GVP Module IX: Evidence, Causality and Clinical Context

Introduction

A validated signal is a safety question, not a conclusion. The next task is to determine what the available evidence actually supports and how much uncertainty remains.

That assessment cannot be reduced to a single causality score or a single statistical result. Signal evaluation requires the integration of individual cases, aggregate patterns, clinical knowledge, epidemiological evidence, biological plausibility and alternative explanations. The relative importance of each source depends on the question being asked and on the limitations of the data.

This is why evidence assessment follows signal validation. Validation establishes that further investigation is justified; scientific assessment determines what can reasonably be concluded from that investigation.

1. What Causality Means in Signal Management

Causality is the relationship between an exposure and an event. In pharmacovigilance, the question is whether a medicinal product may have contributed to the observed event and, at aggregate level, whether the evidence supports an association that is sufficiently credible to affect the product's safety profile.

These are related but different questions.

For an individual case, the reviewer may assess chronology, alternative causes, dechallenge, rechallenge and clinical plausibility. For a signal, the assessment extends across cases and other evidence sources and asks whether the overall pattern supports a product-event association.

An individual case therefore contributes evidence to a signal but does not, by itself, determine the aggregate conclusion.

2. Evidence Has Different Functions

Different evidence sources answer different questions.

A well-documented case may provide information about clinical phenotype and temporal relationship. A case series may reveal a recurring pattern. A clinical trial may provide controlled comparative information. An epidemiological study may estimate relative or absolute risk. Experimental or mechanistic evidence may support biological plausibility.

The objective is not to assign every source a universal rank. It is to understand what each source can establish, what it cannot establish and how its limitations affect the overall conclusion.

A useful assessment therefore asks:

• What question does this evidence address?
• How reliable is the underlying information?
• What biases or confounding could affect it?
• Is the finding consistent with other evidence?
• Does it change the uncertainty surrounding the signal?

3. Individual Case Evidence

Individual cases are often the starting point for a safety hypothesis because they contain clinical detail that aggregate data can obscure.

Useful information can include the patient's clinical history, indication, dose, timing of exposure, latency, concomitant medicines, relevant investigations, outcome, dechallenge and rechallenge, and alternative explanations.

Case quality matters. A serious event with a clear chronology and compatible clinical course may provide more informative evidence than a larger number of poorly characterised reports.

Case evidence should therefore be assessed for information content rather than treated as a simple count.

4. Temporal Relationship

Temporal association is usually necessary for a causal hypothesis to be credible, but it is rarely sufficient on its own.

The relevant question is whether the timing of exposure and event is compatible with the proposed biological mechanism and natural history. Some reactions occur rapidly after exposure; others require prolonged exposure or have a delayed latency.

The absence of an expected temporal relationship can weaken a hypothesis. A compatible temporal relationship increases plausibility but does not distinguish causality from other explanations that can produce the same timing.

Temporal reasoning therefore needs to be integrated with clinical course, background risk and other evidence.

5. Dechallenge and Rechallenge

Information about what happens after treatment is withdrawn or restarted can provide useful evidence in individual cases.

Improvement after dechallenge may support a causal hypothesis when the clinical course is otherwise compatible. Recurrence after rechallenge can provide stronger evidence in some circumstances, although intentional rechallenge may be inappropriate for serious reactions and should never be treated as a routine evidence-generation strategy.

The interpretation depends on the natural history of the event, concomitant treatment and other possible explanations. A negative dechallenge does not necessarily exclude causality, and an apparent positive dechallenge does not automatically establish it.

These observations are therefore evidence within a broader clinical assessment.

6. Alternative Explanations

A central purpose of causal assessment is to determine whether the event could reasonably be explained without the medicinal product.

Possible alternatives include the underlying disease, indication, age, comorbidities, concomitant medicines, infections, environmental exposures, diagnostic procedures and background incidence.

The relevant question is not whether an alternative explanation can be imagined. It is whether the alternative is sufficiently credible to account for the observed pattern and how its credibility compares with the product-related hypothesis.

This prevents both premature attribution and premature dismissal.

7. Clinical Phenotype

A signal becomes easier to interpret when the clinical phenotype is defined precisely.

