Signal Evaluation: GLP-1 Receptor Agonists and Acute Pancreatitis

Introduction

Few safety signals illustrate the difficulty of distinguishing a plausible adverse drug reaction from the background risk of a disease population as clearly as the historical association between glucagon-like peptide-1 receptor agonists and acute pancreatitis.

The signal attracted substantial attention because acute pancreatitis is a clinically important event, because cases were reported after exposure to GLP-1 receptor agonists, and because several biological hypotheses appeared capable of explaining a potential relationship.

The pharmacovigilance question, however, was not simply whether pancreatitis occurred after treatment.

Patients receiving GLP-1 receptor agonists often have:

• obesity;
• type 2 diabetes;
• hypertriglyceridaemia;
• gallstone disease;
• alcohol exposure;
• and other established pancreatitis risk factors.

The exposed population therefore has a substantial background risk.

This creates a classic signal-evaluation problem:

Does the observed incidence of pancreatitis exceed what would be expected from the underlying patient population?

The question becomes even more difficult when evidence from different sources does not point in exactly the same direction.

Spontaneous reports can identify a possible signal.

Clinical trials can provide controlled comparative evidence.

Observational studies can examine much larger populations and longer exposure periods.

Mechanistic studies can establish biological plausibility.

Regulatory authorities can integrate all of these sources when deciding whether product information or risk-minimisation measures should change.

The purpose of this case study is not to retrospectively declare that one side of the historical debate was simply correct.

It is to reconstruct how the signal should be evaluated.

1. Define the Signal

The historical signal can be framed as:

GLP-1 receptor agonist exposure — acute pancreatitis.

GLP-1 receptor agonists are medicines that enhance glucose-dependent insulin secretion and have become important treatments for type 2 diabetes and, more recently, obesity.

The class includes medicines with different molecular structures, pharmacokinetic properties and clinical indications.

For signal evaluation, that matters.

A finding involving one active substance should not automatically be treated as evidence that every member of the class carries an identical risk.

The initial signal therefore needs to distinguish:

• individual product evidence;
• class-level evidence;
• indication;
• duration of exposure;
• patient characteristics;
• and outcome definition.

2. Define the Clinical Outcome

Acute pancreatitis is an acute inflammatory condition of the pancreas.

Common causes include:

• gallstones;
• alcohol;
• hypertriglyceridaemia;
• certain medicines;
• and other metabolic, structural or systemic causes.

A robust evaluation therefore requires more than identifying a diagnostic code.

The evaluator should ask:

• How was pancreatitis defined?
• Was the diagnosis clinically confirmed?
• Were cases adjudicated?
• Were alternative causes assessed?
• Were recurrent episodes counted?
• Was chronic pancreatitis distinguished from acute pancreatitis?
• Was the outcome based on hospitalisation, laboratory criteria or coding alone?

Outcome misclassification can materially influence the apparent strength of a safety signal.

3. Why the Signal Was Plausible

Several observations initially made the hypothesis biologically credible.

GLP-1 receptor agonists influence gastrointestinal and pancreatic physiology.

GLP-1 receptors are expressed in pancreatic tissues, and experimental research explored whether GLP-1 signalling could influence pancreatic growth, ductal biology or other processes relevant to inflammation.

At the same time, some patients treated with GLP-1 receptor agonists developed pancreatitis.

This created a plausible hypothesis:

drug exposure → pancreatic effect → inflammation → acute pancreatitis.

But this was only a hypothesis.

The existence of a plausible mechanism does not demonstrate that the mechanism operates at therapeutic exposure in humans.

That distinction became important as more evidence accumulated.

4. Spontaneous Reports

As GLP-1 receptor agonists entered widespread clinical use, cases of pancreatitis were reported spontaneously.

This was important for pharmacovigilance.

A spontaneous report can identify:

• an unexpected event;
• a potential new safety concern;
• a temporal association;
• unusual clinical characteristics;
• and potentially a pattern that was not apparent during pre-authorisation development.

But spontaneous reports cannot directly provide incidence.

A report tells us:

This event occurred in a patient exposed to the medicine.

