Running a signal from detection to risk communication

A signal is information that suggests a new possibly causal link between a medicine and an event, or a new aspect of a known one. Managing a signal takes it through detection, validation and assessment to the actions that follow when it is confirmed and into the aggregate safety reports. Regulators monitor authorised medicines and act on them, from a change to the product information to a withdrawal, and safety information moves between agencies as well as within one.

What a signal is

The guidelines on good pharmacovigilance practices (GVP), in Module IX on signal management, define a signal as information arising from one or multiple sources, including observations and experiments, which suggests a new potentially causal association, or a new aspect of a known association, between an intervention and an event or set of related events, either adverse or beneficial, that is judged to be of sufficient likelihood to justify verificatory action. ¹ So a signal can concern labelled or listed events as well as unlabelled or unlisted ones for an active substance, and very often applies to every medicinal product containing that substance, including in combinations. ¹ Some signals may apply only to one pharmaceutical form or one indication, as injection and infusion site reactions apply only to products given by injection or infusion. ¹

The identification and investigation of safety signals is a key process in pharmacovigilance. ¹ Signal management is a fundamental process for identifying, validating and addressing safety concerns throughout a medicinal product's lifecycle. ² It draws on both individual case safety reports and aggregate reports. ³ What it looks for is an adverse event that may never have been reported with the product before. ⁴

Monitoring for signals

The sponsor or marketing authorisation holder is required to monitor the safety of its products continuously, both marketed and under clinical development, and describes how it does so in the pharmacovigilance system master file. ^{1, 5} It is responsible for maintaining a robust pharmacovigilance system, and a stepwise assessment approach ensures that safety monitoring is continuous. ²

The cornerstone of pharmacovigilance is the individual case safety report (ICSR), which may come from clinical trials, the literature or spontaneous reporting. ¹ The sponsor or holder also performs active surveillance in the scientific literature, in external safety databases such as the FDA Adverse Event Reporting System (FAERS), VigiBase and EudraVigilance, in same-in-class active substances where applicable, in real-world data such as electronic health records and, most recently, in social media. ¹ Health authority requests can be another source. ¹

A signal detection strategy should be drafted as soon as the product enters first-in-human studies in phase 1, and it must be tailored to the product's development stage, with activities such as literature review and searches of the clinical trial database, the post-marketing safety database and external safety databases. ⁵ As more is learned about the safety profile from earlier development stages or post-marketing surveillance, it is important to update the strategy, by adding or removing activities. ^{5, 1} It should be reviewed periodically or when the product enters a new phase, such as phase 2, phase 3 or marketing authorisation. ¹

The people involved in signal management are most often called the product safety team, and the person who performs signal detection is a safety scientist. ⁶ A good safety scientist must have a deep understanding of the product's safety profile, its mechanism of action and the pathology of the underlying conditions. ⁶

Signal management as a legal duty in the EU

The revised EU pharmacovigilance legislation made signal detection and signal management a legal requirement on all parties, one of its major public health developments. ⁷ The GVP guidance on signal management sets out the requirements on the European Medicines Agency (EMA), the member states and marketing authorisation holders, and it largely follows CIOMS VIII, with signal detection, validation, prioritisation, evaluation and communication. ⁷ Holders need documented processes for signal detection that suit the volume of reports they receive and their product portfolio, and these may be individual case review, statistical analysis, or a combination of both. ⁷ The guidance is clear on which signals to communicate to the authorities and when, and on when to respond to requests. ⁷ Member states carry similar requirements, with clear roles in monitoring EudraVigilance, and every substance authorised or registered in the EU is assigned a lead member state that is responsible for generating and validating signals from EudraVigilance and for the subsequent notification. ⁷ The UK agency considered this a vitally important aspect of the legislation, and held that in the UK it would complement the national pharmacovigilance system rather than replace current practices. ⁷

The stages of signal management

According to the GVP guidelines, the signal management process has six major milestones: signal detection, signal validation, signal confirmation, signal analysis and prioritisation, signal assessment, and recommendations for action. ⁸ For simplicity, these can be grouped into four stages, divided into two phases: ⁸

• Pre-confirmation: signal detection, signal validation and signal assessment. ⁸
• Post-confirmation: recommendations for action. ⁸

A key difference between the stages is the level of evidence and investigation, which rises progressively from detection to assessment. ⁸

Detection

Detection is the continuous monitoring of safety data from diverse sources, such as ICSRs from the post-marketing safety database, clinical trials, the literature and external databases, to identify potential new safety concerns. ⁹ It typically uses qualitative and quantitative methods, which may vary between companies: ¹

