Dipeptidyl Peptidase 4 Inhibitors and the Risk of Bullous Pemphigoid Among Patients With Type 2 Diabetes

Dipeptidyl peptidase 4 (DPP-4) inhibitors are recommended as second- to third-line drugs in the management of type 2 diabetes (1). By inhibiting the DPP-4 enzyme, these drugs increase insulin production, while also lowering the risk of hypoglycemia and having neutral effects on body weight, when compared with other antidiabetic drugs such as sulfonylureas or insulin (1). However, there are concerns that inhibition of the DPP-4 enzyme may also lead to autoimmune-related adverse events such as bullous pemphigoid (BP) (2), a subepidermal blistering disease that has been associated with a doubling of the risk of mortality (3).

Over the years, case reports and pharmacovigilance analyses have suggested a link between the use of DPP-4 inhibitors and the risk of BP (4,5). In response to these safety signals, regulatory agencies such as the European Medicines Agency imposed changes to the product labels of DPP-4 inhibitors (6). However, to date, few observational studies have been conducted to assess this association (7–11). Although these studies reported either a trend toward an increased risk or a significantly increased risk associated with the use of DPP-4 inhibitors (odds ratios ranging between 1.58 and 3.16) (7–11), they had significant methodological shortcomings such as potential selection bias (7,9), potential time-window bias (7–11), or lack of adjustment for potentially important confounders (7–11), and they were underpowered to assess possible duration-response relations (7–11).

As a result of these limitations, and with certain regulatory agencies considering further action (2), additional studies are needed to assess this association. Therefore, the objective of this population-based cohort study was to determine whether use of DPP-4 inhibitors, when compared with use of other second- to third-line antidiabetic drugs, is associated with an increased risk of BP among patients with type 2 diabetes.

This study was conducted using the U.K. Clinical Practice Research Datalink (CPRD), a large primary care database shown to be representative of the general population in the U.K. (12). The CPRD uses the Read code classification to record medical diagnoses and procedures, which have been shown to be of high quality and validity (12). The database also includes demographic and anthropometric data such as smoking status and BMI, laboratory results, as well as prescription information based on the British National Formulary. The study protocol was approved by the Independent Scientific Advisory Committee of the CPRD (protocol 18_166R) and by the Research Ethics Board of the Jewish General Hospital, Montreal, Canada.

We first assembled a base cohort of patients, aged at least 18 years, newly treated for type 2 diabetes with a noninsulin antidiabetic drug (metformin, sulfonylureas, DPP-4 inhibitors, glucagon-like peptide 1 [GLP-1] receptor agonists, sodium–glucose cotransporter 2 inhibitors, prandial glucose regulators, thiazolidinediones, or acarbose) between 1 January 1988 and 31 March 2018. We excluded patients initially treated with insulin, as these patients likely had type 1 diabetes or advanced type 2 diabetes, and women with a history of polycystic ovarian syndrome, as this is another indication for metformin use. All patients were required to have at least 1 year of medical history in the CPRD before their first noninsulin prescription.

Using the base cohort, we assembled a study cohort composed of all patients initiating a new antidiabetic drug class on or after 1 January 2007 (the year the first DPP-4 inhibitor, sitagliptin, entered the market in the U.K.) (13). These patients included those newly treated with an antidiabetic drug and those who switched to or added on an antidiabetic drug from a class not previously used in their treatment history. Cohort entry was defined by the date of this new prescription. At this stage, we excluded patients diagnosed with BP at any time before cohort entry. We also excluded patients diagnosed with pemphigus, epidermolysis bullosa, mucous membrane pemphigoid, bullous dermatoses, or Stevens-Johnson syndrome at any time before cohort entry, as these are possible differential diagnoses for BP (14). Finally, we excluded patients with HIV infection, as it has been previously linked to BP (15). All patients were followed from cohort entry up until an incident diagnosis of BP (identified using Read codes) (Supplementary Table 1), death from any cause, end of registration with the general practice, or the end of the study period (31 March 2018), whichever occurred first.

