Causality Assessment in Pharmacovigilance

Table of contents

• Introduction
• What causality assessment can and cannot do
• Evidence in an individual case
• Temporal relationship
• Clinical phenotype and alternative causes
• Dechallenge and rechallenge
• Dose, exposure and biological plausibility
• From case assessment to aggregate inference
• Case series
• Epidemiology and trials
• Signal-level causality
• Methods and structured tools
• Clinical judgement
• Algorithms and categories
• Counterfactual reasoning
• Communicating uncertainty
• Governance and QPPV use
• Individual case records
• Aggregate and regulatory decisions
• Worked case
• References
• Regulatory Note

Introduction

Causality assessment asks whether a medicinal product contributed to an observed medical event. It is used in individual case review, signal management, aggregate reporting, risk management and benefit-risk evaluation.

It is rarely a proof test. Pharmacovigilance evidence is usually incomplete and observational. The conclusion is a reasoned judgement that integrates chronology, clinical pattern, alternative explanations, pharmacology, dechallenge or rechallenge, class evidence and the quality of the information available.

The right question is often not “did the product cause the event with certainty?” but “how much does the available evidence change the plausibility of a causal contribution, and what action is proportionate?”

What causality assessment can and cannot do

Causality assessment can improve consistency, identify cases needing follow-up, support case-series review and make a safety conclusion transparent. It cannot recover missing facts, remove confounding, establish an incidence rate from spontaneous reports or prove a mechanism from a single narrative.

An individual case conclusion and an aggregate causal conclusion are different. “Related” in a case-processing workflow may reflect a company or regulatory convention for coding and medical review; it does not mean that causation has been proven for the population.

Evidence in an individual case

Temporal relationship

Exposure must precede the event in a biologically plausible way. Consider start date, dose changes, latency, duration, dechallenge and the natural history of the disease. A plausible chronology supports causality, but timing alone is not sufficient.

Clinical phenotype and alternative causes

Ask whether the event is a recognised reaction, whether the clinical pattern is compatible, and whether the diagnosis is secure. Review disease, infection, trauma, comorbidity, concomitant medicines, underlying risk factors and competing explanations.

An alternative cause does not automatically exclude a product contribution. A product may worsen an existing condition or interact with another factor.

Dechallenge and rechallenge

Improvement after withdrawal can support causality, but may reflect natural recovery, treatment of the event or withdrawal of another cause. Lack of improvement may be uninformative when the event is irreversible or the product has a long half-life.

Rechallenge can be informative but is often confounded and may be unethical. Do not encourage rechallenge solely to prove causality.

Dose, exposure and biological plausibility

A dose-response pattern, known pharmacology, class effect or mechanism can strengthen a hypothesis. Absence of a known mechanism does not exclude a reaction, especially for rare or newly recognised events.

From case assessment to aggregate inference

Case series

A case series becomes more informative when cases share a specific phenotype, timing, exposure pattern or outcome and alternative causes are considered. Review completeness, duplicate status, reporting source, geographical spread and ascertainment.

Do not treat the number of cases as a risk estimate. Case counts are influenced by exposure, reporting behaviour, publicity, stimulated reporting and access to diagnosis.

Epidemiology and trials

Clinical trials help compare event rates under controlled conditions but may be underpowered for rare or delayed effects. Observational studies can examine larger populations but require careful consideration of confounding, exposure definition, outcome misclassification, comparator choice and bias.

Causal inference is strengthened when different designs converge, not when one method is declared universally superior.

Signal-level causality

At signal level, the question is whether the totality of evidence supports a new causal association or a new aspect of a known association that warrants action. Evaluate consistency, specificity, temporal patterns, dose or duration, dechallenge, biological plausibility, class information and competing explanations.

Methods and structured tools

Clinical judgement

Expert medical review is essential for complex cases and rare events. The reasoning should be documented rather than replaced by an unexplained label.

Algorithms and categories

Structured algorithms can promote completeness and consistency. They are aids, not substitutes for judgement. A score can be sensitive to missing dates and to how questions are interpreted; it should not be treated as a probability of causation unless validated for that purpose.

Categories such as unlikely, possible, probable or very likely should be defined locally. They should not be presented as interchangeable across companies, regulators or datasets.

Counterfactual reasoning

Counterfactual thinking asks: what would most likely have happened to this patient without the exposure? The answer is not directly observable, but the question makes confounding visible and discourages the assumption that an event after exposure was caused by exposure.

Communicating uncertainty

A defensible conclusion states the evidence and its limits. For example: “The timing and phenotype are compatible, but the patient has a strong alternative cause and follow-up is incomplete; a contribution cannot be excluded.” This is more useful than an unsupported definitive label.

Record what additional information could change the conclusion: diagnostic confirmation, laboratory results, exposure dates, outcome, dechallenge, concomitant treatment or medical history.

Governance and QPPV use

Individual case records

The case record should preserve source information, medical assessment, follow-up requests, causality rationale, coding decisions and any change over time. Quality control should check internal consistency and avoid converting missing information into negative evidence.

Aggregate and regulatory decisions

Causality reasoning should be consistent across case series, signals, PSUR/PBRER and risk-management documents. The QPPV should have visibility of significant changes in causal interpretation, especially where they affect a safety specification, product information, study, communication or benefit-risk conclusion.

Worked case

A patient develops a liver abnormality three weeks after starting a medicine while also taking another hepatotoxic product and experiencing an infection. The chronology supports a possible contribution. Laboratory confirmation, previous liver tests, viral testing, dose history, course after withdrawal and response to treatment would change the assessment.

The correct output is not “the medicine caused the injury” based on timing alone. It is a staged assessment with explicit alternatives and a plan to obtain the missing evidence.

References

• EMA, GVP Module VI: Collection, management and submission of reports
• EMA, GVP Module IX: Signal management
• WHO, The importance of pharmacovigilance
• CIOMS, Practical aspects of signal detection in pharmacovigilance
• ICH E2A, Definitions and standards for expedited reporting

Regulatory Note

Causality labels, algorithms and follow-up requirements vary by jurisdiction, procedure and data source. This article describes scientific reasoning and governance principles; apply the current applicable guidance and the organisation’s approved case-processing procedure.