• Qualitative methods, such as literature review, ICSR review, expert review and review of same-in-class product labels, evaluate incoming information against the reference safety information, such as the investigator's brochure before marketing and the company core data sheet after it. They consider expectedness based on the safety profile, the mechanism of action, the underlying condition and comorbidities, and the event type, such as designated medical events. ^{1, 9}
• Quantitative methods assess trends and patterns, such as changes in the frequency of adverse events, reporting country, source, gender or age group, often by disproportionality analysis such as the reporting odds ratio or the proportional reporting ratio, in the holder's own safety database or in external databases. ⁸

In detection, all the data collected for the period from the data lock point, including case reports, literature and emails, are collated and reviewed for any abnormality in the side effects being reported, with checks of the analysis, causality, relationships and correlations. ⁴ A literature case with a product that the company does not market in the country concerned is not entered in the company's safety database as an individual case, but it is still reviewed for signal analysis, including from the point of view of the therapeutic class or the mechanism of action. ^{10, 11} If a safety scientist identifies a new signal, it is validated further. ⁸

Validation

Validation is a preliminary assessment of the data supporting a detected signal. ¹² Because signal assessment takes significant time and resources, this screening step aims to identify the signals with enough evidence to justify in-depth investigation, and it helps to prioritise resources by filtering out signals that are not credible early. ^{12, 9}

During validation, the safety scientist's steps are to: ⁵

• examine the trigger and its plausibility, based on the time to onset, the mechanism of action and the underlying conditions. ⁵
• search the clinical trial databases, the post-marketing safety databases and the literature. ¹³
• conclude whether the signal is validated and warrants thorough analysis, or is not validated and needs no further investigation. ¹

Validation also asks whether the correlation between the event and the drug is relevant, whether the event has been reported before, and whether other sources corroborate it. ⁴

Assessment

Once validated, a signal undergoes a comprehensive assessment of all the evidence available up to that point in the product's lifecycle. ⁶ Before the assessment begins, the safety team must define the event under evaluation clearly, based on: ⁶

• clinical manifestations ⁶
• aetiology, including alternative explanations ⁶
• pathophysiology, including biological plausibility related to the product ⁶
• diagnosis and differential diagnosis ⁶
• prevention and treatment ⁶

The data analysis methods should also be described, such as the clinical trials included and the literature and safety database search criteria, and the evidence is then evaluated against the event definition. ⁶ The evaluation takes in the factors specific to each trial participant or patient, and its sources include clinical trials, spontaneous reports, the literature and real-world data. ⁹ Multidisciplinary input is often required. ⁹ The evidence differs before and after marketing: ⁸

• Before marketing, the evidence includes non-clinical data, biological plausibility and data from same-in-class molecules. Participant-specific data, such as baseline status, concomitant medications, medical history and concurrent conditions, must be evaluated thoroughly to exclude other causal factors, and drug exposure data should also be considered, especially for rare events. ⁸
• After marketing, the additional sources include spontaneous ICSRs, the scientific literature, social media, epidemiological data, non-interventional studies, market research and patient support programmes, and several subject-matter experts contribute to the assessment. ⁸

The assessment checks the strength and consistency of the evidence: whether the event is a one-off or there is repeated evidence of the drug causing it, what the characteristics of that evidence are, and how it would affect public health. ¹⁴ The evidence is then examined, the association between the drug and the event quantified and more data sought, including from the literature. ¹⁴ The assessment also asks whether the effect could follow from the therapeutic class or from a mechanism of action shared within a class of products, and whether other products in that class can cause it. ^{14, 15}

The safety team then weighs the strength of the evidence and either confirms or refutes the signal. ⁸ Refuted signals return to routine monitoring. ⁸

Recommendations for action

Confirmed signals lead to recommendations for action based on risk and impact. ^{8, 12} If a signal is confirmed, the holder's safety governance boards may decide to implement measures, depending on the impact and importance of the event: ¹³

• Risk categorisation, as appropriate: ⁹
  • an important identified risk or an important potential risk ⁹
  • identified or potential risks not categorised as important ⁹
  • updated information on an existing risk, as when an important potential risk becomes an important identified risk where the evidence supports it ⁹
• Market withdrawal, in cases of major safety concerns. ⁹
• Updates to the reference safety information, including the investigator's brochure, the company core data sheet and regional labels. ⁹
• Risk minimisation measures, such as Dear Healthcare Professional letters, Dear Investigator letters, patient communications, restricted use and pregnancy prevention programmes. ⁹
• Enhanced safety monitoring, such as fixed-frequency event monitoring or an updated analysis in each periodic benefit-risk evaluation report (PBRER). ⁹
• Additional safety studies to characterise the risk further. ⁹

