Causality Assessment in Pharmacovigilance

Introduction

Causality assessment is one of the most important and challenging activities performed in pharmacovigilance. Every major pharmacovigilance process—including individual case review, signal management, aggregate reporting, risk management planning, benefit-risk evaluation and regulatory decision-making—depends upon determining whether an observed medical event is likely to be causally related to a medicinal product.

Unlike many areas of clinical medicine, pharmacovigilance rarely allows direct proof of causality. Decisions are usually made using incomplete information derived from spontaneous reports, clinical trials, pharmacoepidemiological studies, published literature, biological mechanisms and regulatory experience. Individual pieces of evidence are seldom definitive. Instead, causality emerges gradually as multiple independent sources of evidence are evaluated together.

Consequently, causality assessment is not a single test, scoring system or algorithm. It is a structured scientific process that combines medical judgement, pharmacological knowledge, epidemiological principles and critical evaluation of the totality of available evidence.

This article provides a comprehensive overview of causality assessment in pharmacovigilance. It explains the scientific principles underlying causal inference, reviews commonly used assessment methods, discusses their strengths and limitations and describes how experienced pharmacovigilance professionals integrate evidence when evaluating individual cases, safety signals and recognised risks throughout the lifecycle of a medicinal product.

Learning Objectives

After reading this article, you should be able to:

• explain the concept of causality within pharmacovigilance;

• distinguish association from causation;

• understand the strengths and limitations of individual case causality assessment methods;

• explain why aggregate causality differs from individual case assessment;

• evaluate different sources of evidence used to support causal inference;

• understand the role of biological plausibility, temporality, dose-response relationships and alternative explanations;

• describe how causality influences signal evaluation, benefit-risk assessment and regulatory decision-making;

• recognise common errors and misconceptions in causality assessment.

Why Causality Assessment Matters

Every pharmacovigilance activity ultimately attempts to answer a single scientific question:

Did the medicinal product contribute to the occurrence of the observed adverse event?

This question arises repeatedly throughout the lifecycle of a medicinal product.

During individual case review, medical reviewers assess whether a reported adverse event is plausibly related to treatment.

During signal management, reviewers evaluate whether multiple observations collectively suggest a new causal association.

Within aggregate reports such as the Periodic Benefit-Risk Evaluation Report (PBRER), cumulative evidence is reviewed to determine whether the current understanding of recognised risks has changed.

Risk Management Plans identify important identified risks, important potential risks and missing information based upon the strength of available causal evidence.

Ultimately, regulatory authorities rely upon causal assessments when deciding whether product information should be updated, additional pharmacovigilance activities initiated or further regulatory action taken.

Accordingly, causality assessment forms the scientific foundation upon which almost every pharmacovigilance decision is built.

Causality Is Rarely Absolute

One of the most important concepts in pharmacovigilance is that causality is seldom established with complete certainty.

Unlike experimental laboratory sciences, medicines are used in heterogeneous patient populations with different diseases, concomitant medicines, genetic characteristics and environmental exposures.

Adverse events may occur because of:

• the medicinal product;

• the underlying disease;

• concomitant medicines;

• medical procedures;

• genetic susceptibility;

• environmental factors;

• coincidence;

• interactions between several contributing factors.

Consequently, pharmacovigilance usually evaluates the probability and strength of a causal relationship rather than attempting to prove causation absolutely.

Scientific confidence increases progressively as additional evidence accumulates throughout the product lifecycle.

Scientific Foundation

Pharmacovigilance does not ask whether causality has been proven beyond all doubt. It asks whether the cumulative scientific evidence supports a causal relationship strongly enough to justify medical or regulatory action.

Association, Correlation and Causation

One of the most common misunderstandings in pharmacovigilance is the assumption that observing a relationship between a medicinal product and an adverse event automatically implies causation.

In reality, not every observed association represents a true causal relationship. Distinguishing between association, correlation and causation is therefore one of the fundamental responsibilities of pharmacovigilance professionals.

Failure to make this distinction may result in unnecessary regulatory action, inappropriate changes to product information or, conversely, failure to recognise an important safety concern.

What Is an Association?

An association exists whenever two events are observed together more frequently—or less frequently—than would be expected by chance alone.

For example, spontaneous reports may indicate that patients receiving a particular medicinal product experience acute pancreatitis more frequently than anticipated.

At this stage, the observation simply demonstrates that the medicine and the adverse event appear to occur together.

It does not explain why.

The observed association may reflect:

• a genuine adverse drug reaction;

• the underlying disease being treated;

• concomitant medicines;

• characteristics of the treated population;

• increased clinical awareness;

• stimulated reporting;

• random variation;

• systematic bias;

• confounding factors.

