Where the safety information comes from
A pharmacovigilance document draws on many sources of safety information, among them clinical studies, published literature, spontaneous reports, registries, patient support programmes, marketing studies and medication errors. Those sources feed a safety database, and what can be concluded from them is only as sound as the data that database holds.
Sources of safety information
A pharmacovigilance document has to take account of many sources of safety information, and the structure of the document is itself an aid to considering all of them. Clinical studies, medical literature, spontaneous reporting, and observational and epidemiological studies are all sources of safety information, and so are patient support programmes, registries and even marketing studies wherever safety information is reported through them, while post-authorisation safety studies are designed to collect safety information on particular risks.
After marketing authorisation, data are normally collected through spontaneous reporting, which could be one of the main sources of post-marketing data, though a limited one. It depends on patients, physicians or pharmacists reporting to a company or to a health authority, and many spontaneous reports carry very scarce information or are not medically confirmed, so it is important to try to get more precise information in order to perform a proper evaluation. A specific adverse event follow-up questionnaire is one enhanced way to collect data: once a spontaneous report of, say, liver injury arrives, the company could use such a questionnaire to ask for follow-up with more structured data.
The European Medicines Agency's EudraVigilance and the FDA Adverse Event Reporting System collect reports of adverse events for post-marketing surveillance, and the UK's Yellow Card scheme, run in 2025 by the Medicines and Healthcare products Regulatory Agency, was set up many decades ago in response to thalidomide, with analogous spontaneous reporting schemes in many countries. In its annual report for 2014, the European Medicines Agency reported that the capacity for safety monitoring had been strengthened as patients, healthcare professionals and companies followed up on their commitment to report side effects to it: the number of suspected side effects reported in EudraVigilance, the EU adverse drug reaction collection and management system, rose by 6.5% for human medicines and by 27% for veterinary medicines against 2013. With the advent of digital media, the number of adverse events reported has increased dramatically: the FDA received reports of more than 1.8 million adverse events related to drug use in 2017, a 400% increase on the 363,171 reports it received in 2007. To build robust drug safety surveillance systems, pharmaceutical companies were seeking in 2019 to mine big data from other electronic sources, including electronic health records, the medical literature and social media, to identify adverse events.
Virtually any study and any source of information could hide safety information, and for a marketing study the writer can normally rely on the subject-matter experts working on it. So the important thing is to have a reliable contact person and to ask whether any safety information has arisen from such a study.
One experienced pharmacovigilance writer holds that virtually any register can be useful for pharmacovigilance, because registers collect very large amounts of data from large populations. As with marketing studies, the writer needs an expert colleague involved with the register who can say whether it holds safety data relevant to the document. The registers usually referred to are disease registries, because their patient population is likely to resemble the population the medicine is intended to address.
Medication errors are unintended mistakes in the prescribing, dispensing and administration of a medicine that could cause harm to a patient, and they are the most common preventable cause of adverse events in medication practice and a major public health burden. With the entry into force of the EU pharmacovigilance legislation in 2012, reporting all suspected adverse reactions resulting from medication errors became mandatory, and companies and national regulatory agencies in the EU member states are obliged to enter those events in EudraVigilance.
On 14 April 2015 the European Medicines Agency, on behalf of the EU regulatory network, released two draft good practice guides that aim to improve the reporting, evaluation and prevention of medication errors by regulatory authorities and industry throughout the EU, with comments due by 14 June 2015. Their primary purpose is to support industry and regulators in implementing those legal requirements. One of them covers how suspected adverse reactions caused by medication errors should be recorded, coded, reported and assessed, and recommends how marketing-authorisation holders should report information on medication errors brought to their attention that have not caused adverse reactions. That information must be provided in periodic safety update reports and in the risk management plans compulsory for all medicines, so that regulators can evaluate the benefits and risks of a medicine continuously on real-life data. The guidance is one of the key deliverables of the joint action plan on medication errors agreed in 2013 by the agency and the heads of medicines agencies; it was developed in consultation with the European Commission's patient safety and quality of care working group and takes into account stakeholder recommendations gathered at a workshop held at the agency in February 2013.
The safety database
Associating a potential risk with a medicine depends on the quality of the available data; in other words, an efficient pharmacovigilance system requires high-quality data, and the safety database is where those data are kept. The database stores adverse events, adverse drug reactions, medication errors, pregnancy cases, off-label use, misuse, abuse and overdose, among other data, from clinical trials, spontaneous reports by healthcare professionals and consumers, the literature, regulators, programmes following solicited reporting such as compassionate use, and post-marketing surveillance studies. High-quality safety databases are central to pharmacovigilance and essential for its activities, including signal detection, risk assessment and management, and regulatory reporting. Every marketing authorisation holder keeps a safety database, in-house or with a service provider. Based on local requirements, a significant amount of its content is reported to competent authorities, and the holder uses it for meta-analyses of pooled pre- and post-authorisation trial data as the sample size increases, assessing risks and implementing risk minimisation measures through signal detection and evaluation, aggregate reports, risk management plans, responses to health authority requests and label updates. All pharmacovigilance activities contribute to assessing a product's safety profile and benefit-risk balance, and their data significantly affect regulatory decision-making: for example, the Pharmacovigilance Risk Assessment Committee's assessment of a periodic benefit-risk evaluation report can lead to requests for further analyses and monitoring, updates to labels or risk management plans, and urgent market restrictions by any local authority. Accurate and well-managed databases enable robust and conclusive analyses to be performed, supporting regulatory decision-making and patient safety.