The phenotype may include symptoms, signs, laboratory findings, imaging, diagnostic criteria, severity, outcome and relevant timing. Broad event terms can conceal clinically important differences between cases.

For example, several reports using the same preferred term may represent different underlying clinical conditions. Conversely, related manifestations may be coded differently despite representing the same clinical phenomenon.

Clinical review can therefore refine the signal before epidemiological or statistical assessment proceeds.

8. Biological Plausibility

A biologically plausible mechanism can strengthen a causal hypothesis, particularly when it is consistent with pharmacology and the observed clinical pattern.

However, biological plausibility should not be used as a substitute for empirical evidence. A theoretically possible mechanism does not establish that the medicinal product caused the event in patients.

The strongest assessment usually considers mechanistic evidence alongside clinical and epidemiological observations rather than allowing one attractive mechanism to determine the conclusion.

9. Dose and Exposure Relationships

A relationship between dose or exposure and event frequency can provide useful evidence when exposure is measured adequately.

A consistent relationship may strengthen a causal hypothesis, while its absence may weaken one. Neither is decisive in every situation because some adverse reactions are not dose-dependent and exposure measurements can be incomplete or affected by treatment decisions.

The analysis should therefore explain the exposure measure used and its limitations rather than presenting an apparent relationship without context.

10. Latency and Time-to-Onset

The time between exposure and event can be one of the most informative clinical features of a signal.

A plausible latency can support a hypothesis when it is consistent across cases and with the proposed mechanism. An implausible latency may weaken it.

Time-to-onset should nevertheless be interpreted carefully. Immortal-time effects, incomplete exposure histories, delayed diagnosis and changes in treatment can distort apparent patterns.

Temporal analysis is therefore most useful when the exposure history is sufficiently reliable to support it.

11. Epidemiological Evidence

Epidemiological evidence becomes particularly valuable when the question requires comparison with an appropriate background or comparator population.

Spontaneous reports can identify a potential signal but generally cannot provide a simple incidence rate because the denominator of exposed patients is not fully known. A comparative epidemiological study may help determine whether the event occurs more frequently than expected and whether the association persists after accounting for important confounding.

The design must match the question. A study that is poorly suited to the causal question may produce a precise estimate that is nevertheless misleading.

12. Confounding

Confounding occurs when another factor is associated with both the exposure and the outcome and can therefore create or distort an apparent association.

Indication is a common example. A medicinal product may be prescribed to patients with a condition that itself increases the risk of the event being investigated.

Other potential confounders include age, disease severity, comorbidity, concomitant medicines, healthcare utilisation and calendar-time effects.

A credible signal assessment should identify the major plausible confounders and explain how the available evidence addresses them.

13. Clinical Trials and Controlled Evidence

Clinical trials can provide valuable comparative evidence because treatment allocation, data collection and follow-up are generally more controlled than in spontaneous reporting.

Their limitations must also be considered. Sample size may be insufficient for rare events, trial populations may differ from routine clinical practice, treatment duration may be limited and exclusion criteria may reduce generalisability.

The absence of an imbalance in a trial therefore does not automatically refute a signal arising from post-authorisation experience. It must be interpreted in light of the event frequency, exposure and population studied.

14. Literature and External Evidence

Published literature can provide cases, case series, epidemiological studies, mechanistic evidence and independent scientific discussion.

The quality and relevance of a publication should be assessed rather than treating publication itself as evidence of validity. A case report may contain valuable clinical information but limited control for alternative explanations. An epidemiological study may provide stronger comparative evidence but have important methodological limitations.

External evidence should therefore be integrated according to its evidential contribution, not according to its source prestige alone.

15. Consistency Across Evidence Sources

A causal hypothesis becomes more persuasive when different evidence sources converge on the same explanation.

For example, a consistent clinical phenotype may be observed in cases, supported by a plausible latency, accompanied by an exposure-response pattern and compatible with epidemiological evidence.

The reverse is also informative. If one evidence source suggests an association while others repeatedly fail to reproduce it, the discrepancy should be investigated rather than ignored.

Consistency does not mean that every evidence source must point in exactly the same direction. It means that the totality of evidence can be explained coherently.

