Signal Evaluation: DPP-4 Inhibitors and Bullous Pemphigoid

Bullous pemphigoid associated with dipeptidyl peptidase-4 (DPP-4) inhibitors is an instructive signal because the event is rare, clinically recognisable and concentrated in an older population that already has substantial background risk. The evidence therefore cannot be understood from case counts alone. It has to be built in layers: first define the phenotype, then assess individual reports, then ask whether reporting is disproportionate, and finally test the association in comparative data capable of providing a denominator.

The case also demonstrates why a class signal and a molecule-specific signal are not the same thing. Early European regulatory attention was particularly focused on vildagliptin, whereas subsequent observational and randomised evidence supported concern across DPP-4 inhibition more broadly while continuing to suggest heterogeneity between active substances. The appropriate pharmacovigilance conclusion is therefore more nuanced than either “one drug causes bullous pemphigoid” or “all DPP-4 inhibitors carry identical risk.”

Clinical and Pharmacological Context

DPP-4 inhibitors are glucose-lowering medicines used in type 2 diabetes mellitus. By inhibiting the DPP-4 enzyme, they prolong the activity of endogenous incretin hormones and thereby increase glucose-dependent insulin secretion and reduce glucagon secretion. The same enzyme, also known as CD26, participates in several immunological processes. That broader biology made an immune-mediated adverse reaction plausible, but plausibility alone could not establish causality.

Bullous pemphigoid is an autoimmune subepidermal blistering disease. Autoantibodies are directed principally against proteins of the hemidesmosomal adhesion complex, especially BP180 and BP230. The typical clinical picture includes pruritus, urticarial or eczematous lesions and tense blisters; mucosal disease may occur but is less prominent than in some other autoimmune blistering disorders. Diagnosis commonly combines clinical assessment with histopathology, direct immunofluorescence and, where appropriate, serological testing.

This diagnostic specificity matters for signal evaluation. “Rash,” “blistering” and “bullous pemphigoid” are not interchangeable endpoints. A signal becomes harder to interpret if nonspecific skin reactions, other autoimmune blistering diseases and biopsy-supported bullous pemphigoid are pooled together. Conversely, an excessively restrictive case definition may discard clinically credible cases in which specialist investigations were incomplete.

Why background risk matters

Bullous pemphigoid occurs predominantly in older adults. The population treated with DPP-4 inhibitors likewise contains many older people with diabetes, polypharmacy and comorbid disease. Neurological disorders, age and other clinical characteristics associated with bullous pemphigoid may also be unevenly distributed between treatment groups.

This creates the central counterfactual problem: if bullous pemphigoid occurs after DPP-4 inhibitor exposure, would it have occurred anyway in a similar patient who received another glucose-lowering treatment? Spontaneous reports cannot answer that question because they usually lack an appropriate exposed and comparator denominator. Comparative epidemiology is therefore not merely an optional confirmation step; it addresses a different question from spontaneous reporting.

From Individual Cases to a Regulatory Signal

Individual case reports were important because bullous pemphigoid is sufficiently specific and uncommon that repeated well-documented cases can be informative. The evidential value of a report increases when it contains a secure diagnosis, exposure dates, time to onset, relevant comorbidity, concomitant medicines, treatment changes, dechallenge information and a clinically coherent outcome.

Dechallenge deserves cautious interpretation. Improvement after withdrawal of the suspected medicine can support a drug-event relationship, but bullous pemphigoid is usually treated at the same time with corticosteroids or other anti-inflammatory or immunomodulatory therapy. Improvement therefore cannot automatically be assigned to drug withdrawal. Positive rechallenge would ordinarily be more persuasive, but deliberate re-exposure merely to test causality would generally be inappropriate for a serious blistering disease. The absence of rechallenge is consequently not evidence against the association.

The 2016 vildagliptin signal

EMA's Pharmacovigilance Risk Assessment Committee (PRAC) discussed a new signal of pemphigoid with vildagliptin and vildagliptin/metformin at its July 2016 meeting. The signal arose during routine signal-detection activities after a Japanese regulatory labelling notification and consideration of published material and EudraVigilance cases. PRAC recorded 16 EudraVigilance cases in the initial signal context and noted a strong disproportionality signal for vildagliptin.