The actions are chosen on the strength of the evidence, and their goal is to mitigate the risk and ensure the safety of trial participants and patients. ⁹ Such decisions should be well documented in the signal assessment report and in the holder's safety management system for auditing. ⁹

The review also checks for regulatory input. ¹⁵ Once it is confirmed that the product causes the event, the company communicates the risk to doctors and other healthcare professionals, informing them of the probability and possibility of the side effect, and where needed it also recalls the product and puts active monitoring in place. ¹⁵ Risk minimisation informs the stakeholders that the product may not suit a particular age group or class of patients, or that pregnant patients should not take it. ^{15, 16} It defines the vulnerable group and decides who should and should not take the product, and the information is communicated to all the health authorities and to the licensed partners. ¹⁶

Signals in the aggregate reports

Signal management is a core process in understanding a medicinal product's risks and in further characterising its benefit-risk balance to ensure its safe use. ¹⁷ All new information is evaluated in light of the cumulative safety and efficacy data, and the sponsor or holder must draw conclusions on whether the marketing authorisation needs to change or other action on the product is needed. ¹⁷

All new safety information reported or identified during the reporting period of a PBRER or a periodic safety update report (PSUR) must be included in the report. ¹⁷ These reports have sections dedicated to safety signals, which may be titled "Overview of signals: new, ongoing or closed" or "Signal and risk evaluation", depending on the applicable guidance. ¹⁷ Both confirmed and refuted signals must be documented in these sections. ¹⁷ Confirmed signals that have been categorised as risks, whether important potential, important identified, or simply potential or identified, must be presented accordingly, and confirmed signals that give new insight into existing risks must also be described. ¹⁷

The development safety update report (DSUR) has a section for safety signals titled "Summary of important risks", which lists the important identified and potential risks, re-evaluated each year according to the stage of development and existing knowledge. ⁹ An appendix details these risks, with the existing data and the actions for risk minimisation. ⁹

Regulatory action on an authorised medicine

Once a medicine is authorised by the European Commission and prescribed to patients, the EMA and the EU member states continuously monitor its quality and benefit-risk balance and take regulatory action when needed. ¹⁸ The measures can include a change to the product information, the suspension or withdrawal of a medicine, or a recall of a limited number of batches. ¹⁸ In the EU, reported adverse drug reactions are collected in the EudraVigilance database, which is used to detect emerging safety signals. ¹⁹ The EMA's Pharmacovigilance Risk Assessment Committee (PRAC) is responsible for assessing and monitoring the safety of human medicines and for evaluating their safety issues. ^{19, 20}

Withdrawal, revocation and suspension

Zinbryta (daclizumab beta), authorised in 2016 for treating relapsing forms of multiple sclerosis, was suspended and then withdrawn in 2018. ^{21, 22} A PRAC review in 2017 of its effects on the liver had found that unpredictable and potentially fatal immune-mediated liver injury could occur for up to six months after stopping treatment, and had concluded that patients stopping treatment should be followed up. ²¹ After that review, its use was restricted to patients who had tried at least two other disease-modifying treatments and could not be treated with any other multiple sclerosis treatment. ²¹

The 2018 review was initiated at the European Commission's request on 26 February 2018, under Article 20 of Regulation (EC) No 726/2004, and was carried out first by the PRAC. ^{21, 22} On 6 March 2018, while the review was ongoing, the PRAC recommended suspending the marketing authorisation and recalling the product. ²² On 7 March the EMA recommended the immediate suspension of the marketing authorisation in the EU and a recall of batches from pharmacies and hospitals, following 12 reports worldwide of serious inflammatory brain disorders, including encephalitis and meningoencephalitis, three of them fatal. ²¹ A preliminary review of the evidence indicated that the immune reactions in those cases might be linked to the medicine. ²¹ Healthcare professionals were to contact the patients currently being treated immediately, stop their treatment and consider alternatives, and patients stopping treatment had to be followed up for at least six months. ²¹

By then over 8,000 patients had been treated worldwide, most of the EU patients in Germany. ²¹ The PRAC had reviewed 12 cases of immune-mediated inflammatory disorders, most of them occurring within eight months of starting treatment, and the available evidence also indicated that the medicine could be linked to other immune-mediated disorders, such as blood dyscrasias, thyroiditis or glomerulonephritis. ²¹ The recommendation was sent to the European Commission for a legally binding decision, and the holder had already voluntarily requested the withdrawal of the marketing authorisation and told the EMA it intended to stop its clinical studies. ²¹