16. Strength of Association and Its Limits

A strong statistical association can be important evidence, particularly when supported by appropriate design and sensitivity analyses. It does not, however, automatically establish causality.

The magnitude of an association must be interpreted with its confidence interval, study design, potential bias, exposure definition, outcome definition and residual confounding.

Conversely, a modest association can still be clinically important when the event is serious, preventable or affects a large population.

Scientific assessment therefore should not reduce significance to a single numerical threshold.

17. Rare and Serious Events

Rare events create a particular evidential problem because conventional statistical approaches may have little power when only a small number of cases are expected.

In such circumstances, clinical phenotype, temporal relationship, biological plausibility, case quality and consistency across independent observations may become particularly important.

A small number of highly informative cases can therefore justify substantial investigation. At the same time, seriousness alone should not be allowed to convert weak evidence into a causal conclusion.

The appropriate response is proportionate investigation of the potential consequence and the remaining uncertainty.

18. Common Events and Background Risk

For common outcomes, the existence of many reports may provide little information about causality because the event occurs frequently without exposure to the medicinal product.

The key question may instead be whether the frequency, severity or pattern differs from an appropriate background or comparator.

This is why the same number of reports can have very different evidential significance depending on the underlying event frequency and exposed population.

19. Known Risks and New Aspects

Evidence assessment should not stop when the event is already listed as an adverse reaction.

A known risk may generate a new signal if evidence indicates a different severity, population, dose relationship, latency, interaction, clinical manifestation or other material change.

The relevant question is therefore whether the new evidence changes the understanding or management of the existing risk.

This connects signal assessment directly to the safety specification and RMP rather than treating listedness as the end of the investigation.

20. Weighting the Totality of Evidence

The final scientific assessment should explain how the different evidence sources were integrated.

A useful conceptual model is:

Clinical evidence
       +
Case evidence
       +
Epidemiological evidence
       +
Mechanistic evidence
       +
Exposure / temporal evidence
       +
Alternative explanations
       ↓
Totality of evidence
       ↓
Residual uncertainty
       ↓
Scientific conclusion

The purpose is not to calculate a numerical causality score from heterogeneous evidence. It is to make the reasoning transparent enough that another qualified reviewer can understand why the conclusion follows from the evidence.

21. What a Scientific Conclusion Should Contain

A defensible conclusion should identify what the evidence supports, what it does not establish and what uncertainty remains.

For example, the assessment may conclude that the evidence supports a causal association, is insufficient to determine causality, provides evidence against the proposed association or supports an alternative explanation.

The language should match the strength of the evidence. Strong wording unsupported by the underlying data can create unnecessary regulatory consequences; excessively cautious wording can obscure a genuine safety concern.

22. From Scientific Conclusion to Pharmacovigilance Decision

The scientific conclusion is not necessarily the final pharmacovigilance decision.

A conclusion that an association is plausible may lead to different actions depending on seriousness, preventability, exposure, existing risk-minimisation measures and the remaining uncertainty.

Possible consequences include continued monitoring, additional follow-up, targeted studies, RMP reassessment, risk-minimisation changes, product-information assessment or regulatory action.

The decision should therefore preserve the distinction between what the evidence shows and what the organisation should do about it.

23. Documentation and Independent Review

Significant signal assessments should be sufficiently documented to allow reconstruction of the evidence and reasoning.

The record should identify the question, evidence reviewed, important limitations, alternative explanations, clinical assessment, analytical methods where relevant, conclusion, uncertainty and resulting decision.

For important or complex signals, independent review can provide an additional control against confirmation bias and premature closure. The reviewer should have appropriate expertise and sufficient independence from the original assessment to provide meaningful challenge.

24. Illustrative Inspection Scenario: A Causality Score Becomes the Conclusion

A signal assessment contains several individual case causality categories and concludes that the overall association is established because most cases were assessed as probable or likely.

The potential weakness is that case-level causality assessments have been aggregated as though they were independent proof of a product-level causal relationship. The organisation should instead demonstrate how the complete evidence base was evaluated, including alternative explanations and non-case evidence.

The issue is therefore not the use of causality assessment. It is treating a case-level tool as a substitute for aggregate scientific reasoning.

