Historical Signal Evaluation: Pioglitazone and Bladder Cancer

Introduction

The association between pioglitazone and bladder cancer is a particularly useful case study in pharmacovigilance because it demonstrates how difficult causal interpretation can become when the suspected adverse event is relatively uncommon, the underlying disease is itself associated with the event, exposure accumulates over time, and different epidemiological studies produce apparently conflicting results.

Pioglitazone is a thiazolidinedione used to improve glycaemic control in patients with type 2 diabetes mellitus.

The bladder-cancer signal did not emerge from a single decisive experiment.

Instead, concern developed through a combination of:

• observations from preclinical studies;
• clinical-trial data;
• post-marketing experience;
• epidemiological studies;
• analyses examining duration and cumulative exposure;
• and regulatory reassessment as additional evidence accumulated.

This makes the case particularly valuable for signal evaluation.

The central question is not simply:

Does pioglitazone cause bladder cancer?

A rigorous signal evaluation asks several more precise questions:

• Was an association observed?
• How large was the association?
• Was it consistent between studies?
• Was there evidence of a dose-response or duration-response relationship?
• Could diabetes itself or other patient characteristics explain part of the association?
• Was there biological plausibility?
• Did randomised and observational evidence agree?
• How did regulators interpret uncertainty?
• Did the risk-management response change as the evidence evolved?

The answer that emerges is more nuanced than a simple binary causal classification.

The history of pioglitazone demonstrates why pharmacovigilance should distinguish between:

signal detection, epidemiological association, causal inference and regulatory action.

These are related but not identical questions.

1. Define the Signal

The signal evaluated in this case is:

Pioglitazone exposure — bladder cancer.

The concern is clinically important because bladder cancer is a serious disease and because pioglitazone may be used chronically.

The relevant exposure therefore differs from a medicine associated with an acute adverse reaction.

For an acute event, exposure may be assessed in days or weeks.

For a potential malignancy, the evaluator must consider:

• latency;
• cumulative exposure;
• duration of treatment;
• dose;
• previous exposure;
• competing risk factors;
• and the natural history of carcinogenesis.

This immediately makes the signal more difficult to evaluate.

2. Why the Signal Was Biologically Interesting

Pioglitazone is a peroxisome proliferator-activated receptor gamma (PPARγ) agonist.

PPARγ is a nuclear receptor involved in regulation of:

• glucose metabolism;
• adipocyte differentiation;
• lipid metabolism;
• and cellular signalling.

The biological question therefore became whether prolonged PPARγ activation could influence carcinogenesis in the urinary bladder.

The mechanistic evidence is not equivalent to clinical proof.

This distinction is important.

A laboratory finding can establish biological plausibility without establishing that the same pathway causes cancer in treated humans.

Conversely, uncertainty about the precise mechanism does not rule out an epidemiological association.

The signal therefore needs to be assessed across multiple evidence domains.

3. What Was Known From Preclinical Evidence?

Preclinical findings contributed to the early concern.

In animal studies, bladder tumours were observed in male rats exposed to pioglitazone.

The interpretation of this finding was complicated by species-specific mechanisms.

In particular, urinary crystals containing calcium and other components were considered relevant to the rat-specific carcinogenic process.

This raised an important translational question:

Does a carcinogenic finding in one animal species predict human bladder-cancer risk?

The answer cannot be assumed.

Animal carcinogenicity findings can be:

• directly relevant to humans;
• indirectly informative;
• species-specific;
• or mechanistically difficult to translate.

Therefore, the rat findings generated biological concern but did not by themselves establish human causality.

4. Establishing the Baseline Risk

Before assessing the medicine, the evaluator must understand the background risk.

Patients with type 2 diabetes are not a homogeneous population.

They may have:

• older age;
• obesity;
• smoking exposure;
• renal disease;
• cardiovascular disease;
• metabolic abnormalities;
• occupational exposures;
• and multiple medications.

Some of these factors can also influence cancer risk.

This creates a classic pharmacovigilance problem:

The population receiving the drug may differ systematically from the population not receiving it.

This is confounding by indication and confounding by patient characteristics.

If pioglitazone users have a different baseline risk of bladder cancer from non-users, an observed association cannot automatically be attributed to pioglitazone.

5. The Early Clinical-Trial Evidence

The first important clinical question was whether bladder cancer occurred more often in pioglitazone-treated patients in randomised clinical development.

The number of bladder-cancer events was small.

