Risk Identification and Characterization

Risk identification and characterisation is the front end of drug-safety risk management: detecting that a medicine may cause a harm (a safety signal) and then describing that harm precisely enough to act on it — who is affected, how often, how severely, and through what mechanism. It draws on spontaneous reporting, statistical signal-detection methods, and clinical and epidemiological evaluation.

Definition

Risk identification is the detection of a possible new or changed causal association between a drug and an adverse event (a signal); risk characterisation is the subsequent description of that risk's frequency, severity, reversibility, affected population, and plausible mechanism.

Scope

This topic covers what a safety signal is, the data sources and disproportionality and Bayesian methods used to surface signals from spontaneous-report databases, and the steps of validating, prioritising, and characterising a signal into a defined risk. It is framed as reference methodology within pharmacovigilance rather than clinical guidance.

Core questions

• Is there a signal that a drug may cause a particular adverse event?
• Is the association likely causal, or explained by bias, confounding, or chance?
• How frequent and severe is the risk, and who is most affected?
• Which signals should be prioritised for further investigation?

Key concepts

• Safety signal
• Spontaneous reporting systems
• Disproportionality analysis (PRR, ROR)
• Bayesian methods (BCPNN, EBGM)
• Signal validation and prioritisation
• Causality assessment
• Identified versus potential risk

Mechanisms

Most signals first arise from spontaneous reports of suspected adverse reactions collected in national and international databases. Quantitative screening flags drug-event pairs reported more often than expected: frequentist disproportionality measures such as the proportional reporting ratio compare observed-to-expected reporting (Evans et al., 2001), while Bayesian approaches such as the Bayesian confidence propagation neural network shrink unstable estimates from sparse data (Bate et al., 1998). Statistical disproportionality only generates hypotheses; signals are then validated, assessed for causality, and characterised clinically and epidemiologically — defining frequency, severity, risk factors, and mechanism — before being classified as identified or potential risks (Wisniewski et al., 2016; Edwards & Aronson, 2000).

Clinical relevance

The risks defined through this process become the safety information in labelling and communications that clinicians and patients use. This entry explains how those risks are detected and described at the population level and does not provide individual diagnostic or treatment advice.

Epidemiology

Spontaneous reporting is subject to under-reporting and reporting biases, so disproportionality signals reflect reporting patterns rather than true incidence. Characterisation therefore often draws on additional sources — observational studies, registries, and electronic health-care data — to estimate frequency and identify risk factors more reliably.

History

Systematic signal detection grew from spontaneous-reporting schemes established after the thalidomide tragedy, such as the UK Yellow Card system and the WHO Programme for International Drug Monitoring. Quantitative methods matured in the late 1990s and 2000s, with the BCPNN (Bate et al., 1998) and proportional reporting ratios (Evans et al., 2001) becoming standard screening tools, later consolidated into good signal-detection practice (Wisniewski et al., 2016).

Debates

How much can disproportionality statistics tell us?

Disproportionality measures efficiently screen large databases but only generate hypotheses; they are sensitive to reporting biases and database quirks and cannot by themselves establish causality or true frequency.

Key figures

• Andrew Bate
• Stephen J. W. Evans
• I. Ralph Edwards

Seminal works

• bate-1998
• evans-2001
• edwards-aronson-2000

Frequently asked questions

What is a safety signal?

A signal is reported information suggesting a possible new or changed causal association between a drug and an adverse event that warrants further evaluation; it is a hypothesis to be investigated, not a confirmed risk.

Does a disproportionality signal mean the drug caused the event?

No. Disproportionality flags drug-event pairs reported more often than expected, but the association still has to be validated and assessed for causality before it can be treated as a real risk.