Association therefore represents the starting point of scientific investigation rather than its conclusion.

What Is Correlation?

Correlation describes the degree to which two variables change together.

For example:

• increasing dose may be associated with increasing reporting frequency;

• increasing exposure over time may coincide with increasing adverse event reports;

• two laboratory parameters may change in parallel.

Correlation can be measured statistically using methods such as correlation coefficients or regression analysis.

However, statistical correlation alone does not establish causality.

Two variables may correlate because:

• one causes the other;

• both are caused by a third factor;

• the relationship is entirely coincidental;

• systematic bias has created an apparent relationship.

Consequently, statistical correlation should always be interpreted within its broader scientific and clinical context.

What Is Causation?

Causation exists when exposure to a medicinal product contributes directly or indirectly to the occurrence of an adverse event.

Unlike association or correlation, causation implies that altering exposure to the medicinal product would be expected to influence the probability of the adverse event occurring.

Demonstrating causation requires evaluation of multiple complementary sources of evidence rather than reliance upon a single observation.

Examples of evidence supporting causality include:

• consistent temporal relationships;

• positive dechallenge or rechallenge;

• biological plausibility;

• dose-response relationships;

• reproducibility across independent studies;

• pharmacoepidemiological evidence;

• mechanistic understanding;

• consistency across different populations.

No individual criterion is sufficient in every situation.

Instead, confidence in causality increases progressively as multiple independent observations support the same conclusion.

Why the Distinction Matters

Failure to distinguish association from causation has important scientific and regulatory consequences.

Assuming causality too early may result in:

• unnecessary product information updates;

• inappropriate restrictions on patient access;

• diversion of pharmacovigilance resources;

• unnecessary regulatory investigations.

Conversely, dismissing genuine causal associations may delay identification of important safety risks and expose additional patients to preventable harm.

The objective of pharmacovigilance is therefore not to assume or reject causality prematurely but to evaluate competing explanations objectively until sufficient evidence becomes available.

Examples from Pharmacovigilance

The distinction between association and causation can be illustrated using common pharmacovigilance scenarios.

Scenario 1 – Association without Causation

A medicinal product used to treat advanced malignancy is associated with a high number of deaths reported through spontaneous reporting.

Although an association exists, most deaths may result from progression of the underlying disease rather than the medicinal product itself.

Additional evaluation is therefore required before concluding that a causal relationship exists.

Scenario 2 – Correlation without Causation

Following publication of a regulatory safety communication, reporting of liver injury increases substantially.

The increase correlates with the timing of the communication.

However, the apparent increase may reflect stimulated reporting (notoriety bias) rather than a genuine increase in the incidence of hepatotoxicity.

Scenario 3 – Progressive Evidence for Causation

Several independent observations emerge over successive years.

Initially, spontaneous reports suggest a possible association.

Subsequently, similar findings are observed in clinical trials.

Pharmacoepidemiological studies demonstrate an increased relative risk.

Experimental studies identify a biologically plausible mechanism.

The association gradually evolves into a well-supported causal relationship through accumulation of complementary evidence.

Causality Is a Continuum

Rather than viewing causality as a simple yes-or-no decision, experienced pharmacovigilance professionals recognise that confidence develops progressively.

Early observations generate hypotheses.

Additional evidence strengthens or weakens those hypotheses.

Scientific confidence increases as independent evidence converges upon the same explanation.

Equally, confidence decreases when new evidence supports alternative explanations or contradicts previous assumptions.

This progressive accumulation of knowledge underpins signal management, aggregate reporting, benefit-risk evaluation and regulatory decision-making throughout the lifecycle of a medicinal product.

Scientific Writing Principle

An observed association should never be described as causal unless the available evidence justifies that conclusion. Likewise, the absence of definitive proof should not prevent objective evaluation of a plausible causal hypothesis.

The Causal Confidence Continuum

One of the most important concepts in pharmacovigilance is that confidence in causality develops progressively over time.

A medicinal product is rarely considered to cause an adverse reaction after a single observation. Instead, scientific confidence evolves as additional evidence becomes available from different independent sources.

Experienced pharmacovigilance professionals therefore do not think of causality as a binary decision.

Instead, they ask:

"How confident are we that this medicinal product contributed to this adverse event based upon the totality of the available evidence?"

This confidence may increase, remain unchanged or even decrease as new evidence becomes available.

Stage 1 — Initial Observation

The process begins with one or more observations.

