Safety of Semaglutide

Given that the aim of GLP-1RA therapy is mainly to reduce blood glucose levels, it is conceivable that these agents could cause hypoglycemia. However, since GLP-1RA mainly lower blood glucose by stimulating glucose-dependent insulin secretion, hypoglycemia is an infrequent problem. In addition, the inhibition of glucagon release does not occur under hypoglycemic conditions (35). In SUSTAIN-6, severe or plasma glucose-confirmed (<56 mg/dl [3.1 mol/L]) hypoglycemia occurred in similar rates between patients with semaglutide (23.1% in the 0.5 mg group and 21.7% in the 1 mg group) and placebo (21.2%) (28). In comparison, in SUSTAIN-4, severe or confirmed hypoglycemia occurred in 11% of insulin glargine-treated patients, compared with 4–6% in the semaglutide-treated patients (26). Importantly, in SUSTAIN-4 it is reported that hypoglycemia predominantly occurred in subjects using sulfonylurea agents (26). To illustrate: in the group of subjects randomized to semaglutide 1 mg, 9% of subjects using a s sulfonylurea had a severe or blood-glucose confirmed hypoglycemia, versus 2% in those not using a sulfonylurea. Similarly, in SUSTAIN-3, the majority of hypoglycemic events were reported in subjects concomitantly receiving sulfonylureas in both the semaglutide 1.0 mg and exenatide ER 2.0 mg groups. For oral semaglutide, the percentage of patients with severe hypoglycemia was 1.4% with oral semaglutide and 0.8% with placebo in PIONEER 6 (19). Here all severe hypoglycemic events occurred in patients receiving concomitant insulin or sulfonylurea therapy at the time of the event. In other phase 3 trials, no increase in hypoglycemia risk was observed versus comparator groups, including other GLP-1RAs (18–21, 23, 33–36).

Real world data with respect to hypoglycemia are limited to a single observational cohort from Canada (36). In 815 individuals who started semaglutide therapy and were followed for 6 months, there was no change in overall reported hypoglycemia. Although the group of concomitant insulin users also reported no change in hypoglycemia occurrence, this could have been mitigated by the on average 10–20% reduction in total daily insulin dosage (36). Sulfonylurea users did not experience an increase in hypoglycemia events.

Thus, the risk of hypoglycemia appears to be low with subcutaneous and oral semaglutide by themselves, yet the risk is increased when combined with sulfonylurea and/or insulin therapy. Several experts advise to lower the dose of sulfonylurea and short-acting and low-acting insulin analogues prior to or during titration of GLP-1RA therapy, to reduce the risk of (severe) hypoglycemia (37).

In the phase 3 trials, both oral and subcutaneous semaglutide were associated with gastrointestinal disturbances, such as nausea, vomiting and diarrhea, well-known effects from this drug class. When compared with placebo, subcutaneous semaglutide for 30 weeks induced nausea in 11.4 to 20% of the semaglutide-treated patients (placebo 3.3–8%), vomiting in 4 to 11.5% (placebo 2–3%) and diarrhea in 4.5 to 11.3% (placebo 1.5–6%) (10, 27, 31). In SUSTAIN 6, where generally older patients with comorbid conditions were treated for 104 weeks, the incidence of GI disturbances was somewhat higher (28). For oral semaglutide, the placebo-controlled trials found nausea ranged between 5.1 and 23.2% (placebo 5.6–7.1%), vomiting between 2.9 and 9.9% (placebo 2.2–3.8%) and diarrhea between 5.1 and 15% (placebo 2.2–8%) during the on-treatment period (14, 17, 21). These rates were not different when focusing on Japanese patients [PIONEER 9 (23)], but appeared higher in patients with T2D, reduced kidney function (estimated glomerular filtration rate [GFR] of 30–59 ml/min) and comorbidities in PIONEER 5 (18).