It does not tell us:

This event occurred more frequently than expected in comparable patients who were not exposed.

That distinction is fundamental.

5. Reporting Proportionality

A further analytical approach is disproportionality analysis.

If pancreatitis is reported disproportionately often with a particular medicine compared with other medicines in a spontaneous-reporting database, that can support a signal.

However, disproportionality is not a measure of relative risk.

It can be influenced by:

• stimulated reporting;
• publicity;
• reporting restrictions;
• differences in patient populations;
• notoriety bias;
• reporting completeness;
• and competing adverse-event patterns.

A positive disproportionality signal therefore means:

further investigation is justified.

It does not mean:

causality has been established.

6. The Background Risk Problem

This is arguably the central issue in the pancreatitis signal.

Patients with type 2 diabetes already have an increased risk of pancreatitis compared with the general population.

They may also have:

• obesity;
• gallstones;
• hypertriglyceridaemia;
• renal disease;
• alcohol exposure;
• and other comorbidities.

Therefore, if 10,000 patients with diabetes receive a GLP-1 receptor agonist and some develop pancreatitis, the event count alone tells us little.

The relevant comparison is:

How many pancreatitis events would have occurred in a sufficiently comparable population without exposure to the medicine?

This is the epidemiological counterfactual.

7. The First Major Question

The signal evaluation therefore needs to move from:

"Were cases reported?"

to:

"Is there an excess of pancreatitis associated with exposure?"

That requires comparative evidence.

Potential approaches include:

• randomised controlled trials;
• cohort studies;
• case-control studies;
• self-controlled designs;
• active-comparator studies;
• and meta-analysis.

Each answers a slightly different question.

8. Clinical Trial Evidence

Randomised trials have an important advantage.

Randomisation tends to balance known and unknown baseline characteristics between treatment groups.

If pancreatitis occurs more frequently in the GLP-1 receptor agonist group, that can provide stronger evidence than an uncontrolled case series.

However, pancreatitis is relatively uncommon.

Individual trials may therefore be underpowered to detect modest differences.

A trial programme can contain thousands or tens of thousands of participants while still producing too few pancreatitis events to estimate a rare risk precisely.

This is why safety evaluation often requires integration across trials.

9. Meta-Analysis of Randomised Trials

A number of meta-analyses examined pancreatitis across GLP-1 receptor agonist trials.

These analyses generally did not demonstrate a clear statistically significant increase in acute pancreatitis.

However, the absence of a statistically significant difference is not identical to proof of no risk.

The correct interpretation depends on:

• number of events;
• confidence interval;
• follow-up duration;
• ascertainment quality;
• heterogeneity;
• and the size of effect that the studies were capable of excluding.

For rare events, confidence intervals can remain wide even after pooling many trials.

10. A Rare-Event Problem

Imagine that the observed data are:

• 7 pancreatitis events among treated participants;
• 4 among controls.

The point estimate might suggest an increased risk.

But the event count is small.

Random variation could explain the difference.

Conversely, if there are:

• 2 events among treated participants;
• 0 among controls,

the absence of a statistical signal does not prove that the true risk is zero.

This is why pharmacovigilance should avoid simplistic interpretations of small event counts.

11. The Importance of Confidence Intervals

A useful question is:

What risks are compatible with the data?

Suppose the estimated relative risk is:

1.2

with a very wide confidence interval extending from:

0.6 to 2.4.

The correct conclusion is not:

No risk exists.

It is:

The study did not demonstrate a statistically significant increase, but clinically important increases cannot be excluded.

That distinction becomes especially important for rare adverse events.

12. Observational Studies

Observational studies can contribute important evidence because they can include much larger populations than individual clinical trials.

They can also examine:

• real-world prescribing;
• longer exposure;
• different patient populations;
• and uncommon outcomes.

But they introduce other problems.

Patients receiving a GLP-1 receptor agonist may differ systematically from patients receiving other glucose-lowering medicines.

Those differences may affect pancreatitis risk.

This is confounding by indication and treatment selection.

13. Active Comparators

An appropriate comparator could be another glucose-lowering treatment used in similar patients.