This is important because a rare outcome can produce unstable estimates when the number of events is low.

The clinical-trial evidence therefore had limited ability to answer the question definitively.

However, a signal began to attract attention when bladder-cancer events appeared to be numerically more frequent among pioglitazone-exposed patients in some analyses.

A signal does not require certainty.

It requires enough evidence to justify further evaluation.

6. The PROactive Trial

The PROactive trial was a large randomised cardiovascular outcomes study involving patients with type 2 diabetes and established macrovascular disease.

The trial provided an important source of long-term safety information.

During the trial, bladder cancer occurred more frequently in patients assigned to pioglitazone than placebo, although the difference did not establish statistical significance in the overall analysis.

This was important for two reasons.

First, the trial provided randomised evidence.

Second, the number of events remained small.

The trial therefore contributed to the signal without resolving it.

This is a recurring pattern in pharmacovigilance:

Randomisation reduces confounding, but rare outcomes can still leave substantial uncertainty.

7. Why Rare Cancer Signals Are Difficult

Cancer is particularly challenging for signal evaluation because the exposure-to-event relationship may involve long latency.

A study that follows patients for only a few years may not capture the full effect of a carcinogenic exposure.

Conversely, if a study observes an association after long exposure, the evaluator must consider whether:

• cumulative exposure is responsible;
• duration is acting as a marker of underlying disease severity;
• patients receiving long-term treatment differ systematically from short-term users;
• or the observed association reflects another risk factor.

This means that:

duration-response is potentially informative but not automatically causal.

8. The Early Epidemiological Signal

As observational data accumulated, several studies examined bladder-cancer risk among pioglitazone users.

Some studies reported an increased risk.

Others did not.

This produced a familiar signal-evaluation problem:

How should apparently inconsistent epidemiological evidence be interpreted?

The evaluator should not simply count positive and negative studies.

Instead, the studies need to be compared according to:

• population;
• comparator;
• exposure definition;
• duration of follow-up;
• dose;
• cumulative exposure;
• cancer ascertainment;
• adjustment for smoking;
• adjustment for diabetes severity;
• age and sex;
• latency;
• and statistical methodology.

A study reporting a hazard ratio of 1.3 is not directly comparable with another study reporting a hazard ratio of 1.3 if the populations, exposure definitions and adjustment strategies differ substantially.

9. The FDA Observational Study

One of the most influential studies was a large retrospective cohort study conducted using data from the Kaiser Permanente Northern California population.

The study followed patients with diabetes and examined bladder-cancer risk in relation to pioglitazone exposure.

The investigators did not initially find a statistically significant overall increase in bladder cancer.

However, analyses suggested that risk could increase with longer exposure.

This distinction became central.

The question changed from:

Is pioglitazone associated with bladder cancer?

to:

Is there a relationship between cumulative exposure or duration of pioglitazone treatment and bladder cancer?

That is a much more informative question.

10. Duration and Cumulative Dose

A duration-response relationship can strengthen a causal hypothesis.

If risk increases progressively with longer exposure, that pattern may be consistent with a biological process requiring cumulative exposure.

However, duration is not independent of patient characteristics.

Patients who remain on a medicine for many years may differ from those who discontinue it early.

They may have:

• more severe diabetes;
• longer disease duration;
• greater comorbidity;
• more healthcare contact;
• different prescribing patterns;
• and different survival characteristics.

Therefore:

A duration-response relationship is supportive evidence, but it is not proof of causality.

The evaluator must ask whether duration is a true exposure variable or a proxy for something else.

11. Confounding by Diabetes Severity

Diabetes severity is particularly important.

Patients with more difficult-to-control diabetes may require multiple therapies.

Pioglitazone may therefore be preferentially prescribed to patients whose disease is more advanced or more difficult to manage.

Those patients may differ in baseline cancer risk from patients with milder disease.

If a study does not adequately adjust for disease severity, part of the observed association may be due to the underlying population rather than the medicine.

This is why active-comparator studies can be particularly useful.

Comparing pioglitazone users with patients receiving another glucose-lowering therapy can reduce some forms of confounding associated with simply having diabetes.

It does not eliminate all confounding.

12. Smoking as a Confounder

Smoking is one of the strongest established risk factors for bladder cancer.

Therefore, failure to control adequately for smoking can materially distort the association.

This presents a methodological challenge because smoking information may be incomplete in large administrative databases.