We used a time-varying exposure definition in which each person-day of follow-up was classified into one of the following three mutually exclusive categories: current use of a DPP-4 inhibitor (linagliptin, saxagliptin, sitagliptin, vildagliptin, or alogliptin; alone or in combination with other antidiabetic drugs), current use of other second- to third-line antidiabetic drugs (defined as initiation of treatment with either thiazolidinediones, prandial glucose regulators, GLP-1 receptor agonists, acarbose, sodium–glucose cotransporter 2 inhibitors, or combination of antidiabetic drugs, or switch to or add-on of an antidiabetic drug, including insulin, after failure on metformin or sulfonylurea in monotherapy), or current use of metformin or sulfonylureas in monotherapy (as these are typically prescribed as first-line drugs). Of note, concomitant use of multiple classes of antidiabetic drugs could also represent short prescription overlaps between switchers rather than genuine combination treatments. For all categories, exposed person-time was defined by the prescription duration plus a 30-day grace period. Thus, patients were considered continuously exposed if the duration of one prescription overlapped with the date of the next prescription, using the grace period in the event of two nonoverlapping successive prescriptions. Since DPP-4 inhibitors are recommended as second- to third-line drugs for type 2 diabetes (1), the reference category for all analyses consisted of current use of other second- to third-line antidiabetic drugs.

We calculated crude incidence rates of BP with 95% CIs, based on the Poisson distribution, for each exposure category. Time-dependent Cox proportional hazards models were used to estimate hazard ratios (HRs) and 95% CIs of BP associated with use of DPP-4 inhibitors, when compared with use of other second- to third-line treatments. The models were adjusted for the following potential confounders measured at cohort entry: age (modeled as a continuous variable using restricted cubic splines to account for a potential nonlinear relation with the outcome) (16), sex, year of cohort entry, alcohol-related disorders (including alcoholism, alcoholic cirrhosis of the liver, alcoholic hepatitis, and hepatic failure), smoking status (current, past, and never), BMI category (≥30 kg/m² and <30 kg/m²), hemoglobin A_1c level (last measurement before cohort entry), and duration of treated diabetes. The models were also adjusted for the following variables measured at any time before cohort entry: macrovascular diabetic complications (stroke, myocardial infarction, and peripheral vascular disease), microvascular diabetic complications (nephropathy, neuropathy, and retinopathy), cancer (except for nonmelanoma skin cancer), as well as dementia and other neuropsychiatric disorders (multiple sclerosis, Parkinson disease, and depression), as these have been previously associated with BP (17). Multiple imputation was used for variables with missing data (i.e., hemoglobin A_1c, BMI, and smoking) (18,19).

We performed two prespecified secondary analyses. First, we investigated whether there was a duration-response relation according to current use of DPP-4 inhibitors on the incidence of BP. Given uncertainties related to the timing of onset reported in the literature (5), we modeled duration of use as a continuous variable using restricted cubic splines (five knots at the 5th, 27.5th, 50th, 72.5th, and 95th percentiles) (16). Second, we assessed the association between BP and use of individual DPP-4 inhibitors (sitagliptin, saxagliptin, linagliptin, and vildagliptin).

We conducted eight prespecified sensitivity analyses to assess the robustness of our findings. First, to assess possible exposure misclassification, we repeated the analysis after increasing the grace period between successive prescriptions to 60 and 90 days. Second, to minimize potential confounding by disease severity, we restricted the comparison of DPP-4 inhibitors with other second- to third-line antidiabetic drugs by not allowing concomitant use of insulin, given that insulin is often used in patients with advanced disease. Similarly, to exclude the possibility of an incretin effect, we restricted the comparison by not allowing concomitant use of GLP-1 receptor agonists. Fourth, we repeated the analysis by using a stricter outcome definition by additionally requiring a clinically supporting event (i.e., either a referral to a dermatologist and/or prescription for drugs indicated for the treatment of BP [oral or cutaneous glucocorticoids, azathioprine, chlorambucil, methotrexate, mycophenolate, cutaneous tacrolimus, tetracyclines, sulfonamides, or dapsone]) (20) in the 3 months prior to or after the BP diagnosis. Fifth, we conducted a competing risk analysis by death from any cause using the subdistribution Cox proportional hazards model proposed by Fine and Gray (21). Sixth, we censored potential differential diagnoses of BP during follow-up (i.e., pemphigus, epidermolysis bullosa, mucous membrane pemphigoid, bullous dermatoses, or Stevens-Johnson syndrome) (14). Finally, we conducted a marginal structural Cox proportional hazards model to investigate potential time-dependent confounding (22,23). This model included inverse-probability-of-treatment weights as well as inverse-probability-of-censoring weights for each patient, considering the same confounders listed above along with the time-varying use of other antidiabetic drugs. The product of these weights was used to reweigh the cohort, in which we estimated the HRs of BP associated with the use of DPP-4 inhibitors, with 95% CIs calculated using robust variance estimators.