For example:

GLP-1 receptor agonist versus another antidiabetic therapy.

This is generally more informative than:

GLP-1 receptor agonist versus the entire untreated population.

Why?

Because both groups have:

• diabetes;
• healthcare contact;
• treatment indication;
• and many overlapping risk factors.

The comparator does not eliminate confounding.

But it can reduce some important differences.

14. Confounding by Metabolic Risk

Suppose patients with more severe obesity are preferentially prescribed a GLP-1 receptor agonist.

Obesity is associated with:

• gallstone disease;
• metabolic abnormalities;
• and potentially pancreatitis risk.

Then:

obesity → treatment selection

and:

obesity → pancreatitis risk.

A crude analysis could incorrectly attribute some of the risk to the medicine.

This is a classic confounding pathway.

15. Gallstone Disease

Gallstones are one of the most important background causes of acute pancreatitis.

This is particularly relevant because GLP-1 receptor agonists can affect body weight and biliary physiology.

Rapid weight loss itself can increase the risk of gallstone disease.

Therefore, a pancreatitis event following treatment could potentially arise through several pathways.

For example:

GLP-1 receptor agonist

→ weight loss

→ gallstone formation

→ biliary pancreatitis.

Alternatively:

GLP-1 receptor agonist

→ direct pancreatic effect

→ pancreatitis.

Or:

underlying metabolic disease

→ gallstone / metabolic risk

→ pancreatitis.

These hypotheses are not equivalent.

16. Distinguishing Direct and Indirect Mechanisms

This is an important pharmacovigilance teaching point.

A medicine may be associated with an adverse event without directly damaging the organ in question.

The pathway could be indirect.

For pancreatitis, plausible pathways may include:

• direct pancreatic effects;
• biliary effects;
• metabolic changes;
• weight-loss-associated gallstone disease;
• or other patient-level changes.

The evidence should therefore not be reduced to:

"the drug is toxic to the pancreas."

That statement requires evidence that may not exist.

17. What Is Established About GLP-1 Biology?

GLP-1 receptor signalling has established physiological effects on:

• insulin secretion;
• glucagon regulation;
• gastric emptying;
• appetite;
• and energy balance.

The extent to which therapeutic GLP-1 receptor agonism directly produces clinically important pancreatic injury is much less certain.

Experimental studies have produced differing findings depending on:

• model;
• dose;
• duration;
• compound;
• and experimental conditions.

Therefore:

physiological activity is established.

But:

a direct human pancreatic toxicity mechanism is not established simply from receptor biology.

18. Regulatory Attention

The pancreatitis signal attracted regulatory attention relatively early in the development and post-marketing history of GLP-1 receptor agonists.

Regulators considered:

• spontaneous reports;
• clinical trial data;
• epidemiological studies;
• biological plausibility;
• and the underlying background risk.

Product information for relevant GLP-1 receptor agonists included warnings concerning acute pancreatitis.

The precise wording and regulatory history have varied by product and jurisdiction.

That distinction matters.

A class-level article should not imply that every product had an identical label history.

19. The European Perspective

European regulators have evaluated pancreatitis as a potential safety concern associated with GLP-1 receptor agonists and incorporated appropriate warnings into product information where justified.

For historical signal evaluation, the important lesson is not merely that a warning existed.

It is:

what evidence was available when the regulatory decision was made?

This prevents hindsight bias.

A historical signal should be reconstructed according to the evidence available at the time rather than judged only through the lens of later studies.

20. Hindsight Bias

Suppose a later meta-analysis finds no statistically significant association.

It would be incorrect to conclude:

Therefore, the original pharmacovigilance concern was unreasonable.

The original concern may have been entirely appropriate.

Signal detection exists precisely because evidence is incomplete.

A safety system should be capable of escalating a credible concern before definitive epidemiological certainty exists.

The question is whether the action taken was proportionate to:

• the evidence;
• the severity;
• the alternatives;
• and the uncertainty.

21. The Regulatory Meaning of a Warning

A warning does not necessarily mean:

"This medicine causes pancreatitis."