A study with detailed smoking data may therefore have an advantage over a very large database study in which smoking status is poorly captured.

This illustrates a broader principle:

Larger datasets are not automatically better datasets.

For rare outcomes, statistical power matters.

But measurement of important confounders matters too.

13. Surveillance Bias

Another potential explanation is surveillance.

Patients taking pioglitazone may have frequent healthcare interactions because of diabetes.

Greater healthcare contact can increase the probability that symptoms are investigated and cancer is diagnosed.

This can produce differences in observed cancer incidence even if the underlying biological risk is similar.

However, surveillance bias should not be invoked automatically.

The evaluator should examine whether:

• healthcare utilisation differs between groups;
• diagnostic intensity differs;
• haematuria is investigated differently;
• urological referrals differ;
• cancer ascertainment methods are comparable.

A plausible bias is not necessarily an actual bias.

14. The Role of Haematuria

Bladder cancer may present with haematuria.

If treatment or monitoring changes the probability that haematuria is investigated, the apparent incidence of bladder cancer could change.

This is another reason to distinguish:

• occurrence of cancer;
• detection of cancer.

A signal evaluator should ask whether the medicine could alter diagnostic behaviour.

This is particularly relevant for observational studies.

15. Regulatory Review in Europe

The European regulatory assessment evolved as evidence accumulated.

The European Medicines Agency reviewed the emerging evidence and considered the association between pioglitazone and bladder cancer.

The assessment considered:

• clinical-trial evidence;
• observational epidemiology;
• preclinical data;
• potential risk factors;
• duration and cumulative exposure;
• and the overall benefit-risk balance.

The European assessment did not treat the epidemiological association as an isolated statistical finding.

Instead, it considered the totality of evidence.

This is exactly how a mature signal evaluation should operate.

16. Regulatory Risk Minimisation

The regulatory response did not simply become:

Pioglitazone causes bladder cancer and must never be used.

Instead, risk minimisation focused on patient selection and contraindications or warnings relevant to bladder-cancer risk.

The European product information was amended to include warnings concerning:

• active bladder cancer;
• history of bladder cancer;
• unexplained haematuria;
• and consideration of risk factors before treatment.

This is an important pharmacovigilance concept.

When causality is uncertain but a serious potential risk is plausible, regulators may manage the risk through targeted restrictions rather than waiting for absolute scientific certainty.

17. Why Uncertainty Does Not Prevent Action

A common misconception is that regulatory action requires proof beyond reasonable scientific doubt.

Pharmacovigilance does not work that way.

The regulatory question is often:

Is there enough evidence of a clinically meaningful potential risk that the benefit-risk balance can be improved through risk minimisation?

That is different from:

Has every scientific question about causality been resolved?

In the pioglitazone case, the evidence was sufficiently concerning to justify precautions while continued use remained possible in appropriate patients.

This is a classic example of decision-making under uncertainty.

18. The French Study and Subsequent Concern

A large French cohort study published in 2011 reported an association between pioglitazone exposure and bladder cancer.

The study examined more than one million patients with diabetes.

It reported:

• an increased risk associated with pioglitazone;
• higher risk with cumulative exposure;
• and higher risk with longer duration.

The study attracted considerable attention because of its size and because it provided a potential exposure-response relationship.

However, the study also illustrates why large observational studies require careful interpretation.

Even very large studies remain vulnerable to:

• residual confounding;
• exposure misclassification;
• detection bias;
• missing variables;
• and differences in healthcare utilisation.

A large hazard ratio is not automatically a causal hazard ratio.

19. Evidence From Meta-Analyses

Meta-analyses subsequently combined evidence from multiple studies.

The results were not completely uniform.

Some meta-analyses found a modest increased risk.

Others concluded that the evidence was insufficient to establish a clear association.

This heterogeneity is itself informative.

When studies disagree, the evaluator should investigate why.

Possible explanations include:

• different exposure durations;
• different cumulative-dose categories;
• different comparators;
• differences in smoking adjustment;
• differences in cancer ascertainment;
• differences in follow-up;
• differences in population characteristics;
• and random variation.

The correct response to heterogeneity is not to average everything blindly.

It is to understand the sources of heterogeneity.

20. Dose-Response Evidence

Dose-response relationships can be particularly useful in pharmacovigilance.

If bladder-cancer risk increases with cumulative pioglitazone exposure, that pattern is compatible with causality.

However, cumulative dose and duration are closely related.

