (SAE) reports were not considered. The findings were considered to “implicate liraglutide as the cause in at least some of these cases” (17).

Further cause for concern comes from FDA MedWatch data. An excess of acute pancreatitis was already evident for exenatide within 1 year of launch (1), and an updated analysis in 2011 found that, as compared with other non-GLP-1–based diabetes therapies, the reporting rate for acute pancreatitis with exenatide was dramatically increased (P < 2 × 10⁻⁴) (18). This easily checked analysis has not been seriously challenged.

The FDA alert system was designed to identify potential safety problems, not to confirm them. Notwithstanding its limitations, to our knowledge there is no single instance in which a strong sustained signal has turned out to be entirely spurious. When Elashoff et al. (18) was published, there were 971 reported pancreatitis events for exenatide and 131 for sitagliptin. The corresponding numbers are now 2,327 and 718 (Table 2). Recognition of an adverse event undoubtedly increases the reporting frequency, but there was a signal for exenatide long before the first FDA alert was issued, and there was no reason to anticipate a similar problem with sitagliptin. Furthermore, there are now 888 reported pancreatitis events for liraglutide, 125 for saxagliptin, and 43 for linagliptin (Table 2). Every GLP-1–based therapy with sufficient market exposure has generated a signal for pancreatitis, and no other diabetes medication has done so.

We conclude that the balance of evidence does suggest an association between widely used GLP-1–based therapies and acute pancreatitis, suggesting a class effect, and that this is underpinned by a plausible mechanism.

The major concern is not pancreatitis, unpleasant though this is. The concern is that acute events may be no more than the tip of an iceberg, and that these agents might cause subclinical duct proliferation, acinar to ductal metaplasia, and subclinical pancreatitis in a much higher proportion of individuals. Pancreatitis, whether clinical or subclinical, is well known to predispose to pancreatic cancer, and there is a signal for cancer of the pancreas for exenatide in both the FDA and German regulatory databases and for sitagliptin in the FDA database (18,19). The signal has grown stronger with 258 pancreatic cancers reported for exenatide, 63 for liraglutide, 81 for sitagliptin, 18 for saxagliptin, and 1 for linagliptin (Table 2).

Low-grade asymptomatic chronic pancreatitis with associated proliferative changes is not uncommon in the middle-aged target population for this drug class (20), making it likely that the proproliferative actions of GLP-1 therapy will at times be superimposed upon low-grade pancreatitis and its associated dysplastic changes. Some insight into this possibility was gained in the chronic pancreatitis–prone Kras^G12D mouse model in which exenatide therapy accelerated formation and growth of dysplastic intraepithelial neoplasia (PanIN) lesions as well as pancreatitis (8). To date, this is the only study of the actions of incretin treatment in a model of chronic pancreatitis (Fig. 1). In contrast, two studies of short-term GLP-1 exposure superimposed on acute toxin–induced pancreatitis were reported to show a protective effect, but such studies do not address the mechanism of relevance (6,11).

The incidence of pancreatic cancer, as of pancreatitis, is increased in type 2 diabetes (21). Work over the past decade has established that premalignant changes known as pancreatic intraepithelial (PanIN) lesions precede and predict the onset of pancreatic cancer. PanINs are present in up to 50% of the middle-aged population, although relatively few actually progress to cancer (20). Both PanINs and pancreatic cancer express the GLP-1 receptor in humans (8). Since progression of PanINs to pancreatic cancer is via the accumulation of additional somatic mutations, any driver of increased cellular replication in PanINs is likely to increase that probability. This theoretical risk was illustrated by the progression of PanINs in the exenatide-treated Kras^G12D mouse model (8). It is worth noting here that such a potential link between GLP-1 therapy and risk for pancreatic cancer is analogous to estrogen therapy and breast cancer. Estrogen does not initiate breast cancer, but in individuals with premalignant dysplastic ductal changes that bear estrogen receptors, estrogen accelerates growth and malignant conversion in some individuals (22). Likewise, the very high concentration of insulin delivered to the bronchial tree with inhaled insulin was associated with an increased incidence of lung cancer (23). Are estrogen, insulin, or GLP-1 carcinogens? No, but all three can serve as growth factors, and when pharmacological stimulation of growth is imposed on dysplastic lesions, accelerated declaration of cancer is not unexpected.

The regulatory reflex, when presented with a safety concern, is to request further descriptive data from the manufacturers. Our view is that the request for further epidemiologic analysis misses the real point of concern and wastes valuable time. The answer lies in the human pancreas, and (until this answer is known) there are more relevant questions to ask.

One question is this: If subclinical pancreatitis is common (consistent with the episodes of abdominal pain or discomfort described by many users), we might anticipate subclinical increases in pancreatic enzymes. Anecdotally, many clinicians already know this to be the case, but there is only one published case series (24). We accept that pancreatic enzyme levels fluctuate in people with diabetes and that confirmation of increased levels in people exposed to GLP-1–based therapies does not in itself constitute evidence of subclinical pancreatitis, but if a signal is there, we need to know.

