Incretin-Based Drugs and the Risk of Acute Liver Injury Among Patients With Type 2 Diabetes

Dipeptidyl peptidase 4 (DPP-4) inhibitors and glucagon-like peptide 1 receptor agonists (GLP-1 RAs) are the frequently prescribed second- to third-line drugs in type 2 diabetes (1–3). Compared with other antidiabetes drugs, they are associated with a lower risk of hypoglycemia and have neutral to favorable effects on body weight and cardiorenal outcomes (1). However, their increasing use has uncovered potential safety concerns, such as acute liver injury (4–14).

Specific concerns about an association between incretin-based drugs and liver injury emerged with case reports following the administration of these drugs (4–8,10–14). In addition, a pharmacovigilance analysis using the U.S. Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) found an increased frequency of liver injury with specific DPP-4 inhibitors (reporting odds ratios ranging between 1.33 and 6.51) (9). Postmarketing surveillance has also linked DPP-4 inhibitors with fatal outcomes in liver injury cases (15), resulting in the inclusion of liver injury as a label warning in the case of alogliptin (16). Similarly, the U.S. FDA has recently issued a warning against GLP-1 RAs for liver injury based on safety signals identified in the FAERS database (17). However, the interpretation of the current evidence is limited by potential reporting bias. To date, real-world evidence on this possible association is limited (18).

Given the increasing use of incretin-based drugs (3) and the fact that patients with type 2 diabetes are already at an increased risk of acute liver failure (19–21), we conducted a cohort study to determine whether DPP-4 inhibitors and GLP-1 RAs, separately, are associated with an increased risk of liver injury compared with sodium-glucose cotransporter 2 (SGLT-2) inhibitors among patients with type 2 diabetes.

We conducted this study using the GOLD and Aurum databases of the U.K. Clinical Practice Research Datalink (CPRD). The CPRD is a population-based database of electronic health records of >50 million patients across 2,000 general practices in the U.K. In the CPRD, diagnoses and procedures are recorded using the Read code or Systematized Nomenclature of Medicine Clinical Terms (SNOMED-CT) classification, and prescriptions are coded using the U.K. Prescription Pricing Authority Dictionary. Additionally, the CPRD collects information on lifestyle and anthropometric variables (e.g., smoking, alcohol use, BMI), clinical measures (e.g., blood pressure), and laboratory test results (e.g., hemoglobin A_1c). These variables have been validated (15–17), have a high degree of completeness, and the data and practices are audited regularly to ensure high quality. For this study, the CPRD was linked with the Hospital Episode Statistics Admitted Patient Care (HES APC), a database covering all hospital admissions at National Health Services hospitals in England, and the Office for National Statistics (ONS) database, which records vital statistics data in the U.K. Diagnoses in the HES and ONS are recorded using the ICD-10 code classification.

The study protocol was approved by the CPRD Research Data Governance (protocol 21_000691), London, U.K., and by the Jewish General Hospital Research Ethics Board, Montreal, Quebec, Canada.

We assembled two new-user, active-comparator cohorts. The first cohort consisted of new users of GLP-1 RAs (dulaglutide, exenatide, liraglutide [except the 3 mg/0.5 mL formulation indicated for weight loss], lixisenatide, or semaglutide) and new users of SGLT-2 inhibitors (canagliflozin, dapagliflozin, or empagliflozin) from 1 January 2013 (the year the first SGLT-2 inhibitors entered the U.K. market) through 31 December 2019. The second cohort consisted of new users of DPP-4 inhibitors (alogliptin, linagliptin, saxagliptin, sitagliptin, or vildagliptin) and new users of SGLT-2 inhibitors (canagliflozin, dapagliflozin, or empagliflozin) from 1 January 2013 through 31 December 2019. Cohort entry was defined as the date of the first-ever prescription for one of the drugs of interest (GLP-1 RAs or DPP-4 inhibitors) or SGLT-2 inhibitors during the study period.

We selected SGLT-2 inhibitors as the active comparator in both cohorts because they are widely used second- to third-line drugs (3), constitute immediate therapeutic alternatives to GLP-1 RAs and DPP-4 inhibitors (1), and have not been associated with liver injury (22,23). Other comparators, such as metformin, used in early diabetes, and insulin, used in advanced diabetes, were not selected as they are usually used at different stages of the disease, which might result in confounding due to time lag bias (24). Finally, drug classes such as sulfonylureas and thiazolidinediones were not used as comparators as they have been associated with acute liver injury (25–30).