25. Illustrative Inspection Scenario: A Statistical Association Overrides Clinical Evidence

A disproportionality measure exceeds the organisation's alert threshold, and the resulting assessment describes the association as confirmed without examining the clinical phenotype or confounding.

A statistical association is an observation that requires interpretation. The organisation should be able to explain what the result means in the context of exposure, reporting behaviour, indication, alternative explanations and other evidence.

26. Illustrative Inspection Scenario: The Evidence Is Listed but Not Weighed

An assessment contains pages of case reports, literature references and epidemiological results but never explains which evidence is considered persuasive or why conflicting findings do not change the conclusion.

The problem is not insufficient data. It is insufficient reasoning. A scientific assessment should make the weighting of the evidence visible.

27. A Mature Evidence-to-Decision Model

A mature signal assessment can be represented as:

Validated signal
       ↓
Define the causal question
       ↓
Characterise the clinical phenotype
       ↓
Assess individual cases
       ↓
Review aggregate and comparative evidence
       ↓
Examine temporal / exposure relationships
       ↓
Assess biological plausibility
       ↓
Identify confounding and alternative explanations
       ↓
Integrate the totality of evidence
       ↓
State residual uncertainty
       ↓
Scientific conclusion
       ↓
Pharmacovigilance decision
       ↓
Action / monitoring / documented no-action

This model keeps scientific reasoning separate from the operational decision that follows it while preserving the connection between them.

28. Final Review Questions

Before an important signal assessment is finalised, the organisation should be able to answer:

1. What causal question was investigated?
2. What clinical phenotype is being assessed?
3. What are the strongest individual cases?
4. What does aggregate evidence show?
5. What does comparative or epidemiological evidence show?
6. Is the timing biologically and clinically plausible?
7. Is there evidence of a dose or exposure relationship?
8. What alternative explanations are credible?
9. What confounding could materially affect the conclusion?
10. What evidence supports biological plausibility?
11. Where do evidence sources agree?
12. Where do they disagree, and why?
13. What uncertainty remains?
14. What does the totality of evidence support?
15. What does it not establish?
16. What pharmacovigilance consequence follows from the conclusion?
17. Can another qualified reviewer reconstruct the reasoning?

These questions are not a replacement for applicable GVP or regulatory procedures. They provide a practical test of whether the scientific assessment is sufficiently reasoned, proportionate and traceable.

Key Takeaways

Signal assessment is an evidence-integration exercise. No single case, statistical measure, mechanistic theory or study result should automatically determine the conclusion.

Causality is strengthened when independent evidence sources converge and weakened when credible alternative explanations account for the observed pattern. Clinical context determines what the evidence means and what additional evidence is needed.

The strongest assessment makes uncertainty explicit. It distinguishes the scientific conclusion from the subsequent pharmacovigilance or regulatory decision and preserves enough reasoning for the conclusion to be independently understood.

References

1. European Medicines Agency. Good Pharmacovigilance Practices (GVP), Module IX — Signal Management.
2. European Medicines Agency. Good Pharmacovigilance Practices (GVP), Module VI — Collection, management and submission of reports of suspected adverse reactions to medicinal products.
3. European Medicines Agency. Good Pharmacovigilance Practices (GVP), Module VII — Periodic safety update report.
4. European Medicines Agency. Good Pharmacovigilance Practices (GVP), Module V — Risk management systems.
5. Regulation (EC) No 726/2004, as amended.
6. Directive 2001/83/EC, as amended.
7. Commission Implementing Regulation (EU) No 520/2012, as amended.

Regulatory Note

This article explains evidence integration, causality and clinical context within the EU signal-management framework. It distinguishes scientific assessment from regulatory decision-making and recommended operational practice. Current legislation, GVP guidance and EMA implementation material should be verified when applying the framework to a specific medicinal product or safety signal.

Illustrative inspection scenarios are hypothetical and are not presented as documented regulatory findings.

29. Evidence Is Interpreted in Context

The evidential value of an observation depends on the question being asked. A case report can be highly informative about phenotype and chronology while providing little information about comparative risk. An epidemiological study can estimate an association while remaining vulnerable to residual confounding. A clinical-trial analysis can provide a useful comparator while having limited power for rare events.