A patient treated for a longer period generally receives more cumulative exposure.

Therefore, dose-response and duration-response can be difficult to separate.

A rigorous analysis should ask:

• Is cumulative dose independently associated with risk?
• Is duration independently associated with risk?
• Is there a threshold?
• Does risk increase continuously or only at high exposure?
• Is the relationship biologically plausible?
• Does it remain after adjustment for disease duration and severity?

Without these analyses, a simple cumulative-dose relationship should not be overinterpreted.

21. Latency

Carcinogenesis generally does not behave like an acute adverse reaction.

The time between exposure and clinically detectable cancer can be long.

Therefore, the evaluator must distinguish between:

• short-term exposure;
• intermediate exposure;
• prolonged exposure;
• and exposure occurring many years before diagnosis.

If a study defines exposure too narrowly, it may miss relevant risk.

If exposure is defined too broadly, it may include patients whose previous exposure is no longer biologically relevant.

There is no universal correct exposure window.

The appropriate window should be justified using:

• biological plausibility;
• pharmacology;
• disease natural history;
• and the study design.

22. What Does the Preclinical Mechanism Tell Us?

The animal findings provide a biological warning but require careful translation.

The proposed mechanism involving urinary changes and crystal formation in rats illustrates a broader problem in pharmacovigilance:

A mechanistic signal can be real without being directly translatable across species.

Human bladder carcinogenesis may involve different pathways.

Therefore, the correct interpretation is:

Established

Pioglitazone activates PPARγ.

Established in animals

Certain experimental exposure conditions were associated with bladder tumours in male rats.

Supported but not definitive

There are biological mechanisms through which chronic pioglitazone exposure could plausibly influence bladder carcinogenesis.

Not established

That the precise rat mechanism is responsible for human bladder cancer associated with pioglitazone.

This distinction prevents mechanistic overstatement.

23. The Importance of Negative Studies

Several large observational studies did not demonstrate a statistically significant increase in bladder cancer.

These studies should not be treated as inconvenient exceptions.

They are part of the evidence.

A strong signal evaluation should ask:

What would we expect to see if the drug did not cause bladder cancer?

If large well-designed studies repeatedly show no association, the causal hypothesis should weaken.

If positive studies cluster around certain exposure patterns while negative studies have shorter follow-up, the apparent disagreement may become more understandable.

This is why evidence mapping is more useful than simply counting studies.

24. An Evidence Matrix

The evidence can be summarised as follows.

25. How Causality Should Be Expressed

The language used in the final assessment matters.

A weak conclusion would be:

Pioglitazone causes bladder cancer.

That statement is too categorical for an evidence base characterised by observational heterogeneity and residual uncertainty.

Another weak conclusion would be:

There is no evidence of a risk because some studies were negative.

That ignores the positive studies, exposure-response observations and regulatory concern.

A more defensible conclusion is:

The available evidence supports a possible or modest association between pioglitazone exposure and bladder cancer, particularly with longer or greater cumulative exposure, although the magnitude of any causal effect remains uncertain because of heterogeneity between studies and potential residual confounding.

That language communicates both the signal and the uncertainty.

26. The Regulatory Threshold Is Different From the Scientific Threshold

This case demonstrates an important distinction.

A regulator may conclude:

The potential risk is sufficiently credible and clinically important to justify warnings and restrictions.

That does not necessarily mean:

The causal effect has been quantified with high certainty.

The two statements can coexist.

This is fundamental to pharmacovigilance.

Regulatory risk management is often performed under uncertainty.

27. Product Information as a Record of Signal Evolution

Product information can be viewed as a historical record of how the interpretation of a safety concern evolved.

Early information may have contained limited or no bladder-cancer warnings.

As evidence accumulated, warnings were strengthened.

The final language focused on clinically actionable circumstances such as:

• existing bladder cancer;
• previous bladder cancer;
• unexplained haematuria;
• and consideration of individual risk factors.

This illustrates an important principle:

Product information does not need to reproduce the entire scientific debate.

Its purpose is to communicate the clinically relevant consequence of the regulatory assessment.

The underlying signal evaluation may contain substantial uncertainty.

The final warning may nevertheless be precise.

28. Why "Association" and "Causation" Must Remain Separate

This case is particularly useful for teaching epidemiology.

Suppose a study finds:

HR 1.3

That means the observed hazard is approximately 30% higher in the exposed group under the assumptions of the model.