The European Commission issued a legally binding decision suspending the marketing authorisation on 8 March 2018, and withdrew the authorisation on 27 March 2018 at the holder's request. ²² On 18 May 2018 the PRAC confirmed that the medicine poses a risk of serious and potentially fatal immune reactions affecting the brain, liver and other organs, that patients could be at risk from the start of treatment and for several months after stopping it, and that it is not possible to predict which patients will be affected. ²² By then over 10,000 patients had been treated worldwide, most of the EU patients in Germany, and healthcare professionals were expected to continue monitoring the patients who had been treated, in line with the recommendations issued in March 2018. ²²

On 24 May 2019 the PRAC recommended revoking the marketing authorisations of fenspiride medicines, so that they could no longer be marketed in the EU. ²³ These medicines, syrups or tablets available in the EU under various brand names, are used in adults and in children from the age of 2 years to relieve coughs resulting from lung diseases. ²³ The review behind the recommendation, initiated on 14 February 2019 at France's request under Article 107i of Directive 2001/83/EC, confirmed that these cough medicines could cause heart rhythm problems, and at its start the PRAC recommended suspending their supply as a precaution while it was ongoing. ²³ The PRAC considered all the available evidence, including case reports of QT prolongation and torsades de pointes in patients using the medicines, non-clinical studies including hERG channel binding, the results of laboratory studies, the published literature and stakeholder input. ²³ QT prolongation and torsades de pointes are abnormalities of the heart's electrical activity that may lead to heart rhythm disturbances. ²³ Heart rhythm problems can be serious and occur suddenly, and the patients who may be at risk cannot feasibly be identified in advance, whereas fenspiride medicines treat non-serious cough, so the PRAC considered that they should no longer be marketed. ²³ Because the medicines are all licensed nationally, the recommendation was to go to the Co-ordination Group for Mutual Recognition and Decentralised Procedures – Human (CMDh) to decide on its implementation. ²³ The CMDh is a medicines regulatory body representing the EU member states, Iceland, Liechtenstein and Norway, and it is responsible for harmonised safety standards for medicines authorised through national procedures across them. ^{23, 24, 25}

On 15 December 2017 the CMDh endorsed the PRAC's recommendation to suspend the marketing of modified- or prolonged-release products containing paracetamol, alone or combined with the opioid tramadol, which are designed to release paracetamol slowly over a longer period than the usual immediate-release products. ²⁴ The recommendation rested on a review of the data, including a retrospective pharmacokinetic and clinical analysis by the Swedish Poison Information Centre of 53 cases of acute overdose with modified-release paracetamol. ²⁴ The analysis found that the standard treatment protocol, using solely the Rumack-Matthew nomogram or variations of it and based on conventional paracetamol formulations, may not be effective for overdoses with modified-release formulations, and its results confirmed a similar Australian case series. ²⁴ The maximum plasma concentration may occur later, and high concentrations, particularly after large doses, may persist for several days, so the usual sampling and treatment protocols for overdose with immediate-release formulations are not adequate. ²⁴

The CMDh agreed that the advantages of a longer-acting product did not outweigh the complications of managing an overdose, since the treatment procedures for immediate-release products are not appropriate for modified-release paracetamol, and in many cases it may not be known which kind of product an overdose involves, making it difficult to decide how to manage it. ²⁴ It noted the PRAC's conclusion that no practical measures to reduce the risk to patients sufficiently had been identified, and no feasible, standardised way to adapt overdose management across the EU to both kinds of product could be agreed. ²⁴ The position was adopted by majority vote, so it was to go to the European Commission for an EU-wide legally binding decision. ²⁴ The medicines were to remain suspended unless the holders could provide evidence of appropriate and practical EU-wide measures to help prevent overdose and adequately reduce its risks, and immediate-release paracetamol products, which the review did not affect, remained available. ²⁴

Staying on the market under new measures

Hydroxyethyl starch (HES) solutions for infusion are colloids, blood volume expanders used to replace plasma volume after acute blood loss where crystalloids alone are not considered sufficient, to prevent a dangerous drop in blood pressure. ²⁵ In the EU they are approved through national procedures and are available in the member states under various trade names. ²⁵ In January 2018 the PRAC recommended suspending their marketing authorisations, because they continued to be used in critically ill patients and patients with sepsis despite restrictions introduced in 2013 because of the risk of kidney injury and death in these patients. ²⁵

On 29 June 2018 the CMDh agreed with the PRAC's assessment, but decided that the solutions should remain on the market provided that a combination of additional measures to protect patients was implemented. ²⁵ It gave further consideration to the place of HES in the clinical practice of some countries, noted that the previous risk minimisation measures had had some effect, and considered that a combination of new measures would effectively ensure that HES solutions are not used in patients at risk. ²⁵ The new measures are: ²⁵