The assessment should therefore avoid treating evidence as interchangeable units. The appropriate question is what each source contributes to the causal hypothesis and how its limitations affect the conclusion.

30. Evidence That Appears to Conflict

Different evidence streams may point in different directions. A spontaneous-reporting pattern may suggest an association while an epidemiological study finds little evidence of increased risk. Alternatively, a controlled study may show an association that is difficult to recognise in spontaneous reports because the event is uncommon.

Conflict should trigger examination rather than selective evidence use. Differences in population, outcome definition, exposure, latency, data quality, study design and statistical power may explain the apparent disagreement.

A strong assessment makes those differences explicit and explains how they affect the weight assigned to each finding.

31. The Role of Negative Evidence

Evidence that does not support a causal association can be as important as positive evidence.

Repeatedly appropriate studies failing to reproduce an association may reduce confidence in a signal. However, a negative result must be interpreted according to the study's ability to detect the effect in question. A study with insufficient exposure or inadequate power may provide little evidence either way.

The conclusion should therefore distinguish absence of evidence from evidence supporting absence.

32. Strength of Evidence and Strength of Action Are Different

The strength of the scientific evidence and the urgency of the resulting pharmacovigilance response are related but not identical.

A serious, preventable event may warrant rapid action while the evidence is still developing if the potential consequences of waiting are substantial. Conversely, a strong association involving a clinically minor and adequately managed event may not require an immediate change in risk minimisation.

This distinction preserves scientific integrity while allowing proportionate protection of patients.

33. Clinical Context Determines the Meaning of Risk

Risk cannot be interpreted independently of the population in which it occurs.

The same event may have different significance in patients with different underlying diseases, ages, comorbidities or treatment indications. A small relative increase may have substantial public-health consequences when the exposed population is large, while a large relative increase in a very small population may have a different practical impact.

Clinical context therefore informs both scientific interpretation and subsequent risk-management decisions.

34. Vulnerable Populations

Signals involving pregnancy, children, older people, patients with renal or hepatic impairment or other potentially vulnerable groups may require particular attention because baseline risks, exposure patterns and evidence availability can differ from the general treated population.

A signal that appears weak in the overall population may become more informative after appropriate stratification. Conversely, an apparent subgroup association may arise from small numbers or selective reporting.

The assessment should therefore be proportionate to the quality and plausibility of the subgroup finding rather than assuming that any subgroup difference is meaningful.

35. Clinical Review and Statistical Analysis Should Inform Each Other

Clinical review and quantitative analysis are most useful when they are connected.

Clinical review can refine the event definition, identify relevant subgroups and reveal patterns that suggest an analytical question. Quantitative analysis can test whether the observed clinical pattern is unusual, estimate its magnitude or identify characteristics associated with the event.

Neither should be treated as a subordinate validation step for the other. They address different aspects of the same safety question.

36. When Further Evidence Is Needed

A scientific assessment should identify what information would materially reduce the remaining uncertainty.

Possible evidence-generation approaches include targeted follow-up, additional case review, literature monitoring, epidemiological studies, clinical studies, registry analysis or other appropriate data sources.

The choice should follow from the limitation identified in the current evidence. If the problem is an unclear clinical phenotype, improved case characterisation may be most useful. If the problem is confounding, a comparative epidemiological design may be more informative. If the event is too rare for routine quantitative detection, targeted clinical investigation may be required.

Evidence generation should therefore be hypothesis-driven rather than simply additive.

37. Signal Assessment and the Benefit-Risk Balance

A signal assessment may contribute to the overall benefit-risk evaluation of a medicinal product, but the existence of a signal does not by itself establish that the benefit-risk balance has become unfavourable.

The assessment must consider the strength and uncertainty of the safety evidence alongside the seriousness and frequency of the potential risk, the population exposed, available treatment alternatives, treatment benefits and existing risk-minimisation measures.

This is another reason to preserve the distinction between a safety hypothesis, a scientific conclusion and a regulatory decision.

38. Relationship With Risk Management

When a signal changes the understanding of a risk, the implications for the RMP should be assessed systematically.