In one phase 2 trial, subcutaneous and oral semaglutide were compared with each other (38). Here, patients were randomized to oral semaglutide (at a dose of 5, 10, 20 or 40 mg once daily), subcutaneous semaglutide (1 mg once weekly) or placebo. As discussed below, this study also assessed the effect of dose escalation in two additional groups. Unfortunately, the currently advocated oral treatment doses of 7 and 14 mg were not included. When comparing oral 20 mg to subcutaneous 1 mg, the total amount of gastrointestinal disturbances was similar (56% versus 54%, respectively). This was also true for nausea (34% versus 32%), vomiting (16% versus 9%) and diarrhea (20% versus 14%). The proportion of patients with premature discontinuation because of adverse events appeared higher for oral semaglutide 20 mg (27%) than for subcutaneous semaglutide 1 mg (14%). All numbers were similar between the 10 and 20 mg oral dose, except for treatment discontinuation, which was 12% for the lower dosage.

Importantly, for both formulations, higher doses are often associated with more frequent GI adverse effects. For this reason, a dose escalation scheme is advised, starting with a low dose (3 mg). As a clear example in the abovementioned phase 2 study (38), 77% of patients experienced GI adverse effects when a fast 2-week dose escalation was used to reach 40 mg compared with 54% in the slower 8-week dose-escalation group. Generally, the GI complaints with semaglutide occur in the first 8–12 weeks of treatment during dose escalation [in contrast to for example liraglutide, where they occur within 2 weeks (17, 32)], and wane over time (Figure 2). Overall, the adverse effects are mild to moderate in severity and often self-limiting.

Nevertheless, GI complaints are the main adverse-event related cause of drug discontinuation in the phase-3 trials, with rates up to 12% (Table 2). Moreover, cohorts with real-world data show similar numbers. In one retrospective study where 189 patients with T2DM starting subcutaneous semaglutide, 9.5% discontinued therapy because of GI complaints, while in 5.8% such adverse effects limited uptitration (39). In another cohort where 164 T2DM patients were switched from a different GLP-1RA therapy to semaglutide, 10.4% discontinued semaglutide because of adverse GI effects (40). Combined, data from clinical trials and clinical practice suggest that approximately 10% of patients will discontinue semaglutide because of GI complaints, which may be a bit higher compared to other GLP-1 analogues.

Apart from gradual dose titration, data on how to prevent or treat GI disturbances with GLP-1RA are limited. Patients can be counseled to eat slowly with reduced portion size per meal, stop eating when they experience satiety, and to avoid high-fat food (41). Anti-emetic therapy has been found effective in healthy subjects (42), but are not common practice since long-term data are not available. Interestingly, in one systematic analysis, background use of metformin was associated with more nausea and vomiting when using a GLP-1RA (43). However, whether this is also true for the combination with semaglutide, or whether lowering the dose of metformin has effect, has not been studied.

The mechanisms behind nausea/vomiting and diarrhea are incompletely understood. For nausea, a relation with the inhibiting effects on gastric emptying seems plausible. However, nausea also occurs in the fasting state (44), and is not related to measures of gastric emptying speed after meal ingestion (45). An effect on the central nervous system has been suggested as a recent study with modified exenatide—with reduced brain penetrance—showed less vomiting in musk shrews, despite retaining effects on glucose control (46). For diarrhea, studies are lacking. In one study, osmotic diarrhea occurred 8 h after infusion of GLP-1 peptide (47), and GLP-1RAs have been shown to reduce intestinal uptake of glucose and lipids (48, 49). Also, in patients with type 1 diabetes, liraglutide reduced colon transit time (50). So, hypothetically, semaglutide could induce diarrhea by altering nutrient absorption or intestinal motility.

Finally, although nausea and vomiting are perhaps unwanted effects, they may also be partly responsible for aspects of the drug’s efficacy as indicated above. As such, in some studies, nausea induced by GLP-1RAs is linked to weight loss (51, 52). For example, obese subjects treated with high-dose liraglutide who experienced (transient) nausea had on average 2.9 kg (95%-CI 0.5–5.3) more weight loss compared to those without GI events (51). In a mediation analysis of the SUSTAIN 1 to 5 trials, a small component (0.07 to 0.5 kg) of the total treatment difference in weight loss was explained by nausea or vomiting (52). In contrast, when combining data from SUSTAIN 3, 7 and 10, the occurrence of nausea and vomiting was not associated with superior weight loss (53). Whether this route plays a role in the beneficial effects of GLP-1RA on body weight needs further studying.