We also conducted two post hoc sensitivity analyses. First, we assessed the potential impact of unmeasured confounding using the approach proposed by Ding and VanderWeele (24). This approach does not impose any assumptions on the unmeasured confounder or confounders, such as having an unmeasured confounder that is binary, or having no interaction between the effects of the exposure and the confounder on the outcome, or having only one unmeasured confounder. Second, we assessed the risk of BP associated with current use of GLP-1 receptor agonists (alone or in combination with other antidiabetic drugs except DPP-4 inhibitors) compared with other second- to third-line antidiabetic drugs (except DPP-4 inhibitors).

To further investigate the potential impact of residual confounding, we conducted a prespecified ancillary analysis using a propensity score analysis based on sequential cohorts (25,26). Using the study cohort defined above, we identified all new users of DPP-4 inhibitors and new users of second- to third-line antidiabetic drugs (defined by a new prescription for antidiabetic drug class not previously used in the patient’s history) at each calendar month of the study period. This generated a total of 135 sequential DPP-4 inhibitor-comparator cohorts during the study period. We then applied the exclusion criteria listed above at the time of entry in each of these sequential cohorts. Of note, patients in the comparator group eventually adding or switching to a DPP-4 inhibitor were allowed to contribute to the DPP-4 inhibitor group after the time of switch (Supplementary Fig. 1). We estimated the predicted probability (propensity score) of receiving a DPP-4 inhibitor versus another second- to third-line drug using conditional logistic regression, stratified on calendar month, conditional on the variables listed above along with drugs previously associated with BP (i.e., aldosterone antagonists [27], loop diuretics [28], and antipsychotics [27]). We then matched each DPP-4 inhibitor user chronologically to one patient (without replacement) initiating a second- to third-line antidiabetic drug in the same calendar month and propensity score, using a greedy matching algorithm and a caliper of 0.01 (26). The matched sets were followed until an incident diagnosis of BP, treatment discontinuation of cohort entry drug, death from any cause, end of registration with the practice, or end of the study period (31 March 2018), whichever occurred first. Cumulative incidence of BP was plotted using the Kaplan-Meier method. We used a Cox proportional hazards model to estimate the HR and 95% CIs of BP, comparing current use of DPP-4 inhibitors with current use of other second- to third-line drugs. We accounted for within-person correlation using robust SEs (26). All analyses were conducted with SAS version 9.4 (SAS Institute, Cary, NC) and R (R Foundation for Statistical Computing, Vienna, Austria).

The cohort included 168,774 new users of antidiabetic drugs (Supplementary Fig. 2). The median (maximum) follow-up was 3.8 years (11.2), generating a total of 711,311 person-years. During follow-up, 38,325 patients received DPP-4 inhibitors. Overall, 150 patients were newly diagnosed with BP during the study period, corresponding to an incidence rate of 21.1 (95% CI 17.9–24.8) per 100,000 person-years. Most of these events (n = 136; 91%) were accompanied by either a referral to a dermatologist and/or a prescription for a drug indicated for the treatment of BP (Supplementary Table 2). Overall, 30% (n = 45) of patients diagnosed with BP were either hospitalized, developed sepsis, or died within 6 months of their diagnosis (Supplementary Table 3).