The assessment may affect the safety specification, the need for additional pharmacovigilance activities, the choice or intensity of risk-minimisation measures or the need to evaluate their effectiveness.

The important governance point is that the RMP consequence should follow from the assessed evidence. A signal should not automatically trigger a new risk-management measure without considering whether the evidence and clinical significance justify it.

39. Relationship With Aggregate Safety Evaluation

Signal assessment contributes to the broader aggregate evaluation of the medicinal product.

Relevant conclusions may need to be reflected in subsequent PSURs and other periodic or ad hoc safety assessments. Conversely, aggregate analyses can identify patterns that generate new signals.

This means that signal management should not operate as a closed workflow. Information needs to move in both directions between signal management and the wider pharmacovigilance system.

40. Regulatory Assessment May Continue the Scientific Question

When a signal enters a regulatory process, the regulator may ask questions that extend or challenge the MAH's assessment.

The MAH should therefore retain the underlying evidence and reasoning rather than treating the submitted conclusion as the end of the scientific record. Requests for additional analyses, subgroup evaluation, clarification of confounding or further evidence may require reassessment of the original conclusion.

A regulatory question can therefore become a new stage in the evidence-generation cycle.

41. Managing Scientific Disagreement

Reasonable scientific disagreement can occur when evidence is incomplete or competing interpretations are plausible.

The objective of governance is not necessarily to eliminate disagreement. It is to ensure that competing interpretations are considered, documented and resolved through an appropriate decision process.

For a significant signal, the record should make clear what the principal alternatives were, why one interpretation was preferred and whether uncertainty remains.

42. Independent Challenge

Independent scientific challenge can improve the quality of important signal assessments.

The reviewer should have appropriate expertise and sufficient distance from the original assessment to question assumptions, alternative explanations, analytical choices and the strength of the conclusion.

Independent review is particularly useful when the signal could materially affect the benefit-risk balance, regulatory status or major risk-minimisation measures.

43. Illustrative Inspection Scenario: Positive Evidence Is Selected, Negative Evidence Ignored

An assessment cites several supportive case reports and one positive epidemiological analysis but does not discuss two well-conducted studies that found no association.

The potential weakness is not that the organisation reached a particular conclusion. It is that the assessment does not demonstrate consideration of contradictory evidence.

A defensible assessment should explain why the negative studies were considered less informative, if that is the conclusion, including relevant differences in design, population, exposure or statistical power.

44. Illustrative Inspection Scenario: Uncertainty Is Hidden by Definitive Language

An assessment states that a medicinal product "causes" an event even though the evidence consists mainly of spontaneous cases with substantial confounding and no comparative evidence.

The potential weakness is a mismatch between language and evidence. Scientific conclusions should communicate the degree of certainty supported by the data.

45. Illustrative Inspection Scenario: Further Evidence Is Requested Without a Defined Question

A signal remains uncertain and the organisation commissions another analysis without identifying what limitation the new analysis is intended to address.

Additional data do not necessarily resolve uncertainty. The new activity should be linked to a defined information gap and a decision about how its results will change the assessment.

46. A Mature Evidence-Integration Model

The complete assessment can be viewed as:

Safety hypothesis
       ↓
Define causal question
       ↓
Characterise phenotype and population
       ↓
Review case and clinical evidence
       ↓
Assess epidemiological / comparative evidence
       ↓
Assess temporal, exposure and mechanistic evidence
       ↓
Identify confounding and alternative explanations
       ↓
Examine supportive and contradictory evidence
       ↓
Weight the totality of evidence
       ↓
Define residual uncertainty
       ↓
Scientific conclusion
       ↓
Benefit-risk / RMP / regulatory assessment
       ↓
Decision and further evidence, if needed

The model is deliberately iterative. New evidence can return the assessment to an earlier stage rather than simply moving the signal forward through a fixed sequence.

47. Governance Questions for Significant Signals

For a significant or complex signal, governance should be able to establish:

1. Who owns the scientific assessment?
2. Which functions contributed expertise?
3. What evidence was considered?
4. What evidence was excluded and why?
5. How were conflicting findings handled?
6. What assumptions materially affect the conclusion?
7. What uncertainty remains?
8. What further evidence is required?
9. What RMP, PSUR or product-information interfaces were assessed?
10. What regulatory interactions occurred?
11. Who approved the conclusion and resulting action?
12. When will the assessment be revisited?