The volume of data in a safety database depends on the active substance, the number of marketing authorisations held worldwide and the size of the clinical development programme, and can reach thousands of cases within a few years. Every year more medicines are authorised and medicine use is increasing significantly, so the number of adverse events reported to manufacturers and regulators has increased and will continue to increase. That volume of data is the most important reason the stored data should meet the four Cs: correct, comprehensive, consistent and conclusive. Retrieval from a curated safety database is the first step that scientists, writers and regulatory or medical affairs staff take in the activities the database serves, and because data keep accumulating and their volume can be rather large for certain medicines, retrieval needs a systematic approach. Targeted queries are commonly built from MedDRA terms, such as standardised MedDRA queries, high-level terms and high-level group terms, or from customised search queries that compile preferred terms, which facilitates faster, automated or reproducible and, to an extent, less subjective extraction. An analysis may focus on a defined period, the interval, but in most cases the interval data still need to be evaluated against the cumulative data held since development or authorisation began.
The quality of its data
Data entry errors, inconsistent reporting practices and underreporting are among the factors that most affect a safety database, leaving its stored data poor and biased. Errors in data entry and underreporting can significantly affect the accuracy and reliability of safety assessments. Data entry errors include incorrect or partial transfer from sources such as serious adverse event forms, medical information call centres and social media; incorrect mapping of reported verbatim terms to preferred terms and on to lowest level terms, system organ classes and high-level terms; and failure to comply with regulatory guidelines such as good pharmacovigilance practices. Lack of compliance or harmonisation with regulatory guidelines and non-standardised classification of cases are significant obstacles to analysing data across sources such as a company database and EudraVigilance.
If an authority requests a cumulative analysis over a long interval, a compromised database most likely returns many false positives that wrongly meet the search criteria and inevitably misses many more cases that the criteria do not pick up, leaving a dataset that lacks statistical power, and either way a great deal of manual work is needed to remove the false positives or, where possible, retrieve the missed cases. When a final analysis concludes that cases lacked sufficient data to determine causality, or that conclusions cannot be drawn until more data are collected, the information is in many cases missing because fields in the database were not correctly populated from the available source data or from the seriousness or causality assessments of case processing teams and medical reviewers, or because follow-up was not attempted in time. Signal detection is the cornerstone of pharmacovigilance, so a signal refuted on a dataset compromised by poor data entry means that the aim of pharmacovigilance was not met and that professionals unintentionally stopped watching over patients' safety, if only for a while.
To minimise these challenges, data should regularly be validated and reconciled between databases, holders should provide robust training, implement efficient quality checks and keep consistent and detailed data-handling guidance, and targeted follow-up should be attempted to recover missing information, alongside regular clean-ups at the highest level possible.
As for underreporting, the belief that reporting an already known adverse event changes nothing is a mistake, because reporting is the essence of pharmacovigilance: without it, potential new aspects of a marketed medicine, event frequencies that differ from those in the label, new information that clarifies the mechanism of action, and the reasons adverse events occur would go undetected. Common excuses for not reporting include leaving it to the doctor, not being an expert and not wanting trouble, or having no time, and members of the pharmacovigilance community have themselves postponed reporting an event only to realise afterwards that it was too late. Despite initiatives to raise awareness and streamline reporting, more effort is needed from all sides to give healthcare professionals and consumers an incentive to report and to engage more stakeholders in adverse event reporting. The vaster the range of data, the more thorough the analysis and the more robust the conclusions.
Database migration is perhaps one more factor that affects a safety database. Companies commonly migrate to new databases, through collaborations with new service providers or system updates, to improve quality and efficiency, and poor data quality in the existing database lengthens the transfer, adds manual work and ultimately yields worse data.
Databases whose data are correct, comprehensive, consistent and conclusive support the timely detection of new and significant safety issues and speed up the implementation of risk mitigation measures. Setting up a safety database correctly, running effective routine and proactive processes to manage it, complying with regulatory requirements and reporting adverse events in time can maximise the effectiveness of pharmacovigilance. If pharmacovigilance were a chemical reaction turning raw data into an established safety profile, a database of high-quality data would be its catalyst and one of low-quality data its inhibitor.