• A controlled access programme run by the holders, so that only accredited hospitals are supplied, with accreditation requiring that the relevant healthcare professionals receive training on safe use. ²⁵
• Warnings on the packaging and at the top of the summaries of product characteristics, reminding healthcare professionals that the medicines must not be used in patients with sepsis or kidney impairment or in critically ill patients. ²⁵
• Writing directly to healthcare professionals so that they are fully aware of the conditions of use and of the groups of patients who must not receive the medicines because of an increased risk of kidney injury and death. ²⁵

The CMDh also asked the holders to conduct studies checking that only the patients who should be treated with these medicines are receiving them, in addition to ongoing studies of their benefits and risks in patients with trauma and those undergoing elective surgery. ²⁵ The position was adopted by majority vote and was to go to the European Commission for an EU-wide legally binding decision. ²⁵

Restrictions, contraindications and updated advice

On 15 November 2019 the EMA recommended restricting the use of the multiple sclerosis medicine Lemtrada (alemtuzumab) because of reports of rare but serious side effects, including deaths, and recommended new measures to identify and manage them. ²⁶ The side effects include cardiovascular disorders, affecting the heart, circulation and bleeding as well as stroke, and immune-related disorders, caused by the body's defence system not working properly. ²⁶ Lemtrada was authorised in the EU in 2013 to treat adults with relapsing-remitting multiple sclerosis. ²⁶ Its active substance, alemtuzumab, is a monoclonal antibody that attaches to CD52 on white blood cells, causing them to die and be replaced and so reducing the damaging activity of the immune system. ²⁶

Lemtrada should now be used only if the disease is highly active despite treatment with at least one disease-modifying therapy or is worsening rapidly. ²⁶ It must no longer be used in patients with certain heart, circulation or bleeding disorders or in patients with autoimmune disorders other than multiple sclerosis, and it should be given only in a hospital with ready access to intensive care facilities and to specialists who can manage serious adverse reactions. ²⁶ The EMA also recommended updating the physician's guide and the patient information pack with advice on minimising the risk of serious cardiovascular disorders, which may occur shortly after an infusion, and of immune-related conditions, which may occur many months and possibly years after the last treatment. ²⁶ The review was initiated on 10 April 2019 at the European Commission's request and was carried out first by the PRAC, which issued temporary recommendations restricting use while it was ongoing and gave its final recommendations to the human medicines committee (CHMP) on 31 October. ²⁶ The CHMP's opinion was then to go to the European Commission for a final legally binding decision applicable in all EU member states. ²⁶

On 28 June 2019 the EMA recommended that bacterial lysate medicines authorised for respiratory conditions be used only to prevent recurrent respiratory infections, with the exception of pneumonia. ²⁷ Bacterial lysate medicines are made from bacterial cells that are broken down, are taken by mouth, and are intended to stimulate the immune system to recognise and fight infections. ²⁷ In the review, the CHMP considered the results of clinical studies, data on side effects reported with the medicines and advice from an expert group on infectious diseases. ²⁷ It found no robust data showing that the medicines are effective at treating existing respiratory infections or at preventing pneumonia, so they should not be used for these purposes. ²⁷ Although the data are limited, it found some evidence of effectiveness in preventing recurrent respiratory tract infections, with a safety profile in line with what is expected for this type of product. ²⁷ The CHMP therefore recommended that use for prevention could continue, but the companies must provide further data on safety and effectiveness from new clinical studies by 2026. ²⁷

On 9 March 2018 the EMA recommended contraindicating the prostate cancer medicine Xofigo (radium-223 dichloride) with Zytiga (abiraterone acetate) and prednisone or prednisolone, because of an increased risk of death and fractures with this combination. ²⁸ Xofigo is authorised for men whose prostate cancer has spread to the bones and is causing symptoms. ²⁸ The PRAC had reviewed preliminary data from an ongoing clinical study in metastatic prostate cancer patients who had not previously received chemotherapy and had no symptoms or only mild symptoms, such as pain. ²⁸ In the study 34.7% of patients given Xofigo, Zytiga and prednisone or prednisolone had died so far, against 28.2% of those given placebo, Zytiga and prednisone or prednisolone, and fractures had occurred more often with the Xofigo combination, in 26% against 8.1%. ²⁸

Healthcare professionals in the EU must not use the combination, and should stop it in men currently treated with it and review their treatment. ²⁸ They were also warned that the safety and efficacy of Xofigo in combination with second-generation androgen receptor antagonists, such as Xtandi (enzalutamide), have not been established. ²⁸ The patients had completed the Xofigo part of the study, the combination was no longer being used, and all the patients involved were being monitored closely. ²⁸ These were temporary measures until the ongoing in-depth review of the benefits and risks of Xofigo was complete. ²⁸