Table 1 presents the characteristics of the entire cohort stratified by antidiabetic drug use at cohort entry. Users of DPP-4 inhibitors were similar to users of other second- to third-line antidiabetic drugs with respect to age, sex, smoking status, BMI, hemoglobin A_1c levels, and history of macrovascular complications and neuropsychiatric disorders. However, they had a longer duration of treated diabetes (7.7 vs. 5.0 years) and were more likely to have a history of microvascular complications (68% vs. 54%). Table 2 presents the characteristics of patients in the propensity score–matched analysis; all covariates were well balanced between the two exposure groups, with standardized differences ranging between 0.00 and 0.05 (study flow diagram presented in Supplementary Fig. 3).

Compared with use of other second- to third-line drugs, use of DPP-4 inhibitors was associated with a doubling of the risk of BP (incidence rates 47.3 vs. 20.0 per 100,000 person-years; HR 2.21 [95% CI 1.45–3.38]) (Table 3). In the duration-response analysis, the HRs gradually increased with longer durations of continuous use, reaching a peak after 20 months of use (HR 3.60 [95% CI 2.11–6.16]) and decreasing thereafter (Fig. 1). Stratification by individual agent yielded the highest HRs for linagliptin (HR 4.90 [95% CI 2.68–8.96]) and vildagliptin (HR 4.56 [95% CI 1.42–14.64]), although the results on vildagliptin were based on few exposed events. The HRs for saxagliptin and sitagliptin were also elevated, but the CI included the null value (HR 2.16 [95% CI 0.86–5.46] and HR 1.42 [95% CI 0.79–2.53], respectively) (Supplementary Table 4).

The results of the sensitivity analyses are summarized in Supplementary Fig. 4 and presented in detail in Supplementary Tables 5–12. Overall, these analyses produced results that were consistent with those of the primary analysis, except for the marginal structural Cox proportional hazards model that generated a higher point estimate (marginal HR 3.01 [95% CI 1.82–4.97]). Based on a post hoc analysis, these findings are unlikely to be driven by unmeasured confounding under most plausible exposure-confounder and confounder-outcome associations (Supplementary Table 13).

The propensity score–matched analysis yielded results consistent with those of the time-varying analysis, with use of DPP-4 inhibitors being associated with a doubling of the risk of BP, compared with use of other second- to third-line antidiabetic drugs (incidence rates 51.9 vs. 21.7 per 100,000 person-years; HR 2.40 [95% CI 1.13–4.66]) (Table 3). The cumulative incidence curves appeared to diverge after 18 months of use (Supplementary Fig. 5).

The results of this large population-based study indicate that use of DPP-4 inhibitors is associated with an at least doubling of the risk of BP, when compared with use of other second- to third-line drugs. The highest point estimates were observed with linagliptin and vildagliptin, and the association was highest after 20 months of use. Overall, the sensitivity and the propensity score–matched analyses yielded results that were highly consistent with those of the primary analysis.

To date, several case reports and pharmacovigilance analyses have suggested a link between DPP-4 inhibitors and BP (4,5). A potential signal was also observed in the Cardiovascular and Renal Microvascular Outcome Study With Linagliptin in Patients With Type 2 Diabetes Mellitus (CARMELINA) trial, where a numerical imbalance of BP events was reported (linagliptin vs. placebo: seven vs. zero events) (29). To our knowledge, only five observational studies (all case-control designs) have been conducted to assess the association between use of DPP-4 inhibitors and BP; these reported either a trend toward an increased risk (odds ratio 2.48 [95% CI 0.75–8.30]) (7) or statistically significantly increased risks (odds ratios ranging between 1.58 and 3.16) (8–11). However, these studies had significant methodological shortcomings, such as potential control selection bias (7,9), potential time-window bias (7–11), and important confounding (7–11), complicating the interpretation of their findings. Moreover, all five studies were underpowered to assess possible duration-response relations (7–11).