To be included in the cohorts, patients had to be at least 18 years of age and have at least 1 year of medical history in the CPRD before cohort entry; the latter was used as a washout period to identify new users. We excluded patients concomitantly prescribed the study drugs at cohort entry and those who previously used one of the study drugs at any time before cohort entry (i.e., GLP-1 RAs in the DPP-4 inhibitor vs. SGLT-2 inhibitor cohort and DPP-4 inhibitors in the GLP-1 RA vs. SGLT-2 inhibitor cohort). Patients diagnosed with end-stage renal disease at any time before cohort entry were also excluded, as GLP-1 RAs and SGLT-2 inhibitors are contraindicated in these patients. Finally, we excluded patients previously diagnosed with alcohol-related disorders and acute and chronic liver diseases at any time before cohort entry.

All patients were followed while continuously exposed to the cohort entry drugs by using an on-treatment approach. Patients were considered continuously exposed if the duration of consecutive prescriptions overlapped each other. We used a 60-day grace period in the event of nonoverlapping consecutive prescriptions. Thus, patients were followed from cohort entry until an incident diagnosis of acute liver injury (detailed below), treatment discontinuation or switching to one of the study drugs, death from any cause, end of registration with the CPRD, or the end of the study period (31 March 2020), whichever occurred first.

Patients with incident acute liver injury were identified based on the first of the following events during the follow-up period: 1) diagnostic codes adapted from a validated algorithm (based on Read and SNOMED CT codes) (31,32), 2) liver enzyme–based definition in the CPRD (ALT five times the upper normal limit, or alkaline phosphatase twice the upper normal limit, or ALT three times the upper normal limit along with bilirubin twice the upper normal limit) (33,34), or 3) diagnostic codes in the HES APC or ONS databases (identified using ICD-10 codes with positive predictive values of 60–84%) (Supplementary Table 1) (35).

We considered potentially important confounders, all measured before or at cohort entry. These included age (modeled using cubic splines with five interior knots), sex, BMI, smoking status, and calendar year (entered as a categorical variable). We also included variables as proxies for diabetes severity, including hemoglobin A_1c, diabetes duration (defined by the date of the first of either a hemoglobin A_1c ≥6.5%, a diagnosis of type 2 diabetes, or prescription for any antidiabetic drug), microvascular complications (nephropathy, neuropathy, and retinopathy), macrovascular complications (myocardial infarction, other atherosclerotic diseases, stroke, and peripheral arteriopathy), and previous use of antidiabetes drugs in the year before and including cohort entry. The models also included prescription drugs previously associated with liver injury (statins, fibrates, renin-angiotensin system inhibitors, antimicrobials [including antitubercular drugs and highly active antiretroviral therapy], amiodarone, halothane, nonsteroidal anti-inflammatory drugs, steroids [including glucocorticoids, anabolic steroids, oral contraceptives, and hormone replacement therapy], antidepressants, antiepileptics, antipsychotics, anticancer agents, disease-modifying antirheumatic drugs, and biologics). We also included diseases of the biliary tree, cancer (other than nonmelanoma skin cancer), heart failure, autoimmune diseases, and markers of health-seeking behavior (uptake of cancer screening [fecal occult blood testing or colonoscopy, mammography, or prostate-specific antigen testing] and influenza and pneumococcal vaccinations).

We used propensity score fine stratification to adjust for confounding (36). In each new-user cohort, we used multivariable logistic regression to calculate the predicted probability of receiving the drug of interest (GLP-1 RA or DPP-4 inhibitor) versus an SGLT-2 inhibitor conditional on the covariates listed above. Patients in the nonoverlapping regions of the propensity score distributions were trimmed from the cohorts, and 50 strata were created based on the propensity score of the exposed patients (36). Within each stratum, we assigned patients exposed to incretin-based drugs a weight of 1, while patients exposed to SGLT-2 inhibitors were weighted in proportion to the number exposed in the corresponding stratum. The estimand generated by this approach is the average treatment effect among the treated.