On 15 November 2024, after re-examining its initial opinion, the EMA recommended updating the advice aimed at minimising the risks of interaction between the weight-loss medicine Mysimba (naltrexone/bupropion) and opioid-containing medicines, including painkillers such as morphine and codeine, other opioids used during surgery and certain medicines for cough, cold or diarrhoea. ²⁹ Mysimba, authorised on 26 March 2015, is used with diet and exercise to help manage weight in adults who have obesity, with a body-mass index of 30 or more, or who are overweight, with a body-mass index between 27 and 30, and have weight-related complications such as diabetes, abnormally high levels of fat in the blood or high blood pressure. ²⁹ Opioids may not work effectively in patients taking Mysimba, because one of its active substances, naltrexone, blocks their effects. ²⁹ There is also a risk of rare but serious and potentially life-threatening reactions, such as seizures and serotonin syndrome, in people taking Mysimba together with medicines for depression and opioids. ²⁹

Patients and healthcare professionals are reminded that Mysimba must not be used in people who are dependent on opioids, people receiving treatment with opioid agonists such as methadone or buprenorphine, or people going through acute opioid withdrawal. ²⁹ People using Mysimba are to be given a patient card to carry at all times, reminding them to tell their doctor, in case of surgery, that they use it. ²⁹ This is because Mysimba should be stopped for a minimum of three days before starting treatment with opioids, which are often used to prevent pain and discomfort during surgery and medical procedures. ²⁹ The product information was being updated to reflect these changes. ²⁹

On 30 April 2020 the EMA recommended testing patients for lack of the enzyme dihydropyrimidine dehydrogenase (DPD) before starting cancer treatment with fluorouracil given by injection or infusion, or with the related medicines capecitabine and tegafur. ²⁰ Fluorouracil given by injection or infusion and its prodrugs capecitabine and tegafur are used to treat various cancers, and they work by interfering with enzymes involved in making new DNA, so blocking the growth of cancer cells. ²⁰ Fluorouracil applied to the skin is used for skin conditions such as actinic keratosis and warts. ²⁰ Patients who completely lack DPD must not be given any fluorouracil medicine, and for patients with partial deficiency the doctor may consider starting cancer treatment at lower doses than normal. ²⁰ A significant proportion of the general population is deficient in DPD, which is needed to break down fluorouracil and the related medicines capecitabine, tegafur and flucytosine, so in them fluorouracil can build up in the blood after treatment, leading to severe and life-threatening side effects such as neutropenia, neurotoxicity, severe diarrhoea and stomatitis. ²⁰ Patients can be tested by measuring the level of uracil in the blood or by checking for certain mutations in the gene for DPD, and relevant clinical guidelines should be taken into consideration. ²⁰ Therapeutic drug monitoring of fluorouracil may improve clinical outcomes in patients receiving continuous infusions. ²⁰ The recommendations do not apply to fluorouracil medicines used on the skin for conditions such as actinic keratosis and warts, as only very low levels of the medicine are absorbed through the skin. ²⁰

Flucytosine, which is related to fluorouracil, is used to treat severe yeast and fungal infections, including some forms of meningitis. ²⁰ Because treatment of severe fungal infections should not be delayed, pre-treatment testing for DPD deficiency, which may take up to one week, is not required for it, although flucytosine is contraindicated in patients with known complete DPD deficiency because of the risk of life-threatening toxicity. ²⁰ For patients with partial deficiency the doctor may consider stopping flucytosine if severe side effects occur. ²⁰ In case of drug toxicity, stopping flucytosine should be considered, and determining DPD activity may be considered where toxicity is confirmed or suspected. ²⁰ The review was initiated in March 2019 at the request of the French Medicines Agency (ANSM) and was carried out first by the PRAC. ²⁰ It covered fluorouracil medicines given by injection or applied to the skin, the oral prodrugs capecitabine and tegafur, which are converted to fluorouracil in the body, and flucytosine, given by injection or by mouth, some of which is converted into fluorouracil in the body. ²⁰

A newly listed side effect, or a benefit-risk balance unchanged

On 7 April 2021 the PRAC concluded that unusual blood clots with low blood platelets should be listed as very rare side effects of Vaxzevria, formerly COVID-19 Vaccine AstraZeneca, taking into consideration all the evidence then available, including the advice of an ad hoc expert group. ³⁰ The clots occurred in veins in the brain (cerebral venous sinus thrombosis) and the abdomen (splanchnic vein thrombosis) and in arteries, together with low levels of blood platelets and sometimes bleeding. ³⁰ The committee reviewed in depth 62 cases of cerebral venous sinus thrombosis and 24 of splanchnic vein thrombosis reported to EudraVigilance as of 22 March 2021, 18 of them fatal. ³⁰ The cases came mainly from the spontaneous reporting systems of the European Economic Area and the UK, where around 25 million people had received the vaccine. ³⁰ As of 4 April 2021, 169 cases of cerebral venous sinus thrombosis and 53 of splanchnic vein thrombosis had been reported to EudraVigilance, and around 34 million people had been vaccinated in the European Economic Area and the UK. ³⁰