The debate has been conducted in the absence of a single report from the pancreas of a human exposed to GLP-1 therapies. This is where the answer lies (25). Most recently, the first data have become available from human pancreas following a year or more of incretin therapy; 7 individuals treated by sitagliptin and 1 by exenatide compared to 12 individuals with type 2 diabetes treated with other agents and nondiabetes (26). The pancreas was 40% enlarged with increased exocrine pancreas proliferation in incretin-treated individuals. Moreover, there was an increase in the number of PanIN (premalignant) lesions after prior incretin treatment, consistent with the findings in the Kras^G12D mouse model (8). A striking finding in the human pancreas after incretin treatment was marked α-cell hyperplasia with glucagon-expressing microadenomas in 3 of the 8 individuals, and a glucagon-expressing neuroendocrine tumor in 1 of the 8. Given the heavily promoted action of incretin therapy to suppress glucagon secretion, and the prior reports of α-cell hyperplasia and risk for progression to pancreatic neuroendocrine tumors (26), this finding, while of concern, is perhaps not unexpected. No changes were reported in the exocrine pancreata of 10 monkeys that were treated with liraglutide for 87 weeks (13). Treatment was discontinued 2 weeks before the pancreata were obtained. The weight of the pancreata was not however reported (this would have been expected to increase). Long-term treatment of nondiabetic human primates with exenatide has not been published. The concerns raised in this article go well beyond the scope of routine histologic analysis conducted for regulatory purposes, and a full review by independent experts in pancreatic pathology would now seem justified (25).

In summary, a plausible mechanism links GLP-1–based therapy with acute pancreatitis—and a potential risk of pancreatic cancer—in individuals with type 2 diabetes. The model proposes acceleration of pancreatic dysplasia in the setting of low-grade chronic pancreatitis leading to sufficient ductal obstruction in a minority of individuals to provoke an episode of acute pancreatitis. Subclinical changes would be expected in a larger proportion of those exposed. The absence of pancreatitis or pancreatic dysplasia in nondiabetic models or short-term treatment of models of diabetes does not exclude the proposed mechanism. GLP-1 treatment, like estrogen in breast cancer, might promote development of pancreatic cancer in some individuals. Alternatively, periductal α-cell hyperplasia may cause duct obstruction and potentially progress to neuroendocrine neoplasia.

Preclinical registration studies of liraglutide found an increased number of C-cell tumors of the thyroid in rodents. Studies sponsored by the manufacturers have suggested that C cells in humans do not express the GLP-1 receptor; that humans exposed to liraglutide have, in aggregate, little or no rise in calcitonin levels; and that nonhuman primates exposed to liraglutide do not develop thyroid tumors (27). In contrast, analysis of a much larger sample of human thyroid glands and C cells established that a subpopulation of C cells in humans does indeed express the GLP-1 receptor (28) (Fig. 2). It was further established that GLP-1 receptor expression was more abundant in C-cell hyperplasia, a potential precursor of medullary thyroid cancer. Moreover, GLP-1 receptor expression is also present in 20% of those with papillary thyroid cancer, a much more common tumor for which calcitonin levels would be irrelevant. While medullary thyroid cancer is rare (29), a relatively high proportion of the population has apparently quiescent micro foci of papillary thyroid cancer (30).

Once again we must ask whether relatively short-term negative studies of GLP-1 mimetic therapy and thyroid cancer in normal monkeys provide adequate reassurance against the risk of malignancy in humans. As in the pancreas, the concern is that proproliferative actions of GLP-1 superimposed on premalignant lesions (C-cell hyperplasia or micropapillary thyroid cancer) may accelerate the progression of these lesions toward cancer. And, once again, adverse event reporting shows a clear excess of reported thyroid cancer on both exenatide (74 thyroid cancer events) and liraglutide (57 thyroid cancer events), although there is currently no similar signal for the DPP-4 inhibitors (Table 2).

The story is familiar. A new class of antidiabetic agents is rushed to market and widely promoted in the absence of any evidence of long-term beneficial outcomes. Evidence of harm accumulates, but is vigorously discounted. The regulators allow years to pass before they act. The manufacturers are expected—quite unrealistically—to monitor the safety of their own product. We should be thankful that those responsible for aircraft safety do not operate on the assumption that the absence of evidence is evidence of absence.

The safety of the GLP-1 therapies can no longer be assumed, and there will be rapid developments in this area. Drug safety can never be assumed, and the legal principle of “innocent until proved guilty” does not apply. The case presented here does not prove that these agents are unsafe, but it does suggest that the burden of proof now rests with those who wish to convince us of their safety.

P.C.B. is funded by the National Institute of Diabetes and Digestive and Kidney Diseases (DK059579, DK061539, DK077967), The Larry L. Hillblom Foundation, and the Peter and Valerie Kompaniez Foundation. The reanalysis of the FDA database was supported by the National Institutes of Health (NIH)/National Center for Research Resources/National Center for Advancing Translational Sciences UCLA Clinical and Translational Science Institute Grant UL1TR000124. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIH.

E.A.M.G. has provided expert testimony in litigation concerning exenatide. No other potential conflicts of interest relevant to this article were reported.

The authors gratefully acknowledge the technical support of Chi-Hong Tseng (Department of Biomathematics, UCLA) and the editorial assistance of Bonnie Lui (Larry L. Hillblom Islet Research Center, UCLA).