These questions make the governance process visible without turning governance into a substitute for scientific analysis.

48. Final Principle

The objective of signal assessment is not to remove uncertainty artificially. It is to understand uncertainty well enough to make a proportionate and defensible pharmacovigilance decision.

That requires disciplined integration of evidence, clinical context and scientific judgement. It also requires the organisation to remain open to revision when new evidence changes the balance.

A mature signal-management system therefore treats a conclusion as a controlled scientific position that can be strengthened, weakened or revised as the evidence evolves.

49. From Evidence to a Controlled Decision

Once the evidence has been integrated, the assessment should move deliberately from scientific interpretation to the pharmacovigilance decision.

The scientific conclusion should answer what the evidence supports. The decision should then answer what the organisation needs to do about that conclusion. Keeping these questions separate makes the reasoning easier to review and prevents an operational response from becoming evidence for its own justification.

50. Possible Scientific Outcomes

A signal assessment can lead to several scientifically legitimate outcomes.

The evidence may support a causal association, provide evidence against the proposed association, remain insufficient for a conclusion, or indicate that the observation is better explained by an alternative cause or an already characterised risk.

The appropriate outcome depends on the evidence, not on an expectation that every validated signal must become a new risk.

51. Possible Pharmacovigilance Outcomes

The scientific conclusion may lead to continued routine monitoring, enhanced monitoring, targeted follow-up, additional evidence generation, an RMP reassessment, changes to risk minimisation, product-information assessment or regulatory action.

In other situations, the appropriate outcome may be documented no further action.

No further action is itself a decision. The record should show why the evidence does not justify an additional intervention and what monitoring, if any, remains appropriate.

52. Signal Closure Is Not the Same as Forgetting the Signal

Closing a signal should not mean that all evidence associated with it disappears from the pharmacovigilance system.

The closure record should preserve the scientific question, evidence reviewed, conclusion, rationale and any conditions under which reassessment should occur. Relevant information may also remain available to aggregate safety evaluation and future signal detection.

A signal can therefore be closed as an active assessment while its underlying evidence remains part of the continuing safety record.

53. Reopening a Previously Closed Signal

A previously closed issue may need to be reconsidered when new evidence changes the balance.

Triggers can include new cases with a consistent phenotype, a new study, a change in exposure, a new vulnerable population, evidence from another jurisdiction or a mechanistic finding that changes the plausibility assessment.

The organisation should therefore have a controlled mechanism for linking new evidence to previously assessed signals rather than treating every observation as unrelated.

54. QPPV Oversight

The QPPV does not need to personally perform every statistical or clinical analysis. The QPPV does need appropriate visibility of significant safety issues and confidence that the signal-management system is functioning effectively.

For an important signal, governance should allow the QPPV to understand the safety question, the strength and limitations of the evidence, the scientific conclusion, the remaining uncertainty and the resulting pharmacovigilance or regulatory consequences.

This is an oversight responsibility rather than a requirement that the QPPV replace specialist scientific functions.

55. Escalation

Escalation should be based on the potential significance of the issue and the urgency of the decision required.

A potential fatal or rapidly evolving risk may require rapid escalation even before the evidence is complete. A low-impact issue with substantial uncertainty may be appropriate for routine monitoring and planned reassessment.

The escalation mechanism should therefore connect clinical seriousness, evidence strength, potential public-health impact and regulatory consequences rather than relying solely on numerical thresholds.

56. Inspection Evidence

An inspection-ready signal record should make the evidence-to-decision chain reconstructable:

Detection
   ↓
Validation
   ↓
Analysis
   ↓
Clinical assessment
   ↓
Evidence integration
   ↓
Scientific conclusion
   ↓
PV / regulatory decision
   ↓
Action or no-action
   ↓
Follow-up

The records supporting each transition may exist in different systems. What matters is that they can be linked reliably and that responsibilities, dates, decisions and evidence are clear.