It does not mean:

Pioglitazone increases bladder cancer by 30%.

The latter is a causal interpretation.

To make that interpretation, the evaluator must consider:

• confounding;
• bias;
• exposure misclassification;
• selection;
• measurement;
• temporal sequence;
• biological plausibility;
• consistency;
• dose-response;
• and alternative explanations.

This distinction should be maintained throughout a signal assessment.

29. A Worked Example

Consider two hypothetical patients with type 2 diabetes.

Patient A

A 68-year-old patient has:

• longstanding diabetes;
• heavy previous smoking;
• multiple cardiovascular comorbidities;
• pioglitazone exposure for several years;
• haematuria;
• and subsequently diagnosed bladder cancer.

Patient B

A 48-year-old patient has:

• recently diagnosed diabetes;
• no smoking history;
• short-term pioglitazone exposure;
• no haematuria;
• and no bladder cancer.

It would be inappropriate to compare these two patients directly and conclude that pioglitazone caused the cancer in Patient A.

Patient A has multiple independent risk factors.

The correct pharmacovigilance questions are:

• Was exposure temporally plausible?
• Was cumulative exposure substantial?
• What was the baseline bladder-cancer risk?
• What alternative causes were present?
• Was there previous haematuria?
• Were there other medications or exposures?
• Does the case resemble the broader signal?

Individual case assessment and population-level epidemiology must inform each other, but they answer different questions.

30. What a Strong Individual Case Assessment Would Avoid

A good evaluator should avoid:

"The patient was exposed, therefore the drug caused the cancer."

Exposure alone is insufficient.

"The patient smoked, therefore the drug was unrelated."

A competing risk factor does not exclude a contribution from the medicine.

"The cancer occurred after years of exposure, therefore latency proves causality."

Temporal sequence is necessary but not sufficient.

"The study found a hazard ratio above 1, therefore the drug causes cancer."

An association estimate requires causal interpretation.

"The mechanism is plausible, therefore the epidemiology is unnecessary."

Mechanism cannot substitute for human evidence.

31. Benefit-Risk Considerations

Pioglitazone provides glucose-lowering efficacy and has specific metabolic effects that can be clinically useful.

Its use must nevertheless be considered in the context of multiple safety issues, including:

• fluid retention;
• heart-failure risk in susceptible patients;
• fractures;
• and the bladder-cancer concern.

The appropriate clinical decision is therefore not determined by the bladder-cancer signal alone.

This is another important pharmacovigilance lesson:

A safety signal is one component of benefit-risk assessment, not the entire assessment.

The regulatory response therefore focused on identifying patients in whom treatment should be avoided and situations requiring caution.

32. What This Signal Teaches About Risk Minimisation

Risk minimisation can operate at several levels.

Patient selection

Avoid treatment in patients with relevant bladder-cancer risk.

Clinical assessment

Investigate unexplained haematuria appropriately.

Prescriber awareness

Make the potential association visible at the point of prescribing.

Product information

Provide warnings and contraindications.

Ongoing monitoring

Continue surveillance as additional epidemiological evidence emerges.

This is preferable to treating a complex signal as an all-or-nothing decision.

33. Why This Case Is Different From an Acute Safety Signal

Compare pioglitazone–bladder cancer with SGLT2 inhibitor–DKA.

This comparison illustrates why the same signal-management framework cannot simply be applied mechanically to every adverse event.

The evidence strategy must fit the biology of the event.

34. A Practical Signal-Evaluation Framework

For a potential drug-associated malignancy, the evaluator should work through the following sequence.

Step 1 — Define the cancer phenotype

Specify:

• cancer type;
• histology where relevant;
• anatomical site;
• latency;
• stage;
• diagnostic criteria.

Step 2 — Define exposure

Specify:

• current exposure;
• previous exposure;
• cumulative dose;
• treatment duration;
• latency after discontinuation.

Step 3 — Establish baseline risk

Consider:

• age;
• sex;
• smoking;
• occupational exposure;
• disease characteristics;
• previous cancer;
• family history where relevant.

Step 4 — Review the randomised evidence

Ask:

• how many cases occurred?
• what was the follow-up?
• were the populations representative?
• was the outcome prospectively captured?

Step 5 — Review observational evidence

Compare:

• study design;
• comparator;
• adjustment;
• exposure definition;
• follow-up;
• cancer ascertainment.

Step 6 — Examine dose and duration

Look for:

• exposure-response;
• duration-response;
• threshold effects;
• consistency.