Overall, although the results of our study corroborate previous findings (7–11), they provide further insight on the association between DPP-4 inhibitors and BP. Indeed, in three of the previous observational studies (8–11), vildagliptin was strongly associated with BP. This drug differs from other DPP-4 inhibitors in that it has a relatively lower selectivity for the DPP-4 enzyme in comparison with other members of the DPP family, such as DPP-8 and DPP-9 (30). Thus, it has been hypothesized that off-target DPP-8/DPP-9 inhibition could be involved in the pathophysiology of DPP-4 inhibitor–related BP (11,31). However, although we also observed a strong association with vildagliptin, we found an association of equal magnitude with linagliptin, which is highly selective for DPP-4 versus DPP-8/DPP-9 (30). Similar findings were recently reported by Kridin and Bergman (10). Taken together, these results contradict the “off-target” hypothesis. Moreover, although we cannot exclude associations with sitagliptin or saxagliptin, their lower point estimates, as compared with vildagliptin or linagliptin, argue against a homogenous class effect, although more mechanistic evidence is needed in this regard.

Case series and pharmacovigilance analyses have been inconclusive regarding the timing of DPP-4–related BP (5); assumptions about the possible latency period varied greatly, ranging between a few days (5) and 48 months (7) after treatment initiation. Our propensity score–matched analysis indicated a divergence of the Kaplan-Meier curves after 18 months, suggesting a delayed onset of BP. Moreover, our duration-response analysis demonstrated a peak in the association after 20 months of use. Thus, our results provide a clearer pattern of the timing of this adverse event.

This study has several strengths. First, our study design excluded prevalent users, thus eliminating biases resulting from their inclusion (32). Second, the time-varying exposure definition allowed patients to contribute to multiple exposure categories during follow-up, thereby eliminating immortal time bias (33). Finally, although type 2 diabetes is unlikely to be independently associated with BP (17), we nonetheless aggressively adjusted for proxies of diabetes severity, including duration of treated disease, hemoglobin A_1c level, and history of diabetic micro- and macrovascular complications.

Our study has some limitations. First, as with any observational study, residual confounding remains possible (e.g., there are many care reports linking several drug classes with BP). Reassuringly, the use of a marginal structural model that adjusted for potential time-dependent confounders, as well as a propensity score–matched cohort analysis yielded results that were consistent with those of the primary analysis. Moreover, based on a post hoc analysis, our findings are unlikely to be explained by unmeasured or unknown confounding under most plausible exposure-confounder and confounder-outcome associations. Second, prescriptions in the CPRD represent those issued by primary care physicians, and thus misclassification of exposure is possible if patients were also treated by specialists. However, such exposure misclassification is unlikely to have been differential between the exposure groups, as prescribing of DPP-4 inhibitors is not restricted to specialists in the U.K. (34). Finally, to our knowledge, diagnostic codes of BP have not been validated in the CPRD, and the database does not record histologic findings. Thus, some outcome misclassification is possible. However, a sensitivity analysis using a stricter outcome definition requiring BP diagnoses to be accompanied by either a prescription of a drug indicated for the treatment of this condition and/or a referral to a dermatologist led to similar results. Moreover, the incidence rate of BP in our study is comparable with incidence rates reported previously (35).

In summary, the results of this large population-based cohort study indicate that use of DPP-4 inhibitors is associated with an at least doubling of the risk of BP in patients with type 2 diabetes. Although the absolute risk is low, BP is a potentially fatal condition (3), and thus physicians should be aware of this association.

Funding. This study was funded by a Foundation Scheme Grant from the Canadian Institutes of Health Research (FDN-333744). O.H.Y.Y. holds a Chercheur-Clinicien Junior 1 award from the Fonds de Recherche du Québec-Santé (FRQS). K.B.F. holds a Chercheur-Boursier Junior 2 award from the FRQS. L.A. holds a Chercheur-Boursier Senior award from the FRQS. Both K.B.F. and L.A. are recipients of the William Dawson Scholar award from McGill University.

The sponsor 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. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.D. drafted the manuscript. A.D., J.R., H.Y., O.H.Y.Y., K.B.F., and L.A. conceived and designed the study. A.D., J.R., H.Y., O.H.Y.Y., K.B.F., and L.A. analyzed and interpreted data. A.D., J.R., H.Y., O.H.Y.Y., K.B.F., and L.A. critically revised the manuscript for important intellectual content. A.D., H.Y., and L.A. performed the statistical analysis. L.A. acquired data, obtained funding, and supervised the study. 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.