57. Illustrative Inspection Scenario: The Final Decision Has No Scientific Rationale

A signal record states "closed — no action" but contains no explanation of the evidence considered or why further assessment was unnecessary.

The potential weakness is not necessarily the closure decision. It is the inability to demonstrate that the decision resulted from an appropriate scientific assessment.

A concise but reasoned closure record is more useful than a lengthy record that contains data without a clear conclusion.

58. Illustrative Inspection Scenario: The Scientific Assessment Cannot Be Reconstructed

The organisation has the final conclusion but the underlying analytical files, case review and evidence used to reach it are distributed across individual mailboxes and cannot be readily retrieved.

This creates a traceability problem. The organisation should maintain controlled records sufficient to reconstruct the significant elements of the assessment without depending on individual memory.

59. Illustrative Inspection Scenario: New Evidence Does Not Reach the Closed Signal

A later case series contains the same unusual phenotype as a previously closed signal, but the case-processing system has no mechanism for linking the new information to the historical assessment.

The potential weakness is a failure of lifecycle integration. Closure should not prevent new evidence from re-entering the signal-management process when it is relevant.

60. A Mature Clinical and Evidential Governance Model

The complete system can be expressed as:

New information
       ↓
Signal detection
       ↓
Validation
       ↓
Define causal question
       ↓
Clinical + quantitative + epidemiological assessment
       ↓
Alternative explanations / confounding
       ↓
Totality of evidence
       ↓
Scientific conclusion
       ↓
Benefit-risk / RMP / regulatory assessment
       ↓
Action / monitoring / no-action
       ↓
Documented closure
       ↓
Reassessment when new evidence emerges

This is a lifecycle rather than a linear workflow. New evidence can return a closed issue to active assessment, and a regulatory question can require additional scientific analysis.

61. Final Review Questions

A mature signal-management system should be able to demonstrate:

1. A clearly defined safety question.
2. Appropriate clinical characterisation.
3. Proportionate review of individual cases.
4. Appropriate quantitative or epidemiological analysis where relevant.
5. Consideration of biological plausibility.
6. Explicit assessment of confounding and alternatives.
7. Consideration of contradictory evidence.
8. A reasoned statement of residual uncertainty.
9. A scientific conclusion proportionate to the evidence.
10. A separate pharmacovigilance or regulatory decision.
11. Appropriate RMP, PSUR and product-information interfaces.
12. QPPV or governance visibility proportionate to significance.
13. Controlled follow-up and reassessment.
14. Evidence sufficient for independent reconstruction.

These controls demonstrate that the organisation is managing uncertainty rather than merely recording signal status.

Key Takeaways

Evidence, causality and clinical context form the scientific core of signal assessment. The objective is to integrate different evidence sources, understand their limitations, consider credible alternatives and state what the totality of evidence supports.

A mature system does not force every signal into a binary confirmed/refuted outcome. It can retain and govern uncertainty, generate targeted further evidence and revise previous conclusions when new information changes the assessment.

The final measure of quality is therefore not the number of signals closed or confirmed. It is whether the organisation can demonstrate a coherent, scientifically defensible path from safety information to conclusion, action and continuing oversight.

References

1. European Medicines Agency. Good Pharmacovigilance Practices (GVP), Module IX — Signal Management.
2. European Medicines Agency. Good Pharmacovigilance Practices (GVP), Module VI — Collection, management and submission of reports of suspected adverse reactions to medicinal products.
3. European Medicines Agency. Good Pharmacovigilance Practices (GVP), Module V — Risk management systems.
4. European Medicines Agency. Good Pharmacovigilance Practices (GVP), Module VII — Periodic safety update report.
5. Regulation (EC) No 726/2004, as amended.
6. Directive 2001/83/EC, as amended.
7. Commission Implementing Regulation (EU) No 520/2012, as amended.

Regulatory Note

This article explains the scientific assessment of signals within the EU pharmacovigilance framework. It distinguishes evidence assessment, clinical judgement, pharmacovigilance decision-making and regulatory action. Current legislation, GVP guidance and EMA implementation material should be verified when applying the framework operationally.

Inspection scenarios are illustrative unless an authoritative inspection source is specifically identified.