Descriptive statistics were used to summarize the characteristics of the exposure groups before and after propensity score weighting. We used standardized differences to assess covariate balance between the exposure groups, with differences <0.10 indicating good balance (37). For each exposure group, we calculated the incidence rates of acute liver injury with 95% CIs based on the Poisson distribution. We plotted weighted Kaplan-Meier curves to display the cumulative incidence of acute liver injury for each exposure group. Weighted Cox proportional hazards models were fit to estimate hazard ratios (HRs) with 95% CIs of incident acute liver injury, comparing GLP-1 RAs and DPP-4 inhibitors, separately, with SGLT-2 inhibitors. Finally, we calculated the number needed to harm after 1 and 5 years of use using the Kaplan-Meier method (38).

We conducted three secondary analyses. First, we assessed whether there is a duration-response relation between the use of GLP-1 RAs and DPP-4 inhibitors and the incidence of acute liver injury. For this analysis, all exposures were modeled using a time-varying approach, where we updated each person-day according to duration of use. This was obtained by adding the prescription durations of the drugs of interest from cohort entry until the time of the risk set (i.e., time of the event). This was then stratified into the following categories: <6 months, 6 months–1 year, and >1 year. It is important to note that as part of the exposure definition, it was possible for patients to contribute person-time in each of these duration categories, thereby eliminating the possibility of immortal time bias. Second, we assessed whether the association varies with individual drugs within each class. Third, we examined potential effect measure modification by age (<75 years and ≥75 years), sex, use of drugs previously associated with acute liver injury, and history of autoimmune diseases. We tested effect measure modification by including interaction terms between the exposure and these variables in the models.

We conducted four sensitivity analyses to assess the robustness of our assumptions. First, we repeated the analysis by varying the grace period between nonoverlapping consecutive prescriptions to 30 and 90 days. Second, we used time-varying inverse probability of censoring weighting, with weights updated every 60 days to assess the impact of potential informative censoring (39–41). This involved taking the product of the weights calculated from the conditional probabilities of treatment discontinuation or switching, death from any cause, and administrative censoring conditional on the covariates listed above. Third, we used an intention-to-treat exposure definition, censored at 1 year to limit exposure misclassification. Finally, to account for the possibility that our results may be driven by differential testing rates, we restricted our outcome to diagnostic codes of acute liver injury recorded in the CPRD and the HES and ONS databases. As an ancillary analysis, we repeated the analyses using neuropathy as a negative control outcome to assess the impact of residual confounding. This outcome has been associated with disease severity, but not with the drugs of interest. This analysis was based on the same cohort as the one used in the primary analysis but involved excluding patients diagnosed with neuropathy at any time before cohort entry. All analyses were conducted with SAS 9.4 software (SAS Institute, Cary, NC).

The first cohort consisted of 106,310 new users of DPP-4 inhibitors and 27,277 new users of SGLT-2 inhibitors (Supplementary Fig. 1). The propensity score distribution for the DPP-4 inhibitor and SGLT-2 inhibitor users are shown in Supplementary Fig. 2. Before weighting, DPP-4 inhibitor users were older, less likely to be obese, and had a longer duration of diabetes than SGLT-2 inhibitor users (Table 1). After weighting, the exposure groups were well-balanced across all covariates (Table 1). This cohort was followed for a median of 1.1 years (interquartile range 0.4–2.4), during which time 529 patients experienced incident acute liver injury events.

The results of the primary analysis are presented in Table 3. The use of DPP-4 inhibitors was associated with an overall 53% increased risk of acute liver injury compared with the use of SGLT-2 inhibitors (2.5 vs. 1.7 per 1,000 person-years, respectively; HR 1.53, 95% CI 1.02–2.30). The cumulative incidence curves diverged relatively soon after treatment initiation (Fig. 1A), with the use of DPP-4 inhibitors associated with a 78% increased risk after 1 year of use (HR 1.78, 95% CI 0.99–3.22), although the CI included the null (Supplementary Fig. 3). The number needed to harm was 1,990 at 1 year and 245 at 5 years.

In secondary analyses, there was no effect modification by age, the use of drugs previously associated with liver injury, and autoimmune diseases. However, an increased risk of acute liver injury was observed among women (HR 3.22, 95% CI 1.67–6.21), but not among men (HR 1.13, 95% CI 0.70–1.83), although CIs overlapped (Supplementary Fig. 3). Among the individual drugs, linagliptin was associated with a high risk of liver injury (HR 2.29, 95% CI 1.34–3.92) (Supplementary Fig. 3). Overall, the results of the sensitivity analyses were generally consistent with the primary analyses, except for the analysis restricting the outcome definition to diagnostic codes of acute liver injury generating an elevated HR of 2.58 (95% CI 1.43–4.65) (Supplementary Fig. 4).