Most of the reported cases had occurred in women under 60 within two weeks of vaccination, and the review had not then identified any specific risk factor, such as age, gender or a previous medical history of clotting disorders. ³⁰ One plausible explanation for the combination of blood clots and low blood platelets is an immune response, leading to a condition similar to the heparin-induced thrombocytopenia sometimes seen in patients treated with heparin. ³⁰ The PRAC requested new studies and amendments to ongoing ones, and said it would take any further action necessary. ³⁰ The EMA reminded healthcare professionals and people receiving the vaccine to remain aware of the possibility of very rare cases of blood clots with low platelets within two weeks of vaccination. ³⁰ The PRAC stressed the importance of prompt specialist medical treatment: healthcare professionals who recognise the signs of blood clots and low blood platelets and treat them early can help those affected recover and avoid complications. ³⁰ COVID-19 is associated with a risk of hospitalisation and death, the reported combination of blood clots and low platelets is very rare, and the overall benefits of the vaccine in preventing COVID-19 outweigh the risks of side effects. ³⁰

In July 2021 the PRAC concluded that there was no evidence that Zynteglo causes acute myeloid leukaemia (AML). ³¹ Zynteglo is a one-time gene therapy for beta thalassaemia in patients aged 12 and over who require regular blood transfusions, and it uses a viral vector to deliver a working gene into the patient's blood cells. ³¹ It was granted conditional marketing authorisation in May 2019, under which the company is to provide more evidence about the medicine to the EMA for review. ³¹ The PRAC reviewed two cases of AML in patients treated with another investigational medicine, bb1111, in a clinical trial for sickle cell disease. ³¹ There had been no reports of AML with Zynteglo, but both medicines use the same viral vector, and there was a concern that the vector might be implicated in the development of the cancer through insertional oncogenesis. ³¹

The review found that the viral vector was unlikely to be the cause: in one patient the vector was not present in the cancer cells, and in the other it was present at a site, VAMP4, that does not appear to be involved in cancer development. ³¹ More plausible explanations for the AML cases included the conditioning treatment the patients received to clear out bone marrow cells and the higher risk of blood cancer in people with sickle cell disease. ³¹ Patients having Zynteglo also need conditioning treatment, so healthcare professionals should explicitly inform them of the increased risk of blood cancers from the medicines used in conditioning. ³¹ The CHMP agreed with the PRAC's recommendation that healthcare professionals check their patients for signs of blood cancers at least once a year for 15 years. ³¹ The review was initiated on 18 February 2021 at the European Commission's request, under Article 20 of Regulation (EC) No 726/2004, and the PRAC, working closely with experts from the Committee for Advanced Therapies, concluded that the benefits of Zynteglo continue to outweigh its risks. ³¹ As for all medicines, the PRAC will monitor any new data on its safety and update the advice for patients and healthcare professionals when necessary. ³¹

Incidents with a potentially serious impact on public health

The EMA, with the Heads of Medicines Agencies (HMA) and the European Commission, established the EU incident management plan (EU-IMP) in 2009 to enable rapid and effective action across the EU in case of an event or new information on medicines authorised in the EU with a potentially serious impact on public health. ³² Such incidents can affect the safety, quality, efficacy or availability of a medicinal product, and their causes may include the product's safety profile, manufacturing compliance or supply chain issues. ³² When an incident is suspected, a group of experts from the EMA and its scientific committees, the European Commission and the national competent authorities, called the Incident Review Network (IRN), convenes within the shortest possible time to assess the potential impact on public health and to recommend the appropriate regulatory pathway and the most appropriate communications. ³²

A ten-year analysis of the plan concluded that the EU medicines network has a robust regulatory framework, with defined processes and clear responsibilities, for handling public health incidents. ³² In the first ten years of the plan's operation, 78 incidents were managed through the IRN. ³² Of these, 70% were triggered by information that came to the EMA from national competent authorities, followed by information from marketing authorisation holders, at 17%. ³² More than half of the issues concerned the safety of medicines, while quality and non-compliance with good manufacturing practice accounted for over a third. ³² Almost half of the incidents resulted in a variation to the marketing authorisation, risk minimisation measures for the medicine, or both; 22% led to no change to the marketing authorisation, 10% to its suspension and 9% to its revocation. ³² The analysis also found that the revised pharmacovigilance legislation implemented in 2012 has provided robust regulatory instruments and clear roles and responsibilities for managing most safety issues directly, without going through the IRN. ³²