Within years of the introduction of GLP-1RAs, these agents were linked to the occurrence of acute pancreatitis, and suggested to potentially cause pancreatic cancer (6). In the subsequent years, many pharmacovigilance and database studies followed, with conflicting results (54–60). Given the nature of observational studies, data could have been confounded, since patients with diabetes whom have an indication for GLP-1RA therapy often have concomitant risk factors for pancreatitis (notably obesity, longer diabetes duration and co-medication). As such, the longer term CVOTs were a welcome addition to the discussion. When focusing on semaglutide, no signals of pancreatic AEs were present with blinded adjudication. In SUSTAIN 6, acute pancreatitis occurred in 9 semaglutide-treated patients, and in 12 placebo-treated patients. Pancreatic cancer occurred in one and four patients, respectively (28). In PIONEER 6, acute pancreatitis occurred in one semaglutide-treated patient, and in three placebo-treated patients (19). The incidence of pancreatic cancer was not reported. When combining all phase 3a data, pancreatitis occurred in five semaglutide-treated patients in PIONEER (six in the comparator group), and in 15 patients in SUSTAIN (13 in the comparator group). However, it is possible that for a relatively rare complication (the background incidence of pancreatitis and pancreatic cancer in T2D patients is 422 and 15–24 per 100,000 person-years, respectively (61, 62)), the CVOTs and phase 3 studies are of insufficient power to show differences between groups. When combining all available CVOT data (including those from non-semaglutide GLP-1RAs) in a meta-analysis, a hazard ratio of 1.05 (95% confidence interval [CI] 0.78–1.40) was found for pancreatitis and 1.12 (95% CI 0.77–1.63) for pancreatic cancer (63). These data thus argue against an effect of GLP-1RA on pancreatitis and pancreatic cancer incidence. However, one can wonder whether the follow-up duration in the CVOTs (ranging from a median of 1.3 to 5.4 years) is long enough for patients to develop pancreatic cancer.

While establishing (the absence of) a link with pancreatitis and pancreatic cancer in large clinical studies was one aspect in this field of research, others focused on animal studies and more mechanistic findings. One consistent finding is a subtle and asymptomatic increase in plasma lipase and amylase level (64, 65), which occurs within hours of administration (66). In a 26-week randomized controlled trial, oral semaglutide dose-dependently increased lipase levels by 9 to 55% and subcutaneous semaglutide by 36% (38). An increase in enzyme levels was not associated with occurrence of pancreatic events in trials with liraglutide (67, 68). Moreover, our group previously demonstrated that the liraglutide-induced increase in pancreatic enzymes is not associated with changes in pancreatic exocrine function or pancreas size measured by magnetic resonance imaging (69). Such studies have not yet been conducted for semaglutide.

A handful of preclinical studies showed that GLP-1RAs induce pancreatic inflammation, cellular proliferation and intra-epithelial neoplasia (PanIN) (70–72). However, the majority of animal studies did not find any effect of GLP-1RAs on pancreatic physiology, even with doses up to 240 the normal human dose (73–76). Preclinical studies with semaglutide also found no adverse signals in pancreatic tissue (76). Although pancreatic adverse events are difficult to completely rule out, an assessment by the FDA and the EMA concluded that a causal association between incretin-based drugs and pancreatitis or pancreatic cancer is inconsistent with the current data (77).

Both formulations of semaglutide have received an official box warning for thyroid C-cell tumors in the US. This caution is solely based on data from rodent studies and is not unique for semaglutide amongst the GLP-1RA. In rodents, the thyroid C-cells (neuroendocrine parafollicular cells which secrete calcitonin) highly express the GLP-1 receptor (78). Stimulation leads to upregulation of the calcitonin gene expression, calcitonin synthesis, C-cell hyperplasia, and increased risk of medullary adenomas and carcinomas (78). Initial studies found expression of the GLP-1 receptor in healthy human thyroid tissue, as well as in medullary thyroid carcinoma (MTC) and C-cell hyperplasia (79, 80). However, these studies were later refuted, as incompletely validated GLP-1 receptor antisera were used (81). When using validated antibodies, the GLP-1 receptor is only marginally expressed in thyroids of non-human primates and humans (78, 82). Supporting this is the observation that monkeys treated with >60 times the human dose of liraglutide do not develop C-cell abnormalities after 20 months (78).