It may instead mean:

"A potential risk has been identified that warrants clinical vigilance."

This distinction is important when interpreting SmPC language.

A warning may function as:

• risk communication;
• clinical monitoring;
• a trigger for diagnostic evaluation;
• or a risk-minimisation measure.

It should not automatically be interpreted as a definitive causal statement.

22. Later Evidence

As exposure to GLP-1 receptor agonists increased, larger epidemiological datasets became available.

The resulting evidence was mixed.

Some studies reported an association.

Others did not.

Some meta-analyses suggested a small possible increase.

Others concluded that there was insufficient evidence to demonstrate a causal relationship.

This inconsistency is itself informative.

It suggests that:

• the underlying effect, if present, may be small;
• confounding may materially affect estimates;
• product differences may exist;
• outcome ascertainment may differ;
• or the original signal may have been partly generated by background risk.

23. Product-Specific Versus Class-Level Evidence

A major methodological error would be:

Drug A was associated with pancreatitis, therefore all GLP-1 receptor agonists cause pancreatitis.

That inference is too strong.

Class effects require evidence across:

• different active substances;
• different populations;
• different study designs;
• and ideally biologically coherent mechanisms.

Conversely, a negative study for one product does not necessarily exclude a risk for another.

Signal evaluation should therefore preserve the distinction between:

class hypothesis

and:

product-specific evidence.

24. The Role of Exposure Duration

A causal relationship should have a plausible temporal pattern.

Relevant questions include:

• Does risk appear shortly after treatment initiation?
• Does risk increase with cumulative exposure?
• Does risk persist after discontinuation?
• Is there evidence of recurrence on rechallenge?
• Does risk decline after treatment withdrawal?

For rare events, these questions may be difficult to answer reliably.

But the absence of a coherent temporal relationship can weaken a causal hypothesis.

25. Dechallenge and Rechallenge

Individual cases can provide particularly useful information when the temporal relationship is clear.

For example:

drug started → pancreatitis → drug stopped → recovery

supports a temporal relationship.

But pancreatitis often resolves naturally.

Therefore, recovery after discontinuation does not establish causality.

Rechallenge can be more informative.

If pancreatitis recurs after re-exposure, that may strengthen causal suspicion.

However, intentional rechallenge would generally be inappropriate for a serious suspected reaction simply to establish causality.

26. Alternative Causes in Individual Cases

For a patient who develops pancreatitis during treatment, the assessment should consider:

• gallstones;
• alcohol;
• triglycerides;
• hypercalcaemia;
• structural abnormalities;
• infection;
• concomitant medicines;
• prior pancreatitis;
• and underlying disease.

A case report that simply states:

pancreatitis occurred after GLP-1 receptor agonist initiation

is incomplete.

The strength of the case depends heavily on how well alternative causes were investigated.

27. Case Narrative Quality

A high-quality pharmacovigilance case narrative should describe:

• indication;
• dose;
• start date;
• latency;
• clinical symptoms;
• laboratory findings;
• imaging;
• pancreatic enzyme results;
• relevant medical history;
• gallstone status;
• alcohol exposure;
• triglyceride concentration;
• concomitant medication;
• treatment;
• dechallenge;
• outcome;
• and any rechallenge.

This is not administrative detail.

It determines whether the case contributes meaningful evidence to the signal.

28. What Would Strengthen the Signal?

Evidence supporting causality would become stronger if studies demonstrated:

• consistent increased risk across different populations;
• appropriate active comparators;
• a coherent temporal relationship;
• dose or exposure-response;
• recurrence on inadvertent rechallenge;
• reduced risk after withdrawal;
• biologically coherent mechanisms;
• and consistency across clinical trials and observational data.

Evidence weakening causality would include:

• no association in large well-controlled studies;
• strong association with baseline risk factors;
• disappearance of the association after appropriate adjustment;
• inconsistent findings across active comparators;
• and lack of a plausible exposure-response pattern.

29. What Does a Negative Study Mean?

A negative study should be interpreted according to its design.

A study can fail to find an association because:

1. no association exists;
2. the true effect is very small;
3. the study lacks statistical power;
4. exposure is misclassified;
5. outcomes are incompletely captured;
6. residual confounding obscures the effect.