These may include:

• an Individual Case Safety Report;

• an unexpected laboratory finding;

• a clinical trial observation;

• a published case report;

• a regulatory communication;

• an observation made during routine medical practice.

At this stage, no conclusion regarding causality should be reached.

The observation merely raises a scientific question that deserves further evaluation.

Stage 2 — Safety Hypothesis

As similar observations accumulate, a scientific hypothesis begins to emerge.

For example:

"Could this medicinal product increase the risk of acute pancreatitis?"

The objective at this stage is not to confirm causality but to determine whether the hypothesis deserves systematic investigation.

Alternative explanations should remain equally plausible until sufficient evidence becomes available.

Stage 3 — Safety Signal

When available evidence suggests a new potentially causal association requiring further evaluation, the hypothesis becomes a safety signal.

Importantly, a safety signal is not evidence that causality has been established.

Rather, it indicates that sufficient concern exists to justify formal scientific evaluation.

Signals may subsequently be:

• refuted;

• confirmed;

• refined;

• merged with existing risks;

• closed without further action.

Signal detection therefore represents the beginning of structured causal assessment rather than its conclusion.

Stage 4 — Possible Causal Association

Following formal evaluation, the available evidence may support a possible causal relationship.

Typical characteristics include:

• consistent temporal association;

• biologically plausible mechanism;

• several compatible case reports;

• limited supporting literature;

• absence of definitive epidemiological confirmation.

Important uncertainty usually remains.

Further evidence is generally required before regulatory conclusions can be strengthened.

Stage 5 — Probable Causal Association

As additional evidence accumulates, confidence may increase substantially.

Supporting evidence may include:

• reproducible findings in independent datasets;

• pharmacoepidemiological studies;

• positive dechallenge;

• occasional positive rechallenge;

• dose-response relationships;

• mechanistic evidence;

• consistency across different patient populations.

Although uncertainty may still exist, the cumulative evidence increasingly favours a causal interpretation.

Stage 6 — Established Causal Association

Eventually, the totality of evidence may support a well-established causal relationship.

Characteristics often include:

• consistent findings across multiple evidence sources;

• biological plausibility;

• reproducibility;

• regulatory recognition;

• incorporation into product information;

• routine consideration during clinical practice.

At this stage, the scientific discussion usually focuses less on whether causality exists and more on characterising the risk.

Examples include identifying:

• susceptible patient populations;

• risk factors;

• preventability;

• reversibility;

• clinical management;

• effectiveness of risk minimisation measures.

Stage 7 — Well-characterised Risk

The final stage is not simply recognition of causality.

Instead, the objective becomes comprehensive understanding of the adverse reaction.

Questions now include:

• Which patients are at greatest risk?

• Is the reaction dose dependent?

• Can early warning signs be recognised?

• Can the risk be prevented?

• How should affected patients be managed?

• Does the benefit-risk balance differ between indications?

This transition reflects an important change in scientific thinking.

The principal question is no longer:

"Does the medicine cause this reaction?"

Instead, it becomes:

"How can this recognised risk be characterised and managed most effectively?"

Confidence May Also Decrease

The continuum is not one-directional.

Scientific confidence may decrease when:

• larger studies fail to confirm earlier observations;

• alternative explanations become more convincing;

• reporting bias is identified;

• methodological limitations are recognised;

• mechanistic evidence is inconsistent;

• epidemiological studies contradict earlier hypotheses.

Accordingly, pharmacovigilance requires continuous re-evaluation rather than permanent acceptance of earlier conclusions.

This principle explains why important potential risks may be removed, signals may be closed and previously suspected associations may ultimately be considered unlikely.

Causality Is Dynamic

Causal assessment is therefore best viewed as a dynamic scientific process rather than a fixed decision.

Knowledge evolves throughout the lifecycle of a medicinal product.

Each new study, safety report, regulatory assessment or scientific publication contributes to a progressively more complete understanding of the medicine's safety profile.

Experienced pharmacovigilance professionals remain willing to strengthen, weaken or revise previous conclusions whenever the cumulative evidence justifies doing so.

Scientific Insight

Causality is not established by a single piece of evidence. It emerges through the gradual convergence of independent observations, scientific reasoning and critical evaluation of the totality of evidence over time.

The Philosophy of Causation

Before discussing specific causality assessment methods, it is useful to understand what scientists mean by "causation."

Although the term appears straightforward, causation has been debated extensively in medicine, epidemiology, philosophy and the natural sciences. Modern pharmacovigilance does not rely upon a single definition of causality. Instead, it integrates concepts from several scientific disciplines to evaluate whether exposure to a medicinal product has contributed to an observed adverse event.