The second cohort consisted of 9,470 new users of GLP-1 RAs and 26,936 new users of SGLT-2 inhibitors (Supplementary Fig. 5). The propensity score distribution for the GLP-1 RAs and SGLT-2 inhibitor users is shown in Supplementary Fig. 6. Before weighting, GLP-1 RA users were more likely to be women, obese, and have a longer duration of diabetes. After weighting, the covariates were well balanced between the exposure groups (Table 2). Over a median follow-up of 0.9 years (interquartile range 0.4–1.9), this cohort generated 87 incident acute liver injury events.

Table 3 presents the results of the primary analysis. GLP-1 RA use was not associated with an overall increased risk of liver injury compared with the use of SGLT-2 inhibitors (2.0 vs. 1.8 per 1,000 person-years, respectively; HR 1.11, 95% CI 0.57–2.16). The cumulative incidence curves overlapped throughout the follow-up period (Fig. 1B), with no clear association in the analysis stratified by duration of use (Supplementary Fig. 7).

In secondary analyses, women, but not men, were at an increased risk of acute liver injury (HR 3.23, 95% CI 1.44–7.25 and HR 0.69, 95% CI 0.30–1.58, respectively), although the CIs overlapped. There was no effect measure modification by age, use of drugs previously associated with liver injury, or autoimmune diseases (Supplementary Fig. 7). Similarly, no individual drug was associated with an increased risk of acute liver injury (Supplementary Fig. 7). Similar findings were observed in sensitivity analyses (Supplementary Fig. 8).

In the ancillary analysis using neuropathy as a negative control outcome, there were 98,379 and 28,285 users of DPP-4 inhibitors and SGLT-2 inhibitors, respectively, and 8,954 and 28,165 users of GLP-1 RAs and SGLT-2 inhibitors, respectively. In both cohorts, the use of either DPP-4 inhibitors (HR 0.93, 95% CI 0.85–1.02) or GLP-1 RAs (HR 1.13, 95% CI 0.98–1.30) was not associated with an increased risk of neuropathy when compared with SGLT-2 inhibitors (Supplementary Table 2).

The results of this large population-based cohort study indicate that compared with SGLT-2 inhibitors, DPP-4 inhibitors were associated with a 53% overall increased risk of acute liver injury. This increased risk was particularly elevated among women and linagliptin users. In contrast, GLP-1 RAs were not associated with an overall increased risk of acute liver injury, although an increased risk was observed among women.

Large cardiovascular outcome trials of DPP-4 inhibitors or GLP-1 RAs have not reported any clear signal of acute liver injury, although the EXAMINE (EXamination of CArdiovascular OutcoMes: AlogliptIN vs. Standard of CarE in Patients with Type 2 Diabetes Mellitus and Acute Coronary Syndrome) trial of alogliptin reported a numerically higher incidence of liver enzyme elevation compared with placebo (ALT more than three times the upper limit of normal: alogliptin 2.4%, placebo 1.7%) (42), raising concerns (6). This was compounded by postmarketing surveillance reports of acute liver injury events (15). To our knowledge, only one real-world study investigated whether incretin-based drugs are associated with an increased risk of acute liver injury (18). In that study, saxagliptin was compared with non–DPP-4 inhibitor antidiabetes drugs using two U.S. (HealthCore Integrated Research Database, Medicare) and two U.K. (CPRD, The Health Improvement Network) databases (18). However, that study was underpowered as no cases of acute liver injury were recorded in the saxagliptin group (18). Overall, our findings suggest an increased risk of acute liver injury with the use of DPP-4 inhibitors. This corroborates prior case reports and pharmacovigilance analyses that have documented this potential association with this drug class (6–14). Our study adds to the existing literature by finding that the absolute risk associated with this potentially life-threatening adverse event is reassuringly low. The exact mechanism behind this increased risk remains to be elucidated. Toxicological studies of DPP-4 inhibitors in rats led to hepatic changes, including increased alkaline phosphatase, increased liver weights, and centrilobular hepatocellular hypertrophy (43). A possible hypothesis explaining this effect may be that inhibition of the DPP-4 enzyme leads to increased levels of proinflammatory chemokines such as eotaxin-1 (44–46), a molecule thought to play a major role in the pathogenic process of drug-induced liver disease (47,48). However, further research is needed to confirm this hypothesis.