Step 7 — Examine biological plausibility

Separate:

• established mechanisms;
• plausible mechanisms;
• unresolved mechanisms.

Step 8 — Assess alternative explanations

Explicitly identify:

• confounding;
• bias;
• reverse causality;
• surveillance;
• selection.

Step 9 — Assess regulatory significance

Ask:

• Is the signal serious?
• Is the magnitude clinically meaningful?
• Can the risk be reduced?
• Does the benefit-risk balance remain favourable?

Step 10 — State uncertainty explicitly

Do not force a binary conclusion when the evidence does not support one.

35. What the Historical Record Shows

The pioglitazone signal illustrates a gradual evolution.

Stage 1 — Preclinical concern

Animal bladder tumours raised a biological question.

Stage 2 — Clinical observation

Bladder-cancer events appeared during clinical development.

Stage 3 — Epidemiological investigation

Large observational datasets produced mixed but concerning findings.

Stage 4 — Exposure-response analysis

Longer duration and higher cumulative exposure became important analytical questions.

Stage 5 — Regulatory assessment

European regulators considered the totality of evidence and introduced targeted restrictions and warnings.

Stage 6 — Continued evaluation

Subsequent studies continued to examine whether the association represented causation, residual confounding or a combination of factors.

This is how a pharmacovigilance signal should evolve.

The conclusion at one time point should not be treated as immutable.

36. What a QPPV Should Take From This Case

36.1 A cancer signal requires a different analytical approach

Latency, cumulative exposure and background cancer risk matter.

36.2 Large observational studies are powerful but imperfect

Large numbers improve precision but do not eliminate bias.

36.3 Confounding must be actively investigated

It should not simply be mentioned as a generic limitation.

36.4 Duration-response is useful but not definitive

Longer treatment can mean greater exposure, but it can also identify a different patient population.

36.5 Smoking adjustment can materially change interpretation

A major confounder should be treated as a central analytical issue, not a footnote.

36.6 Negative studies are evidence

They should be incorporated into the causal assessment rather than ignored.

36.7 Mechanistic plausibility is supportive

It should not be converted into proof.

36.8 Regulatory action can be proportionate to uncertainty

Warnings and restrictions can be appropriate even when the precise causal effect remains uncertain.

36.9 Product information captures the actionable conclusion

It does not need to reproduce the entire epidemiological debate.

36.10 Historical interpretation matters

The evidence available at the time of a regulatory decision should be distinguished from evidence published later.

37. Overall Assessment

The historical evidence surrounding pioglitazone and bladder cancer supports a clinically relevant safety signal.

The signal is strengthened by:

• animal carcinogenicity findings;
• observations from clinical development;
• positive findings from several epidemiological studies;
• reported relationships with cumulative exposure and treatment duration;
• and regulatory assessments that considered the potential risk sufficiently important to warrant restrictions and warnings.

However, the causal interpretation is complicated by:

• the background association between diabetes and cancer;
• smoking and other patient-level risk factors;
• differences between study populations;
• residual confounding;
• exposure misclassification;
• surveillance differences;
• and heterogeneity between epidemiological studies.

The most defensible conclusion is therefore not that every bladder cancer occurring during pioglitazone treatment was caused by pioglitazone.

Nor is it that the signal was unsupported.

A balanced conclusion is:

The accumulated evidence supports a possible and clinically relevant association between pioglitazone exposure and bladder cancer, with some studies suggesting greater risk with longer or greater cumulative exposure. However, the magnitude and certainty of the causal effect remain difficult to establish because of heterogeneity between studies and the potential for residual confounding, particularly from smoking, diabetes severity and other patient characteristics. The regulatory response therefore appropriately focused on targeted risk minimisation and patient selection rather than treating the epidemiological signal as proof of a uniform causal effect in all exposed patients.

38. Lessons for Signal Management

Lesson 1: Match the method to the biology

A possible carcinogenicity signal cannot be evaluated like an acute adverse reaction.

Lesson 2: Establish background risk first

The observed rate means little without understanding the baseline population risk.

Lesson 3: Define latency explicitly

Cancer signals require an exposure window that is biologically defensible.

Lesson 4: Treat duration and cumulative dose carefully

They can support causality but can also introduce or reflect confounding.

Lesson 5: Identify major confounders before analysing the result

For bladder cancer, smoking is a particularly important example.