That history illustrates what a regulatory signal means. The case series and disproportionality result were sufficient to justify formal analysis and prioritisation; they did not themselves quantify incidence or prove that vildagliptin had a higher biological risk than every other DPP-4 inhibitor. PRAC requested an integrated review including spontaneous and clinical-trial data, literature, time to onset, dechallenge and rechallenge, background incidence, biological plausibility and the possible need to amend product information and risk-management documentation.

The sequence is important:

case recognition → database pattern → regulatory signal → structured cumulative evaluation.

This is a model of how signal management should escalate uncertainty rather than prematurely convert it into certainty.

What Disproportionality Can and Cannot Show

Disproportionality analysis asks whether a drug-event pair is reported more often than expected relative to other reports in the same database. Measures such as the reporting odds ratio, proportional reporting ratio or information component can make an unusual reporting pattern visible within a very large spontaneous-reporting system.

A positive disproportionality result supports prioritisation when it is clinically coherent. It does not measure incidence, absolute risk or causal relative risk. Reporting can be influenced by market share, duration on the market, country-specific use, stimulated reporting, publicity, changes in product information and differences in the types of events reported for comparator products.

The early concentration of concern around vildagliptin therefore had to be separated into two questions:

1. Was vildagliptin–pemphigoid reported disproportionately? The regulatory record shows that it was.
2. Did vildagliptin have a higher true clinical risk than other DPP-4 inhibitors? That required comparative evidence and remained more uncertain.

This distinction prevents a common signal-evaluation error: treating a ranking in a spontaneous-reporting database as if it were a ranking of clinical incidence.

Moving From Reporting Patterns to Comparative Risk

Once the signal had been recognised, the next question was whether patients receiving DPP-4 inhibitors developed bullous pemphigoid more often than sufficiently comparable patients receiving alternative diabetes therapy. That transition matters because a comparative study supplies a denominator and can address some of the background-risk problem that spontaneous reports cannot.

Case-control studies added early support. They compared patients with bullous pemphigoid with controls and repeatedly found greater exposure to DPP-4 inhibitors among cases. Several studies also suggested that the association might be stronger for particular active substances. These findings strengthened the hypothesis, but case-control designs remained vulnerable to selection, exposure misclassification and residual confounding.

A more informative design is an active-comparator new-user cohort. “New user” means that patients are followed from treatment initiation rather than being selected after they have already tolerated treatment for an uncertain period. “Active comparator” means that the comparison group receives another treatment for the same underlying disease. Together, these design choices can reduce survivor bias and some differences in indication and healthcare use.

A large active-comparator study

A large US cohort published in 2020 compared patients initiating DPP-4 inhibitors with patients initiating second-generation sulfonylureas. Across the source populations, 1,664,880 patients were included. After propensity-score matching, bullous pemphigoid occurred at approximately 0.42 per 1,000 person-years among DPP-4 inhibitor initiators and 0.31 per 1,000 person-years among sulfonylurea initiators. The pooled hazard ratio was 1.42 (95% CI 1.17–1.72).

This result contributes something that the spontaneous-report signal could not: both a comparative estimate and an indication of absolute event frequency in the studied population. The event remained uncommon. The approximate incidence difference in that analysis was 0.11 cases per 1,000 person-years, or about 11 per 100,000 person-years. That figure should not be transplanted uncritically to every population, because absolute risk depends on age, case ascertainment, treatment patterns and baseline risk.

The same study reported higher rates in some subgroups, including older patients and linagliptin users. Such subgroup findings are hypothesis-informing rather than automatic proof of biological susceptibility. Multiple subgroup analyses increase the possibility of unstable estimates, and prescribing patterns can differ between molecules.

Randomised evidence

Randomised trials are often thought of as the strongest source of comparative safety evidence, but very rare adverse reactions pose a power problem. No individual diabetes trial was designed to establish the risk of bullous pemphigoid. Pooling many trials can nevertheless be informative.

A 2020 meta-analysis identified 138 eligible DPP-4 inhibitor trials; only six reported at least one pemphigoid event. There were 17 events among DPP-4 inhibitor recipients and one among controls, producing a Mantel-Haenszel odds ratio of 4.44 (95% CI 1.31–15.00). The very wide confidence interval reflects the small event count. A separate linagliptin analysis was also positive, but the authors explicitly cautioned that data were insufficient to define the risk of other individual agents with precision.