Therefore:

negative ≠ disproven.

Likewise:

positive ≠ proven.

The strength of evidence depends on the totality of the design and results.

30. A Signal Evaluation Matrix

31. A Hypothetical Individual Case

Consider a 58-year-old patient with type 2 diabetes and obesity.

The patient begins a GLP-1 receptor agonist.

Six months later the patient develops severe epigastric pain.

Lipase is markedly elevated.

CT confirms acute pancreatitis.

Gallstones are identified.

The patient drinks little alcohol and has no major hypertriglyceridaemia.

How should the case be interpreted?

Evidence supporting the drug

• temporal association;
• known historical signal;
• plausible biological relationship.

Evidence supporting an alternative cause

• gallstones;
• an established cause of acute pancreatitis.

The appropriate conclusion would therefore not be:

GLP-1 receptor agonist-induced pancreatitis.

A more defensible conclusion would recognise that:

The event occurred during exposure and is temporally compatible with the historical signal, but gallstone disease provides a strong alternative explanation and substantially weakens attribution to the medicine.

This illustrates why individual case assessment cannot be separated from clinical diagnosis.

32. A Different Hypothetical Case

Now consider a patient with:

• no gallstones;
• no alcohol exposure;
• normal triglycerides;
• no hypercalcaemia;
• no structural pancreatic disease;
• no relevant concomitant medicines.

The patient develops pancreatitis shortly after treatment initiation.

The medicine is discontinued.

The patient recovers.

Later, inadvertent re-exposure occurs and pancreatitis recurs.

This case would provide substantially stronger evidence of individual causality.

Even then, it would remain one case.

Individual cases generate evidence.

They do not establish population incidence.

33. Signal Detection Versus Signal Evaluation

This distinction is particularly important.

Signal detection asks:

Is there something unusual that warrants investigation?

Signal evaluation asks:

Does the totality of evidence support a causal relationship, and what action is justified?

A spontaneous report may be sufficient for detection.

It is rarely sufficient for definitive evaluation.

This is why a mature signal-management system moves through:

detection → validation → prioritisation → analysis → regulatory/clinical action → follow-up.

34. What Would a QPPV Need to Know?

A QPPV reviewing the pancreatitis signal would need visibility of:

• individual case reports;
• seriousness;
• case completeness;
• alternative causes;
• reporting trends;
• exposure estimates;
• clinical trial findings;
• epidemiological studies;
• relevant regulatory assessments;
• literature;
• and product-specific differences.

The QPPV should also understand the limitations of each evidence source.

A dashboard saying:

"Pancreatitis cases increased 30%"

is not enough.

The QPPV needs to know:

• exposure increased by how much?
• reporting changed?
• stimulated reporting occurred?
• the patient population changed?
• diagnosis increased?
• background incidence changed?

The denominator and context matter.

35. Reporting Rate Versus Incidence

This is one of the most important distinctions in spontaneous-reporting analysis.

Suppose reports increase from:

100 to 150 cases.

That appears concerning.

But suppose exposure also increases from:

1 million to 3 million patient-years.

The reporting rate may actually have decreased.

Conversely, reports could remain stable while exposure falls dramatically.

Therefore, raw case counts should never be interpreted as incidence.

This principle applies far beyond pancreatitis.

36. The Importance of Exposure Data

For a quantitative assessment, the evaluator should seek:

• number of exposed patients;
• patient-years;
• treatment duration;
• prescribing volume;
• indication;
• age distribution;
• and other relevant denominators.

Without exposure data, the evaluator cannot reliably distinguish:

more patients treated

from:

more risk per patient treated.

37. Signal Strength Versus Clinical Importance

Even if the relative risk were small, the signal might remain clinically important because pancreatitis can be serious.

Conversely, a statistically significant association does not automatically justify broad treatment restrictions.

The regulator must consider:

• absolute risk;
• severity;
• treatment alternatives;
• therapeutic benefit;
• susceptible populations;
• and feasibility of risk minimisation.

This is benefit-risk evaluation rather than signal detection.

