Each day, millions of patients rely on medicines that are already on the market. However, a medicine’s safety profile doesn’t cease once it reaches the market. Rare adverse events, long-term effects, and unexpected drug interactions may only emerge after widespread use, making continuous safety monitoring essential.
The challenge of the pharmacovigilance market isn’t the lack of data. It’s the sheer overwhelming volume of it. From spontaneous adverse event reports and clinical studies to scientific literature, identifying genuine safety matters among thousands of data points has become increasingly complex.
This is where safety signal management serves a key role. A safety signal is information about a new or known adverse event that warrants additional investigation to determine whether a medicine may pose a potential risk.
As regulatory criteria continue to evolve under GVP Module IX, pharmacovigilance teams are also exploring how artificial intelligence (AI) can strengthen signal detection. AI can rapidly examine vast amounts of structured and unstructured data, helping identify patterns, prioritise possible signals, and support faster decision-making.
Before we delve deeper, let’s understand what a safety signal is.
What is a safety signal?
A safety signal is information about a known or new adverse event associated with a medicine that needs further investigation. EMA and other regulatory bodies in Member States, as well as MAH, are responsible for detecting and managing safety signals.
These safety signals are detected from multiple sources, including spontaneous reports, clinical studies, and the medical literature. Here, the EudraVigilance database is a central source of information on suspected AEs and signals.
In case there is a safety signal, it does not imply that the medicine has caused the adverse event. It could also be caused by an illness or another medicine. When an assessment of the safety signal is conducted, it determines whether there is a relationship between the medicine and the reported AE.
The assessment of safety signals is part of PV and is important for ensuring that regulatory bodies have up-to-date information on medicines’ benefits and risks.
A safety signal is a key part of signal management as outlined in GVP Module IX.
Brief overview of signal management under GVP Module IX
GVP is divided into modules that govern various aspects of pharmacovigilance processes. GVP Module IX on signal management lays out general guidance and requirements on the scientific and quality aspects of signal management.
This applies to Marketing Authorisation Holders (MAHs) for medicines authorised by the EMA and by EU member states. In the absence of formal regulations for signal management by health regulators, these guidelines are considered de facto.
GVP IX signal detection
GVP Module IX notes that common data sources for signal detection include spontaneous reports (the MAH’s safety database and national/global repositories such as FAERS, EudraVigilance, and VigiBase), active surveillance systems, clinical and pharmacoepidemiology studies, and the scientific literature.
Which sources to use should be determined by clinical judgement and the product’s characteristics. Signal detection processes should combine manual review of Individual Case Safety Reports (ICSRs), statistical analyses, or both. In practice, disproportionality analysis should be applied alongside additional data summaries and clinical assessment so that statistical signals are interpreted in the context of clinical and pharmacovigilance evidence.
Moving forward, AI is an integral part of signal detection in this era. Let’s understand it.
Where AI fits in signal detection
Signal detection has remained a pivotal part of pharmacovigilance for years now. AI is best suited as a decision-support tool in signal detection. It can help triage literature, cluster cases, identify duplicates, spot emerging trends, and rank items for review. One core example of AI involvement in signal detection is as follows.
Imagine you are trying to search all of Twitter for a specific treatment side effect or going through reading every journal that refers to a certain condition. These data sets are often huge and unstructured, requiring signals to be detected from an often-unstructured narrative rather than from completed fields.
To adapt to this variable data source and quality, several pharmacovigilance teams are adopting AI, ML, NLP, and analytics tools that can be trained to detect signals based on keywords, phrases, and trends. For example, NLP can be applied to the full-text context of a possible signal. One can use algorithms to detect AEs based on sentence syntax and context within a report. With proper training, these algorithms can perform quick, accurate analysis, identifying signals that may impede the process.
Core best practices of signal detection
Here are the core controls of AI in pharmacovigilance. Keep AI as augmentation, define a narrow intended use, validate against real PV workflows and require human oversight at key checkpoints.
In signal detection, AI should be used to augment, not replace, qualified PV assessment. This is because it involves detecting and assessing safety signals rather than letting the software draw safety conclusions on its own.
The optimal approach is a human-in-the-loop workflow: AI flags patterns, ranks cases, and surfaces trends, while trained reviewers make the clinical and regulatory judgements.
Here are a few best practices for signal detection:
- Specify clear intended use and system scope so the model is used only for the signal management tasks for which it was built and validated.
- Validate models with real pharmacovigilance workflows and datasets, since signal management depends on practical performance in detection, validation, and evaluation, not just offline accuracy.
- Maintain audit trails for inputs, outputs, overrides, and final decisions so every signal decision can be reconstructed and defended during review or inspection.
- Establish human monitoring at every critical step so qualified reviewers can accept, modify, or reject AI outputs before they influence signal conclusions.
Regulatory and compliance considerations in signal management
- Document the full signal management lifecycle, viz., detection, validation, prioritisation, assessment, and actions. Ensure validated signals and outcomes are tracked in agency-accepted systems (e.g., EPITT) as required by GVP Module IX.
- Record roles, responsibilities, and frequencies for routine monitoring (e.g., monthly for routine EV outputs). This is most often for MPs under additional monitoring, with auditable timelines for when signals were detected, validated, and communicated to regulators.
- Maintain versioned SOPs that describe where AI is used, intended use limits, and acceptance/withdrawal criteria so functionality remains within the scope regulators expect.
Step-by-step Implementation of signal detection and labelling changes
Effective signal management and timely safety labelling changes are critical for PV. Here is a step-by-step guide on GVP module IX focusing specifically on signal detection.
Step 1: Insight into signal management
Signal management is an organised process through which PV firms identify, assess, and manage safety information from various sources. A few sources that contribute to the signal detection process are spontaneous reporting systems, clinical trial reports, literature reviews, and social media.
Step 2: Assessing signals for regulatory norms
When a possible signal has been identified, the next phase is an evaluation to determine the implications and significance of product safety. This evaluation takes into account several elements, such as the strength of the evidence, the relevance of the product label, and probable indications.
Step 3: Carrying out safety labelling changes
After a signal has been assessed, the next step is to implement safety labelling changes if warranted. This can range from minor updates showing newly understood risks to major modifications that require changes to indications or contraindications. Other actions include further data collection, additional investigations, enhanced monitoring, PASS, risk minimization measures, or no regulatory action if the signal is not confirmed.
Step 4: Keeping a record of changes
Each step in signal management needs to be documented carefully. When implementing safety labelling changes, organisations must follow protocols that comply with FDA and EMA requirements. This documentation includes:
- Signal detection analysis reports.
- Summary of the evaluation for each signal.
- Justification for any labelling changes.
- Updated SmPC or product labelling.
- Stakeholder interaction plan.
Step 5: Working with regulatory authorities
When safety labelling changes are implemented, it is important to engage with regulatory authorities like the FDA and EMA. When engaging with authorities, organisations must be prepared with all pivotal findings, assessments, and suggested label changes in a transparent manner.
Step 6: Continual Administration
Post-approval commitments may include additional studies, Risk Evaluation and Mitigation Strategies (REMS), or further clinical trials to confirm the product’s safety following labelling changes. These commitments should be tracked and reported in line with FDA requirements or those of other relevant regulatory authorities.
Conclusion
As regulatory expectations for signal detection continue to evolve, pharmacovigilance organisations must strengthen their processes. Secure prompt identification, evaluation, and management of emerging safety signals while maintaining conformity with global regulatory requirements. Signal detection is a one-time destination. It is an ongoing process that helps restrain adverse reactions to a drug. Is signal detection something you’re struggling with?
Take the first step and connect with us now for a FREE consultation.