This is a good example of why “randomised” does not mean “certain.” Randomisation reduces confounding, but it cannot manufacture information when events are sparse. The trial evidence supported the association while leaving the magnitude and molecule-specific pattern imprecise.

Class Effect Versus Molecule-Specific Risk

A class effect is plausible when medicines share a pharmacological target, when similar events occur across several members of the class and when the biological hypothesis is coherent. But shared pharmacology does not imply identical risk. Molecules differ in selectivity, off-target activity, pharmacokinetics, exposure patterns and the populations in which they are used.

For DPP-4 inhibitors and bullous pemphigoid, the evidence supports awareness at class level while also showing heterogeneity. Early European regulatory assessment focused on vildagliptin. Later comparative studies and randomised evidence supported an association involving other DPP-4 inhibitors, including linagliptin. These observations make two extreme conclusions inappropriate:

• “The signal belongs only to vildagliptin.” Later evidence does not support such a narrow statement.
• “Every DPP-4 inhibitor has the same risk.” The evidence does not establish equal magnitude across molecules.

An experienced signal evaluator therefore maintains two evidence tables: one for the class hypothesis and one for each active substance. A class-level conclusion can coexist with uncertainty about molecule-specific risk.

Mechanistic Evidence and Clinical Phenotype

DPP-4/CD26 participates in immune regulation and cleaves multiple peptide substrates. Experimental and clinical studies have proposed several pathways by which DPP-4 inhibition might alter immune tolerance or antigen processing relevant to BP180. Some reports have also described phenotypic differences between DPP-4 inhibitor-associated and conventional bullous pemphigoid.

These observations are supportive, not decisive. Mechanistic research can make a signal biologically coherent, help explain latency or phenotype and generate hypotheses about molecule specificity. It does not substitute for comparative clinical evidence. A mechanism that is possible in vitro may not contribute materially at therapeutic exposure in humans.

The practical use of mechanism is therefore integrative: it should answer whether the observed clinical pattern is compatible with known or plausible biology, not whether a mechanistic story can be constructed after the event.

Translating the Signal Into Product Information

Regulatory action followed the accumulating evidence. European product information for DPP-4 inhibitors has incorporated bullous pemphigoid warnings or adverse-reaction information at product level following regulatory assessment. For example, EMA's procedural record for sitagliptin documents a 2017 product-information update adding a warning on bullous pemphigoid after PRAC assessment.

A product-information change should not be misread as a numerical statement about risk. Its function may be to enable recognition and appropriate action when a rare but clinically meaningful reaction is suspected. The regulatory decision integrates seriousness, plausibility, available evidence, preventability and the consequence of delayed recognition; it does not require a perfectly precise incidence estimate.

That is particularly relevant to bullous pemphigoid because recognition changes management. Persistent pruritic or blistering lesions require clinical assessment rather than being treated indefinitely as a nonspecific rash. Where bullous pemphigoid is suspected, the relevant product information should guide treatment decisions and the need for discontinuation; the pharmacovigilance professional should not replace product-specific clinical instructions with a generic class rule.

Practical Signal-Evaluation Workflow

The case can be translated into a reusable workflow for rare, clinically specific adverse reactions:

The order is not rigid. New evidence may force the evaluator back to redefine the phenotype or reassess the class hypothesis. Signal management is iterative rather than a one-directional checklist.

Failure Modes in Evaluating This Signal

The DPP-4 inhibitor–bullous pemphigoid history exposes several recurring analytical errors.

The first is phenotype dilution: combining biopsy-supported bullous pemphigoid with nonspecific rash or blistering. This can inflate counts while weakening clinical coherence. The second is denominator substitution: treating the number of spontaneous reports as if it were incidence. The third is disproportionality over-interpretation: assuming that the active substance with the highest reporting statistic necessarily has the highest clinical risk.

A fourth error is class inflation. Evidence involving one molecule can justify examining the class, but it does not automatically prove equal class-wide risk. The opposite error—refusing to consider a class hypothesis because the original signal was molecule-specific—is equally problematic once independent evidence accumulates across agents.