Understanding these concepts helps explain why causality assessment cannot be reduced to a checklist or numerical score.

Deterministic and Probabilistic Causation

In some areas of science, causes produce predictable effects.

For example, under defined experimental conditions, a specific chemical reaction consistently produces the same products.

This is known as deterministic causation.

Medicines rarely behave this way.

The same medicinal product may produce an adverse reaction in one patient while thousands of other patients experience no similar event despite receiving identical treatment.

This occurs because drug safety depends upon numerous interacting factors including:

• age;

• genetics;

• organ function;

• concomitant medicines;

• underlying diseases;

• immune responses;

• environmental exposures;

• treatment duration;

• dose.

Consequently, pharmacovigilance primarily evaluates probabilistic causation.

Exposure to a medicinal product changes the probability that an adverse event will occur, rather than guaranteeing that it will occur.

Necessary and Sufficient Causes

An important concept in epidemiology is the distinction between necessary and sufficient causes.

A necessary cause is a factor that must be present for a particular outcome to occur.

Without that factor, the outcome cannot develop.

A sufficient cause is a complete set of circumstances capable of producing the outcome.

In pharmacovigilance, very few adverse drug reactions have a single necessary or sufficient cause.

Instead, adverse reactions usually develop because several contributing factors act together.

For example, drug-induced liver injury may depend upon:

• exposure to the medicinal product;

• genetic susceptibility;

• concurrent illness;

• alcohol consumption;

• interacting medicines;

• individual immune responses.

The medicinal product contributes to the outcome but is rarely the only contributing factor.

Multifactorial Causation

Most adverse drug reactions are multifactorial.

Rather than asking whether a medicine alone caused an event, pharmacovigilance asks whether exposure contributed meaningfully to its occurrence.

Multiple contributing factors may coexist.

Examples include:

• advanced age;

• renal impairment;

• hepatic impairment;

• polypharmacy;

• infection;

• malignancy;

• autoimmune disease;

• inherited susceptibility.

Recognition of multifactorial causation explains why apparently similar patients may experience very different clinical outcomes despite receiving the same treatment.

Rothman's Causal Pie Model

One of the most influential concepts in modern epidemiology was proposed by Kenneth Rothman.

Rather than viewing diseases as resulting from a single cause, Rothman described outcomes as arising when several component causes combine to form a sufficient cause.

A medicinal product may therefore represent only one component within a larger causal pathway.

Other components may include:

• genetic predisposition;

• environmental exposures;

• underlying disease;

• concomitant medicines;

• lifestyle factors.

Removal of one important component may prevent the outcome even though other contributing factors remain.

This model reflects many adverse drug reactions encountered in routine pharmacovigilance and helps explain why individual susceptibility differs substantially between patients.

Counterfactual Thinking

Modern causal inference frequently uses the concept of the counterfactual.

The fundamental question is:

"What would probably have happened if this patient had not received the medicinal product?"

Unfortunately, this alternative reality cannot usually be observed directly.

Instead, pharmacovigilance attempts to estimate the answer using available evidence.

Evidence contributing to this assessment may include:

• dechallenge;

• rechallenge;

• epidemiological studies;

• comparison with untreated populations;

• historical controls;

• biological mechanisms;

• cumulative clinical experience.

Counterfactual reasoning explains why causality assessment is fundamentally an exercise in scientific inference rather than direct observation.

Why Absolute Proof Is Rare

Absolute proof of causality is uncommon outside exceptional situations.

Randomised controlled trials may demonstrate an increased incidence of an adverse event, but they often lack sufficient power to detect very rare reactions.

Spontaneous reports provide valuable clinical observations but are susceptible to bias and confounding.

Observational studies improve understanding but introduce additional methodological challenges.

Consequently, pharmacovigilance depends upon the convergence of multiple independent sources of evidence rather than a single definitive experiment.

Scientific confidence therefore develops progressively as different forms of evidence support the same causal explanation.

Practical Implications for Pharmacovigilance

Understanding these principles changes how causality is evaluated.

Rather than asking:

"Did this medicine definitely cause the adverse event?"

Experienced pharmacovigilance professionals ask:

• Has exposure increased the probability of the event?

• How strong is the available evidence?

• Which alternative explanations remain plausible?

• How much uncertainty remains?

• Would additional evidence be expected to strengthen or weaken the current hypothesis?

These questions reflect the scientific reasoning underlying modern pharmacovigilance.

Scientific Foundation

Causality assessment is not the search for absolute proof. It is the systematic evaluation of whether the totality of available evidence supports a causal contribution by the medicinal product while recognising alternative explanations and remaining uncertainty.