Lesson 6: Do not equate statistical significance with causality

A statistically significant association still requires causal interpretation.

Lesson 7: Do not equate statistical non-significance with absence of risk

Rare outcomes can produce wide confidence intervals.

Lesson 8: Examine why studies disagree

Heterogeneity can reveal the mechanism of uncertainty.

Lesson 9: Use active comparators where appropriate

They can reduce some forms of confounding associated with the underlying disease.

Lesson 10: Separate biological plausibility from proof

Animal and mechanistic data should inform the assessment without dominating it.

Lesson 11: Regulatory action can precede complete scientific resolution

Risk management often operates under uncertainty.

Lesson 12: State uncertainty as part of the conclusion

Uncertainty is not a failure of the assessment.

It is an important result of the assessment.

Conclusion

The pioglitazone–bladder cancer signal is a useful demonstration of what difficult pharmacovigilance looks like in practice.

The concern did not emerge from one definitive study.

It developed through multiple evidence streams that were individually imperfect but collectively important.

The signal was biologically plausible.

Some clinical and epidemiological evidence supported an association.

Other studies did not.

Longer duration and cumulative exposure were important findings in several analyses but were also vulnerable to confounding.

Smoking and underlying diabetes characteristics complicated interpretation.

Regulators therefore had to make a decision under uncertainty.

The resulting approach was neither complete reassurance nor complete withdrawal.

Instead, risk minimisation focused on avoiding treatment in patients with relevant bladder-cancer concerns and increasing awareness of the potential risk.

This is an important lesson for signal evaluators.

The purpose of signal evaluation is not to manufacture certainty.

It is to determine, as rigorously as possible:

• what the evidence shows;
• what it does not show;
• what alternative explanations remain;
• how much uncertainty remains;
• and what action is proportionate to the evidence and the consequences of being wrong.

For a potential drug-associated malignancy, the quality of the assessment often depends less on finding a single decisive study than on understanding why the entire evidence base looks the way it does.

That is the essence of historical signal evaluation.

References

1. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP), Module IX — Signal Management.

2. European Medicines Agency. PRAC recommendations on safety signals.

3. European Medicines Agency. Pioglitazone-containing medicines — referral and safety information.

4. European Medicines Agency. Assessment report concerning the pharmacovigilance signal for bladder cancer associated with pioglitazone.

5. European Medicines Agency. Product information for pioglitazone-containing medicinal products.

6. U.S. Food and Drug Administration. FDA Drug Safety Communication: Updated drug labels for pioglitazone-containing medicines.

7. Dormandy JA, Charbonnel B, Eckland DJA, et al. Secondary prevention of macrovascular events in patients with type 2 diabetes in the PROactive Study: a randomised controlled trial. The Lancet. 2005;366:1279–1289.

8. Lewis JD, Ferrara A, Peng T, et al. Risk of bladder cancer among diabetic patients treated with pioglitazone: interim report of a longitudinal cohort study. Diabetes Care. 2011;34:916–922.

9. Neumann A, Weill A, Ricordeau P, Fagot JP, Alla F, Allemand H. Pioglitazone and risk of bladder cancer among diabetic patients in France: a cohort study. Diabetologia. 2012;55:1953–1962.

10. Ferwana M, Firwana B, Hasan R, et al. Pioglitazone and risk of bladder cancer: a meta-analysis of randomized clinical trials and observational studies. Diabetic Medicine. 2013;30:1026–1032.

11. Colmers IN, Bowker SL, Majumdar SR, Johnson JA. Use of thiazolidinediones and the risk of bladder cancer among people with type 2 diabetes: a meta-analysis. CMAJ. 2012;184:E675–E683.

12. Tuccori M, Filion KB, Yin H, et al. Pioglitazone use and risk of bladder cancer: population based cohort study. BMJ. 2016;352:i1541.

13. Bosetti C, Rosato V, Buniato D, et al. Cancer risk for patients using thiazolidinediones for type 2 diabetes: a meta-analysis. The Oncologist. 2013;18:148–156.

14. Levin D, Bell S, Sund R, et al. Pioglitazone and bladder cancer risk: a systematic review and meta-analysis. Diabetologia. 2015.

15. European Commission. Commission Implementing Regulation (EU) No 520/2012 on the performance of pharmacovigilance activities.

16. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP), Module V — Risk management systems.

17. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP), Module XVI — Risk minimisation measures: selection of tools and effectiveness indicators.