Finally, an evaluator can overvalue dechallenge in a disease that is treated simultaneously. If a suspected medicine is stopped and systemic corticosteroids are started on the same day, improvement is compatible with a drug relationship but cannot be attributed solely to withdrawal.

Inspection and Governance Perspective

An inspector assessing a company's management of a similar signal would be interested less in whether the final conclusion exactly matches a later publication than in whether the process was scientifically defensible and traceable. Illustrative questions include:

• Was the event definition clinically appropriate and applied consistently?
• Were serious and well-documented cases medically reviewed rather than reduced to coded terms?
• Were class and molecule-specific analyses separated?
• Was disproportionality described as a reporting measure rather than a risk estimate?
• Were relevant literature and comparative epidemiological studies incorporated when they became available?
• Could the organisation reconstruct why the signal was validated, prioritised, further assessed or closed?
• Did new evidence trigger appropriate reconsideration of product information, risk management and communication?

These are illustrative inspection questions, not a list of legally mandated documents or predetermined inspection findings. The underlying governance principle is traceability: the evidence, assumptions, uncertainty and decision should be reconstructable from the signal record.

For the QPPV, the oversight question is similarly one of system effectiveness. The QPPV need not personally reproduce every epidemiological analysis, but should be able to understand the important evidence, challenge unsupported certainty, recognise where a product-level conclusion has been generalised to the class without justification, and ensure that material regulatory commitments are implemented and tracked.

Key Takeaways

Bullous pemphigoid with DPP-4 inhibitors shows how a rare, specific phenotype can progress from individual cases to a well-supported safety concern without ever producing perfectly uniform evidence across molecules.

The most important lessons are:

• a clinically specific rare event can generate a meaningful signal from a small number of high-quality cases;
• spontaneous-report disproportionality supports detection and prioritisation but does not measure incidence or comparative risk;
• active-comparator epidemiology adds the denominator and addresses part of the background-risk problem;
• randomised evidence can support a rare-event association while remaining imprecise because event counts are small;
• class-level concern and molecule-specific heterogeneity can coexist;
• regulatory action can be appropriate before the exact mechanism and magnitude of risk are completely resolved.

The case therefore teaches a broader principle of pharmacovigilance: signal evaluation is not a contest to produce the strongest possible causal statement. It is a disciplined process for converting incomplete evidence into the most proportionate and revisable conclusion the evidence can support.

References

1. European Medicines Agency. Minutes of the PRAC meeting, 4–8 July 2016. Signal of pemphigoid with vildagliptin/vildagliptin-metformin, EPITT 18692. EMA; 2016.
2. European Medicines Agency. Januvia: procedural steps taken and scientific information after authorisation. Includes the 2017 worksharing variation adding a bullous pemphigoid warning following PRAC assessment.
3. Lee H, Chung HJ, Pawar A, et al. Evaluation of Risk of Bullous Pemphigoid With Initiation of Dipeptidyl Peptidase-4 Inhibitor vs Second-generation Sulfonylurea. JAMA Dermatology. 2020. PMID: 32697283.
4. Silverii GA, Dicembrini I, Nreu B, Montereggi C, Mannucci E, Monami M. Bullous pemphigoid and dipeptidyl peptidase-4 inhibitors: a meta-analysis of randomized controlled trials. Endocrine. 2020;69(3):504–507. doi:10.1007/s12020-020-02272-x. PMID: 32236820.
5. Kridin K, Bergman R. Association of bullous pemphigoid with dipeptidyl-peptidase 4 inhibitors in patients with diabetes: estimating the risk of the new agents and characterizing the patients. JAMA Dermatology. 2018;154(10):1152–1158.

Regulatory Note

This article distinguishes regulatory signal-management decisions from scientific certainty. EMA/PRAC records are regulatory sources. Observational studies, randomised-trial meta-analyses and mechanistic publications are scientific evidence and do not themselves create EU legal obligations.

Product-information requirements are product- and procedure-specific. Statements about clinical recognition or discontinuation should therefore be checked against the current authorised product information for the relevant medicine and jurisdiction. The operational workflow and inspection questions in this article are explanatory recommended practice, not additional legal requirements.