Overall, while the use of GLP-1 RAs was not associated with an increased risk of acute livery injury, an increased risk was observed among women. This appears to corroborate the regulatory warning issued by the U.S. FDA in late 2021 based on signals generated in the FAERS database (17). Importantly, this aligns with the pattern observed in the two case reports of GLP-1 RA–induced acute liver injury (4,5). In both case reports of female patients, an immune-mediated mechanism was implicated, with one report also demonstrating in vitro lymphocyte transformation specifically induced by the suspect agent (4). Potential mechanisms for this might be based on immune reaction incited by the 3–50% difference in protein sequence that GLP-1 RAs have from native GLP-1.

A possible explanation of female preponderance of acute liver injury with both DPP-4 inhibitors and GLP-1 RAs is consistent with the female susceptibility of drug-induced liver injuries (49). Several reasons for this sex-based difference have been proposed. They include sex differences in pharmacokinetics of different drugs, preponderance of aberrant immune responses, such as immune-mediated hepatotoxicity, in women, and hormone-mediated interaction with signaling molecules that may affect drug safety (49). However, further studies need to corroborate the results of higher risk of acute liver injury with DPP-4 inhibitors and GLP-1 RAs in women given the CIs overlapped with the effect estimate in men.

Overall, our study has several strengths. First, the use of the CPRD and linked databases allowed us to identify acute liver injury using diagnostic codes and laboratory results in the outpatient setting as well as in the inpatient setting and death certificates. Second, the use of the CPRD also allowed us to control for potentially important confounders (e.g., smoking, BMI) often absent in administrative databases. Finally, we used an active comparator that has not been associated with acute liver injury, reducing the potential for confounding by indication.

However, our study has some limitations. First, exposure misclassification is possible because the CPRD is a general practitioner database and does not record prescriptions written by specialists. However, this is unlikely to be an important source of misclassification since general practitioners almost entirely manage type 2 diabetes in the U.K. (50).

Second, outcome misclassification is a possibility. However, the definition of liver injury identified using Read and ICD-10 codes has been validated in multiple studies (31,35). Consequently, it is unlikely that outcome misclassification was substantial and differential between the exposure groups.

Finally, as with any observational study, residual confounding is possible. To mitigate this limitation, the propensity score models considered a wide range of potential confounders. Furthermore, as acute liver injury is not a well-known adverse effect of incretin-based drugs, it is unlikely that patients on these drugs experienced any substantial degree of channeling, reducing the risk of confounding. Reassuringly, our analyses using neuropathy as a negative control outcome generated point estimates close to the null value, suggesting good control of confounding.

In summary, the results of this large population-based study indicate that overall, DPP-4 inhibitors, but not GLP-1 RAs, are associated with an increased risk of acute liver injury. However, an increased risk was observed among women, which was the case for both classes of incretin-based drugs. While our results need replication, physicians should balance the potential low absolute risk of acute liver injury of the incretin-based drugs with their known clinical benefits in patients with type 2 diabetes.

Funding. R.P. is the recipient of a Fonds de Recherche du Québec–Santé Doctoral Award. O.H.Y.Y. holds a Fonds de Recherche du Québec–Santé Chercheur-Boursier Clinicien Junior 1 Award. L.A. holds a Fonds de Recherche du Québec–Santé Chercheur-Boursier Senior Award and is the recipient of a McGill University William Dawson Scholar Award. This study was funded by a Canadian Institutes of Health Research Foundation Scheme grant (FDN-143328).

The sponsors had no influence on design and conduct of the study, collection, management, analysis, and interpretation of the data, and preparation, review, or approval of the manuscript.

Duality of Interest. L.A. received consulting fees from Janssen Pharmaceuticals and Pfizer for work unrelated to this project. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. R.P. wrote the manuscript. R.P., O.H.Y.Y., and L.A. designed the study. R.P., H.Y., and L.A. did the statistical analyses. L.A. designed the study and acquired the data. All authors analyzed and interpreted the data, critically revised the manuscript, conceived and designed the study, approved the final version of the manuscript, and agreed to be accountable for the accuracy of the work. L.A. 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.