Exchange of safety information between agencies

The EMA and the US Food and Drug Administration (FDA) have set up a cluster on pharmacovigilance topics, to provide a forum for a more systematic and focused exchange of information on the safety of medicines. ³³ It builds on the experience of their earlier regular videoconferences in this area and on the creation of the PRAC, the EMA's dedicated committee for pharmacovigilance. ³³ Clusters are regular collaborative meetings between the EMA and regulators outside the EU that focus on specific topic areas identified as requiring an intensified exchange of information and collaboration. ³³

In the pharmacovigilance cluster, the agencies discuss any pharmacovigilance issue monthly by teleconference, which allows them to work swiftly on the safety of medicines and to coordinate their communication activities. ³³ Canadian and Japanese regulatory authorities take part in the cluster's meetings as observers, and the information exchange is covered by the confidentiality arrangements between the EMA and the FDA. ³³ The cluster is a step in the two agencies' wider approach to expanding and reinforcing international collaboration. ³³ The EMA holds that collaboration between medicines regulators is essential in an increasingly globalised pharmaceutical market, and that regulators are interdependent: any action taken in one territory has repercussions on the rest of the world. ³³ The FDA holds that protecting the health and safety of the American people cannot be done in isolation, and is part of a larger collaborative global effort with its international regulatory partners. ³³

Automating case processing and signal detection

Signal management has historically relied on spontaneous reporting systems and manual assessments, but the growing availability of electronic health records and unstructured health data, driven by digitalisation, has created a demand for more innovative and automated approaches. ³⁴ Growing data volumes challenge the traditional methods and are prompting the adoption of artificial intelligence (AI) techniques. ¹ The evolution of signal management reflects this changing landscape of healthcare data, with traditional manual assessments increasingly supplemented by AI capabilities. ² Although still emerging, AI capabilities have already shown their usefulness and are expected to contribute further to automation in pharmacovigilance, and machine learning, deep learning and natural language processing have shown effectiveness in analysing large datasets. ³⁴

The volume of post-marketing safety data is large, its sources mix structured and unstructured data, and substantial processing is needed to make the data usable. ¹⁹ Sponsors, independent pharmacovigilance providers and regulators use AI and machine-learning systems for case processing, for case evaluation, assessing the likelihood of a causal relationship between the drug and the adverse event, and for case submission. ¹⁹ The FDA uses the Sentinel system, launched in May 2008, for the safety surveillance of medicinal products, and it uses natural language processing and other machine-learning approaches to extract and process relevant information from submitted adverse event reports. ¹⁹ The goal set out in its five-year strategy for 2019 to 2023 is to incorporate emerging data sciences and electronic health records data into safety surveillance. ¹⁹

The Uppsala Monitoring Centre's vigiMatch algorithm identifies potential duplicate reports by scanning similarities between existing cases in VigiBase, the World Health Organization's global safety database for medicines and vaccines, and the centre also applies natural language processing to reports containing unstructured data. ³⁴ Data-mining techniques, including neural networks, decision trees and clustering algorithms, aim to accelerate signal detection in large databases compared with traditional methods that rely on manual exploration or predefined queries. ³⁴ These approaches proved particularly useful during the COVID-19 pandemic, when accelerated vaccine approvals required rapid capture and analysis of safety data from sources such as social media. ³⁴

For pharmacovigilance, the EMA's reflection paper on AI emphasises the potential of AI tools in managing adverse event reports and enhancing signal detection, particularly through models with continuous learning capabilities. ³⁴ The FDA notes that AI algorithms could support signal detection by identifying clusters of signs and symptoms and by identifying cases with safety concerns in real time. ³⁴

The integration of AI into routine pharmacovigilance faces technical limitations, such as hallucination in natural language processing models, which perceive patterns or objects that do not exist, as well as concerns about privacy and bias. ³⁴ The regulatory framework for AI in healthcare is still underdeveloped, with only preliminary reflections and ethical principles issued by organisations such as the WHO and the EMA, and establishing accountability and determining liability within AI-driven processes is a further challenge. ³⁴ So transparency and auditability of AI models may become prerequisites for regulatory approval, which implies that decision-making pathways, data sources and risk assessments should be well documented. ³⁴ As the volume and complexity of safety data grow, a structured, harmonised and flexible regulatory framework that evolves alongside technological progress is increasingly needed, so that AI can be used fully and ethically in pharmacovigilance. ^{34, 2, 1} Industry and regulators should collaborate in developing harmonised models and methods that adapt to large amounts of data and ensure effective signal management. ²