In the SUSTAIN program, three adjudicated events of malignant thyroid neoplasm were identified, two in semaglutide-treated patients (combined n = 5,933), and one in the comparator group (n = 4,736) (10–13, 22, 26–32). None of these were medullary carcinoma. Serum calcitonin was measured during these trials, and no notable difference in mean levels was seen between the treatment arms. In the PIONEER program, four thyroid malignancies occurred in semaglutide-treated patients, versus one in the comparator group (14–21, 23, 24). In one instance, a MTC developed in a patient with preexisting nodules and elevated calcitonin at baseline (19). When looking at long-term data from the LEADER trials, the CVOT for liraglutide, there was no difference between liraglutide and placebo regarding calcitonin levels and C-cell malignancies (83).

It should be noted that MTC is rare (estimated incidence of 0.2 cases per 100,000 patient-years), and as such, it is very difficult to definitively rule out an association between GLP-1RA and thyroid malignancies. Therefore, regulatory authorities required additional pharmacovigilance activities, by systematically monitoring the annual incidence of MTC in the US for at least 15 years (MTC-22341, results expected by 2035–2037). In the meantime, semaglutide is contraindicated in patients with a personal or family history of MTC, as well as in patients with multiple endocrine neoplasia (MEN) type 2 in the US.

In the SCALE-trial, high-dose liraglutide for the treatment of obesity was associated with an increased risk of gallbladder events compared to placebo (2.5% versus 1.0% of patients, respectively) (84). Based on AEs as reported in the European EudraVigilance database, gallbladder disease is likely not limited to liraglutide, but affects all incretin-based therapies (85). A recent meta-analysis observed an increased risk of 28% for cholelithiasis with GLP-1RA treatment (86), but a breakdown for each agent was not given. In the SUSTAIN program, 83 patients (1.4%) treated with semaglutide developed a gallbladder event, compared with 39 patients (1.9%) in the placebo group (10, 27, 28, 31, 32). The events mainly included cholelithiasis. In the PIONEER program, cholelithiasis occurred more often in the semaglutide-treated group (0.6% versus 0.1% with placebo), while the risk of cholecystitis was similar (data derived from the summary of product characteristics (SmPC) (87), as the manuscripts did not describe these data). Importantly, none of the gallbladder events have been linked to mortality. Cholelithiasis has been included in the SmPC of both subcutaneous and oral semaglutide).

Initially the gallbladder events were attributed to GLP-1RA-induced weight loss, as for example in the SCALE and LEADER trials, the patients with gallbladder events had more than average weight loss (84, 88). However, as gallbladder disease is not an issue with sodium-glucose co-transporter-2 (SGLT-2) inhibitors (with similar weight loss) (89), and gallbladder events also occurred in GLP-1RA-treated patients well after weight reduction (90), other mechanisms are possibly in play. One option could be lower gallbladder motility, which enhances biliary sludge formation and bile stones. In acute intervention studies, exenatide and albiglutide reduced cholecystokinin-induced gallbladder emptying (91, 92). However, after 12-week liraglutide intervention, we were unable to demonstrate an effect on gallbladder emptying (93), while Nexøe-Larsen et al. observed that liraglutide prolonged the time to reach maximum gallbladder emptying (94). Another mechanism is a change in bile salts, leading to supersaturated bile. While we observed changes in deoxycholic acid levels in plasma and fecal samples after liraglutide treatment, the clinical relevance remains unclear (93). Fascinatingly, exendin-4 appears to stimulate cholangiocyte proliferation through the GLP-1 receptor, hereby preventing cholangiocyte apoptosis in models of bile acid-induced damage and models of ductopenic cholangiopathies (95, 96). Although these data are considered beneficial, it also indicates that GLP-1RA could have direct adverse effects on the biliary tree. What the exact mechanism is behind the gallbladder events requires further study, but probably encompasses a combination of factors.

All GLP-1RAs increase heart rate, and this is not different for semaglutide. In SUSTAIN 6, a placebo-corrected heart rate increase of 2.75 beats per minute (bpm) was observed for semaglutide 0.5 mg, and 3.2 bpm for the 1.0 mg dosage (97). This increase was not associated with adverse cardiac events.