38. What the Historical Record Teaches

The pancreatitis signal demonstrates several important features of pharmacovigilance.

First, a plausible signal can emerge before epidemiological certainty exists.

Second, spontaneous reports can trigger appropriate investigation without proving causality.

Third, background disease can create major confounding.

Fourth, rare events can be difficult to evaluate even in large clinical trials.

Fifth, biological plausibility can strengthen a signal without resolving causality.

Sixth, regulatory warnings may remain appropriate even when the scientific literature remains mixed.

Finally, a historical signal should be evaluated according to the evidence available at each stage rather than reconstructed with hindsight.

39. Overall Signal Assessment

The historical association between GLP-1 receptor agonists and acute pancreatitis represents a credible pharmacovigilance signal that warranted investigation and regulatory attention.

The signal was supported by spontaneous reports, temporal associations and biological hypotheses.

However, the exposed population already carries substantial background risk for pancreatitis.

Major alternative causes—including gallstones, alcohol exposure and metabolic disease—make causal attribution difficult.

Randomised trials and subsequent observational studies have generally not provided a simple, consistent estimate of a large class-wide increase in risk.

The epidemiological evidence is therefore best interpreted in the context of:

• rare-event statistics;
• comparator selection;
• confounding;
• outcome ascertainment;
• and product-specific differences.

The most defensible conclusion is:

GLP-1 receptor agonist-associated acute pancreatitis has been an important historical safety concern, and the signal warranted regulatory evaluation and continued clinical vigilance. However, the available evidence does not justify treating the occurrence of pancreatitis during treatment as proof of drug causality. The underlying population has substantial baseline risk, and alternative causes must be actively assessed. Evidence for a uniform class-wide causal effect has remained uncertain, and interpretation should be based on the individual product, clinical context and totality of evidence.

40. Final Perspective

The pancreatitis signal illustrates a central principle of pharmacovigilance:

The harder the event is to distinguish from background disease, the more important the design of the evaluation becomes.

A spontaneous case may be clinically compelling.

A disproportionality analysis may be statistically striking.

A mechanistic experiment may be persuasive.

A randomised trial may be reassuring.

An observational study may point in the opposite direction.

None should automatically dominate the others.

The task is to understand what question each source of evidence can actually answer.

A mature evaluation therefore asks:

1. What was observed?
2. What was the expected background risk?
3. Who received the medicine?
4. What alternative explanations existed?
5. How was the outcome defined?
6. What did clinical trials show?
7. What did observational studies show?
8. Did comparator choice alter the result?
9. Is the mechanism established, plausible or speculative?
10. What did regulators conclude?
11. What uncertainty remained?
12. Was the resulting risk-management action proportionate?

That is the difference between identifying a signal and evaluating it.

References

1. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP) Module IX — Signal Management.
2. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP) Module V — Risk Management Systems.
3. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP) Module XVI — Risk Minimisation Measures: Selection of Tools and Effectiveness Indicators.
4. U.S. Food and Drug Administration. Information on incretin mimetic drugs and reports of pancreatic toxicity.
5. Egan AG, Blind E, Dunder K, et al. Pancreatic Safety of Incretin-Based Drugs — FDA and EMEA Assessment. New England Journal of Medicine. 2014;370:794–797.
6. Li L, Shen J, Bala MM, et al. Incretin treatment and risk of pancreatitis in patients with type 2 diabetes: systematic review and meta-analysis of randomised and non-randomised studies. BMJ. 2014;348:g2366.
7. Monami M, Dicembrini I, Mannucci E. Glucagon-like peptide-1 receptor agonists and pancreatitis: a meta-analysis of randomized clinical trials. Diabetes Research and Clinical Practice. 2014;103(2):269–275.
8. European Medicines Agency. Relevant product information and regulatory assessment documents for GLP-1 receptor agonists.
9. U.S. Food and Drug Administration. Prescribing information and safety communications for GLP-1 receptor agonists.
10. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP) Module VI — Collection, management and submission of reports of suspected adverse reactions to medicinal products.
11. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP) Module VII — Periodic safety update report.