Signal Detection

The foundational quantitative engine of signal detection. It identifies drug-event combinations reported more frequently than expected.

• Frequentist methods: Proportional Reporting Ratio (PRR) and Reporting Odds Ratio (ROR) provide straightforward calculations.
• Bayesian methods: Empirical Bayes Geometric Mean (EBGM) applies Bayesian shrinkage to reduce false positives from low-count pairs.
• Compares observed counts against a statistical expectation derived from the entire database background.

Robust signals are rarely identified from a single source. The process requires cross-referencing multiple data streams.

• Spontaneous reporting databases: FAERS, EudraVigilance, and VigiBase provide global ICSR volumes.
• Literature monitoring: Systematic screening of medical journals for published case reports.
• Supplementary sources: Social media listening and electronic health records offer real-world data context.

Statistical association alone does not confirm a signal. Clinical evaluation examines the evidence for a causal relationship.

• Temporal relationship: Did the adverse event occur after drug administration?
• Dechallenge/Rechallenge: Did the event abate upon withdrawal and recur upon re-administration?
• Confounding by indication: Is the underlying disease, not the drug, the true cause of the event?

Detected signals must be triaged based on impact and verified through a rigorous validation process.

• Seriousness criteria: Events resulting in death, hospitalization, or disability are prioritized.
• Expectedness: Is the event already listed on the product label? Unexpected signals demand urgent action.
• Signal validation confirms the new causal association before regulatory escalation and aggregate reporting.

Signal detection depends on the precise, consistent coding of adverse events using controlled terminologies.

• MedDRA: The global standard for coding adverse event reports throughout the regulatory lifecycle.
• Standardised MedDRA Queries (SMQs): Pre-defined groupings of terms for retrieving complex conditions like anaphylaxis.
• Enables accurate aggregation and comparison of events across different products and regions.

Spontaneous reporting data is inherently noisy. Effective signal detection must account for systematic biases.

• Notoriety bias: A drug already in the media may see a spike in reporting for all events.
• Weber effect: Reporting peaks in the first two years after a drug's market launch.
• Confounding by indication must be ruled out to avoid attributing a disease symptom to the drug.

The foundational quantitative methodology for signal detection in spontaneous reporting databases. It identifies drug-event combinations reported more frequently than expected compared to a background reference set.

• Frequentist methods: Proportional Reporting Ratio (PRR) and Reporting Odds Ratio (ROR)
• Bayesian methods: Empirical Bayes Geometric Mean (EBGM) using the Multi-item Gamma Poisson Shrinker (MGPS)
• Stratification by age, sex, and reporter type reduces confounding
• Thresholds like PRR ≥ 2, Chi-squared ≥ 4, and n ≥ 3 trigger initial alerts

A statistical technique that shrinks observed disproportionality scores toward a null value to reduce false-positive signals from drug-event combinations with very low report counts.

• Protects against spurious associations driven by sparse data
• The EB05 and EB95 credible intervals quantify signal precision
• A lower bound of the 90% credible interval (EB05) ≥ 2 is a standard signal threshold
• Essential for analyzing rare events or newly marketed drugs with limited exposure data

The FDA Adverse Event Reporting System, a publicly accessible repository containing millions of spontaneous adverse event and medication error reports submitted by healthcare professionals, consumers, and manufacturers.

• Supports quarterly data mining for emerging safety signals
• Reports coded using MedDRA preferred terms for standardized analysis
• Susceptible to reporting biases: Weber effect, notoriety bias, and under-reporting
• Complements EudraVigilance and VigiBase in global signal detection workflows

The rigorous process of evaluating a detected safety signal to confirm the existence of a new causal association or a new aspect of a known association before regulatory action.

• Strength of evidence: Statistical robustness and consistency across databases
• Biological plausibility: Pharmacological mechanisms and preclinical data
• Clinical context: Patient demographics, co-morbidities, and concomitant medications
• Includes causality assessment factors like dechallenge/rechallenge and temporal relationship

A critical bias in pharmacovigilance where the underlying disease being treated, rather than the drug itself, is the true cause of an observed adverse event.

• Example: Asthma medications associated with asthma-related deaths
• Channeling bias: Patients with severe disease preferentially prescribed certain drugs
• Mitigation requires stratified analysis and comparison to active comparators
• Complicates signal interpretation and requires clinical adjudication

The systematic, global screening of published scientific and medical literature to identify Individual Case Safety Reports (ICSRs) and new drug-safety findings.

• MEDLINE/PubMed and Embase are primary sources
• NLP models extract adverse event mentions, drug names, and patient demographics
• Required by ICH E2D guidelines for post-market surveillance
• Identifies signals before they appear in spontaneous reporting databases