In addition, no increase in cardiovascular outcomes were observed in SUSTAIN 6 and PIONEER 6, which is reassuring given the initial fear of adverse cardiac events with increased resting heart rates. Large epidemiological studies have found that an increase in 5 bpm is associated with an increase of 17% in mortality (98). It is unclear whether this association holds true for drug-induced heart-rate acceleration. The α-blocking agent doxazosin increases heart rate by ~25% (99), and is associated with an increase in heart failure incidence (compared with the diuretic agent chlorthalidone) (100). In contrast, lowering heart rate by approximately 10 bpm using the cardiac funny-channel inhibitor ivabradine did not affect mortality in patients with stable coronary artery disease. At this point, it is clear that the beneficial effects of GLP-1RA on cardiovascular risk factors and physiology outweigh a potential risk of the associated heart rate increase. Liraglutide has been on the market for 10 years, but cardiovascular safety beyond this has not been studied yet.

The increase in heart rate is also of importance in patients with heart failure (HF). While the semaglutide CVOTs did not show an increased incidence of hospitalization for HF compared to placebo (101), in earlier smaller studies with liraglutide in patients with HF with reduced left ventricular ejection fraction, the GLP-1RA was associated with increased incidence of serious cardiac events (rhythm disorders, worsening of HF) (102, 103). Since patients with HF with New York Heart Association class IV were excluded from the CVOTs, it is unclear whether safety risks could occur in semaglutide-treated patients. However, a recent meta-analysis of all current CVOTs, showed that GLP-1RAs as a group were associated with a (non-significant) reduction in HF (104).

Several clinical mechanistic trials provided conflicting evidence while aiming to understand the GLP-1RA-induced heart rate-increase. Some studies found systemic vasodilation (with likely consequent reflex tachycardia), while others failed to show this (105–107). Similarly, discrepant findings are available for activation of the (cardiac) sympathetic nervous system (106, 108–111). Our own group previously hypothesized a direct effect of GLP-1RAs on sino-atrial cells (106), after exclusion of other potential causes. This postulation was later confirmed in a mouse model, where stimulation of GLP-1 receptors on atrial cells induced a chronotropic effect, but only when neuronal input was present (112).

Most novel drugs also undergo testing for their effect on the QT interval, as QT prolongation is a marker for potential ventricular fibrillation. Compared with placebo, subcutaneous semaglutide had no effect on this ECG measure in healthy volunteers, with doses above what is used in daily practice (113).

Initial case reports suggested that GLP-1RA treatment could cause acute kidney injury (AKI) in some patients (114). Mechanistically, this was explained by dehydration caused by nausea, vomiting and diarrhea (see above). Also, very recently it was shown that the GLP-1RA, dulaglutide decreased fluid intake (115). Furthermore, GLP-1RA potentially further compromise fluid homeostasis by increasing renal sodium excretion (116). Combined, this could induce renal failure, especially in frail patients or those with medication such as renin–angiotensin–aldosterone system inhibitors, non-steroidal anti-inflammatory drugs or diuretic drugs.

In the SUSTAIN program, acute kidney failure was only reported in SUSTAIN 6, where its occurrence was similar between semaglutide and placebo (28). In PIONEER, AKI was a safety event of interest, and reported in all papers (14, 17–19, 21, 23). In PIONEER 6, AKI occurred in 2.0% of patients treated with oral semaglutide and 2.3% of placebo-treated patients (19). Whether this is statistically or clinically significant has not been evaluated yet.

In contrast to the incidental cases of AKI, the CVOTs mainly demonstrate a beneficial effect on renal outcomes, likely because of effects on cardiovascular risk factors (117). As recently reviewed, GLP-1RAs reduce progression to macro-albuminuria and lead to (subtle) reductions in the decline in renal function (118). In a recent post-hoc analysis of SUSTAIN 6, semaglutide was associated with less events of nephropathy, independent of baseline blood pressure (119). Thus, while it is conceivable yet not statistically confirmed that semaglutide could cause AKI in selected patients, there is plenty of evidence that it reduces nephropathy in the long term. A dedicated kidney trial (the FLOW study; NCT03819153↗) is currently ongoing, studying the effects of subcutaneous semaglutide on renal outcomes in people with T2D and chronic kidney disease.

In the SUSTAIN-6 trial, an increase in DRP complications, defined as a composite of need for retinal photocoagulation or treatment with intravitreal agents or vitreous hemorrhage or diabetes-related-blindness, was reported for semaglutide compared to placebo (hazard ratio 1.76; 95% CI 1.11–2.78). In a large systemic review and network analysis, including several GLP-1RAs, subcutaneous semaglutide was the only glucose-lowering drug for which this signal was observed (120). However, in the LEADER trial, a non-significant trend towards DRP was observed for liraglutide (121). In PIONEER 6, unadjudicated DRP occurred in 5.8% of oral semaglutide-treated patients and in 4.8% of the placebo-treated patients (19).

Villsbol and colleagues further investigated the DRP signal in the SUSTAIN program (122). In SUSTAIN-6, nearly 30% of patients had previous documented DRP, with 6% proliferative DRP. This percentage was not surprising given the inclusion of patients with previous cardiovascular disease, usually associated with long-standing diabetes. In semaglutide-treated patients, 3% (versus 1.8% in the placebo group) of patients reached an adjudicated endpoint of DRP. Across all DRP categories as indicated above, more events with semaglutide were noted. Participants that were prone to develop DRP had pre-existing DRP, longer diabetes duration, higher HbA_1c levels at baseline, and more often used insulin therapy. Particularly, participants with pre-existing DRP who were using insulin therapy had the highest risk for a new DRP event.

This analysis further assessed whether the increase in DRP was a GLP-1 specific effect, or rather caused by a robust and early glucose lowering as suggested by several other studies, where acute and large reductions in glucose concentrations may initially and transiently worsen DRP, yet prevent or delay onset or progression of DRP in the long term (123–128). Patients that met a DRP endpoint had strongest glucose lowering during the trial, independent of their randomization to semaglutide or placebo. A post-hoc mediation analysis adjusting for HbA_1c reduction at week 16 showed that glucose reduction at this time point explained the increased incidence. Limitations of DRP assessment during the trial were the absence of assessment of retinal changes over time, while the severity was not graded on baseline. Nevertheless, based on the data brought forward, it seems safe to conclude that the phenomenon of early worsening of pre-existing DRP was secondary to the initial and rapid improvement in glycemic control that occurred in SUSTAIN-6. This was confirmed in the recent AngioSafe study which showed no effect of GLP-1RA therapy on angiogenesis and no association between GLP-1 exposure and severe DRP was shown (129).

Currently, a large trial is ongoing assessing the long-term effects of semaglutide on DRP in patients with T2D as primary outcome (FOCUS trial, NCT03811561↗). This study will provide important data with respect to semaglutide safety on the retina. Until that time, caution should be exercised when using semaglutide in patients with DRP. It may be sensible to perform a fundoscopy prior to semaglutide therapy, and existing DRP should be treated where necessary. In addition, given the strong effects of semaglutide on glucose levels, down titrating insulin will prevent rapid decreases in glucose concentrations thereby reducing the risk of acute DRP worsening.

Although every subcutaneous injection can induce injection-site reactions, there are no signals that this is higher with semaglutide compared with placebo (130). In phase 3 studies, any site reaction was present in 0.6% of patients on the 0.5 mg dose, 0.3% on the 1 mg dose, and 0.8% in the comparator groups. The local site reaction includes bruising, discoloration, induration, and pain (130). In SUSTAIN-6, none of these injection site-reactions was considered severe, and it was never a reason to withhold therapy.

Given the immunogenic potential of protein-based drugs, it is important to monitor allergic reactions with GLP-1RAs. Allergic reactions were reported in four patients in the SUSTAIN program. However, at closer inspection, these reactions were more likely caused by the (concomitant) use of angiotensin-converting enzyme inhibitors or an infection (130). Across the phase 3a PIONEER trials, less subjects with oral semaglutide (2.9%) had allergic reactions compared with the comparators (4.6%) (131). No cases of anaphylactic reactions have yet been attributed to semaglutide; one patient using semaglutide had an anaphylactic shock attributed to cefazolin in SUSTAIN-6.